Optimized arrangement method of autoclave curing tool for composite material component

Abstract

The invention belongs to the technical field of composite materials, and particularly relates to a composite material component autoclave curing tool optimized arrangement method. According to the method, a neural network model is constructed, a complex mapping relation between tool features and the size of an autoclave platform is learned, and the optimal placement coordinates of a plurality of tools are directly output. Therefore, the consistency, the stability and the design efficiency of a tool arrangement scheme are remarkably improved while the curing quality of the composite material workpiece is guaranteed.

Description

Technical Field

[0001] This invention belongs to the field of composite material technology, specifically relating to an optimized arrangement method for composite material component autoclave curing fixtures. Background Technology

[0002] The curing process of aerospace composite components involves numerous complex issues such as heat conduction and chemical reactions, making it a crucial process in composite material molding. Autoclave curing is currently the most common curing method. An autoclave is a key piece of equipment used in manufacturing high-performance composite components. Its main function is to ensure uniform curing of the resin matrix within a mold by precisely controlling temperature and pressure. This process typically includes three stages: heating, pressurization, and cooling, ensuring the composite material achieves the designed mechanical properties. In actual production, to improve production efficiency and autoclave resource utilization, multiple parts with the same curing parameters are usually placed in the autoclave simultaneously for curing. Optimizing the airflow distribution within the autoclave when multiple parts are curing simultaneously is a crucial process step in ensuring part quality. Many defects in composite parts, such as delamination, porosity, deformation, and cracks, are caused by uneven curing. To optimize the uniformity of the airflow during the simultaneous curing of multiple composite parts, in addition to reasonable tooling design and optimized process parameters, methods for optimizing the arrangement of tooling for multiple parts can be used. Currently, tooling arrangement within the autoclave largely relies on manual experience, resulting in poor consistency and low efficiency. Summary of the Invention

[0003] The purpose of this invention is to address the problems of poor consistency and low efficiency in tooling arrangement for simultaneously loading multiple aerospace composite parts into an autoclave, which relies on manual experience. This invention proposes a deep learning-based method for optimizing the arrangement of tooling for autoclave curing. This method constructs a neural network model to learn the complex mapping relationship between tooling features and autoclave platform dimensions, directly outputting the optimal placement coordinates for multiple tooling components. This significantly improves the consistency, stability, and design efficiency of the tooling arrangement scheme while ensuring the curing quality of the composite parts.

[0004] The technical solution to achieve the objective of this invention is as follows: a method for optimizing the arrangement of tooling for autoclave curing of composite material components is provided, comprising the following steps: Step 1) Data Acquisition and Preprocessing: Collect multi-source historical data, including tooling geometry features, autoclave platform dimensions, and verified placement positions, and normalize the data to form standardized input parameters required by the model, thus forming a structured dataset in text form. Step 2) Model building: Establish a feedforward neural network model based on a multilayer perceptron. This model receives the numerical feature vectors of the tooling and the platform, achieves feature fusion through multilayer nonlinear transformation, and outputs the two-dimensional coordinate prediction values ​​of each tooling on the autoclave platform. That is, the position coordinates (x, y) of each tooling are output through the regression layer. Step 3) Model Training: The backpropagation algorithm and stochastic gradient descent method are used to optimize the model parameters. During the training process, a composite loss function that includes coordinate prediction error and physical constraint violation term is used to ensure that the model can accurately predict the position while meeting the actual process constraints. The training set and the validation set are divided in an 8:2 ratio, and the model performance is continuously optimized through iterative training. Step 4) Model Validation: The trained model is evaluated on an independent test set. The effectiveness of the model is comprehensively evaluated by comparing the model output scheme with the manually optimized scheme in terms of placement accuracy, constraint satisfaction rate and computational efficiency. Step 5) Model Application and Iteration: Integrate the trained model into the actual tooling layout decision system to provide operators with optimized placement schemes, reduce manual design and calculation time, and continuously optimize the model based on feedback from actual applications.

[0005] Furthermore, the tooling geometry features mentioned in step 1) include length, width, height and windward orientation information, the autoclave platform dimensions include platform length and width, and the verified placement position refers to the historical data of the manually optimized position coordinates (x, y) of the tooling on the platform.

[0006] Furthermore, the normalization process mentioned in step 1) refers to cleaning the original data and verifying its compliance with constraints to ensure that the data meets the process requirements; The normalization formula is: x_norm = (x - x_min) / (x_max - x_min), where x_norm is the normalized value, x is the original data value, x_min is the minimum value of the dataset, and x_max is the maximum value of the dataset. This formula is applied to the length, width, and height of tooling, and the length and width of autoclave platforms, after manual optimization of the coordinates.

[0007] Furthermore, in step 1), the orientation of the tooling's windward side is subjected to periodic feature processing: θ_sin = sin(2π*θ / 360), where θ_sin is the sine of the corresponding angle, and θ is the input angle in degrees.

[0008] The orientation of the tooling's windward side refers to the angular position of the surface of the tooling structure that directly bears the wind force inside the autoclave, relative to a fixed reference direction (usually the direction of the autoclave door). This angle is measured clockwise or counterclockwise and ranges from -45° to 45°.

[0009] Furthermore, the constraints described in step 2) include: the windward side of the tooling faces the same direction as the autoclave door; the minimum tooling spacing is within 50cm; the minimum distance between the tooling and the front end of the platform is 50cm, and the minimum distance between the tooling and the rear end of the platform is 50cm; the tooling is arranged in ascending order of maximum height along the direction from the door to the tail of the platform; and all tooling is located inside the platform boundary.

[0010] Furthermore, the neural network model described in step 2) includes an input layer, multiple hidden layers, and an output layer, wherein the output layer corresponds to the position information of each tooling. The input feature vector of the model includes 9 nodes, namely, tooling length, tooling width, tooling height, θ_sin, autoclave platform length, autoclave platform width, minimum tooling spacing, minimum distance between the tooling and the front end of the platform, and minimum distance between the tooling and the rear end of the platform.

[0011] Furthermore, in step 2), the hidden layer contains 3 fully connected layers, each with 128 neurons and the activation function is ReLU; the output layer has 2 nodes (predicting X and Y coordinates) and the activation function is Sigmoid.

[0012] Furthermore, the composite loss function described in step 3) consists of mean squared error loss and constraint loss. During training, the loss function uses labels (human-verified placement positions) to calculate the MSE; during application, the model relies solely on the aforementioned ontology and platform features for prediction. The mean square error loss function

[0013] In the formula L MSE Let Σ{i=1}^N be the mean squared error loss function, and let Σ{i=1}^N be the summation over all samples (from the 1st to the Nth). For the first i Variables of a sample X The squared prediction error, For the first i Variables of a sample Y The squared prediction error, 1 / N To calculate the average of the total error.

[0014] The constraint losses include tooling spacing constraints, tank door / tank tail distance constraints, height sorting constraints, and windward orientation constraints.

[0015] Tooling spacing constraints

[0016] In the formula, i, j are the index numbers of the tooling, i, j ∈ {1,2,...,N}, Xᵢ', Yᵢ' are the coordinate positions of the i-th tooling on the platform predicted by the model, X j ', Y j ' represents the coordinate position of the j-th tooling on the platform as predicted by the model. d min The minimum tooling spacing required by the process. Let max(0, ...) be the Euclidean distance between the two fixtures. d min - distance) is a penalty when the actual distance is less than the minimum spacing, otherwise it is 0; Platform boundary constraint loss

[0017] In the formula, Yᵢ' represents the coordinate position of the i-th tooling in the longitudinal direction (Y-axis) of the platform. d front To maintain the minimum safe distance between the tooling and the front end of the platform, d ᵣ ea ᵣ represents the minimum safe distance between the tooling and the platform backend. L p The total length of the autoclave platform, ( L p - d ᵣ ea ᵣ) represents the location of the platform's backend security boundary; Height-ordering constraint loss

[0018] In the formula, H sorted This is an array of tool heights sorted by Y-coordinate from smallest to largest. H sorted[i] Let this be the height of the tool at the i-th position after sorting. H sorted[i+1] Let max(0, ) be the height of the tool at position i+1 after sorting (the tool further back on the Y-axis). H sorted[i] - H sorted[i+1] This is a penalty that occurs when the height of the preceding tooling is greater than the height of the following tooling. Windward orientation constraints

[0019] In the formula, θ sin,i Let be the sinusoidal encoding value of the windward orientation of the i-th tooling. θ target The sine code value for the target orientation; In summary, the composite loss function is:

[0020] In the formula, λ1, λ2, λ3, and λ4 are the weight coefficients of mean square error, tooling spacing, platform boundary, and height sorting constraint, respectively.

[0021] Furthermore, in step 3), the training process employs an early stopping strategy to prevent overfitting and periodically saves model checkpoints; early stopping method: training is terminated when the training set loss does not improve within 10 epochs.

[0022] Furthermore, the training set samples mentioned in step 3) all contain placement labels that have been manually optimized and determined.

[0023] Compared with the prior art, the significant advantages of this invention are: The production of aerospace composite components is usually multi-variety and small-batch. In order to improve production efficiency and reduce production energy consumption, it is necessary to put multiple composite component toolings with the same curing parameters into the autoclave for curing at the same time. This invention realizes the optimization of the arrangement of multiple toolings of different specifications and sizes into the autoclave.

[0024] For the first time, geometric constraints (spacing, tank door / tank tail distance) and process constraints (height sorting, windward orientation) are integrated into the loss function, significantly improving the rationality of the layout.

[0025] An MLP structure is adopted, combined with a Sigmoid activation function and an inverse normalization strategy, to ensure that the output coordinates satisfy the physical boundary conditions.

[0026] Compared with manual layout, it no longer relies on human experience, ensuring the consistency and stability of tooling layout, improving layout efficiency, and is of great significance for ensuring product curing quality and improving production efficiency. Attached Figure Description Figure 1 Flowchart of the tooling arrangement method for autoclave curing; Figure 2 This is a schematic diagram of the internal working principle of an autoclave. Figure 3 This is a schematic diagram of the tooling features in Example 1; Figure 4 This is a schematic diagram showing the position of the tooling after it has been placed on the platform. Figure 5 This is a schematic diagram of the tooling features in Example 2. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0029] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0030] Due to the large amount of sample data, only two sets of sample data are selected for illustration in the attached figures. The other sample data show similar patterns. This invention provides an optimized arrangement method for the autoclave curing fixture of aerospace composite material components, such as... Figure 1 As shown, it includes the following steps.

[0031] Example 1: 1) Data Acquisition and Preprocessing: Collect 2000 sets of historical tooling layout data. Each set includes: tooling characteristics (length, width, height, windward orientation), autoclave platform dimensions (length, width), and manually optimized placement coordinates (X, Y). For example... Figure 2 As shown, the leftmost point at the front of the autoclave platform is taken as the origin. Similarly, among the four corner points of the tooling, the corner point closest to the origin of the coordinate system is taken as the coordinate of the tooling placement position, with the unit being meters (m). The data is normalized to form a structured dataset, as illustrated in Table 1. The tooling feature diagram in this embodiment is shown below. Figure 3 For example.

[0032] Table 1

[0033] 2) Data Normalization and Constraint Processing: The above data is processed into structured input vectors acceptable to the model. Each sample includes tooling attributes and corresponding platform dimensions. All data must meet the following process constraints: the tooling's windward side faces the same direction as the autoclave door; the minimum spacing between tooling fixtures is 50cm; the minimum distance between tooling fixtures near the door and the front end of the platform is 50cm, and the minimum distance between tooling fixtures near the tail end and the rear end of the platform is 50cm; tooling fixtures are arranged from low to high along the direction from the door to the tail end; all tooling fixtures are located within the platform's physical boundaries. A schematic diagram of the tooling placement on the platform is provided. Figure 4 For example.

[0034] 3) Model Construction: A deep learning model based on a multilayer perceptron (MLP) is built on a computer using the PyTorch framework. The model structure includes: an input layer (receiving feature vectors from the tooling and platform), three fully connected hidden layers (each equipped with a ReLU activation function), and an output layer (regressing and outputting the continuous two-dimensional coordinate values ​​(x, y) of each tooling).

[0035] 4) Model Training: The mean squared error loss function was used, with a constraint penalty term added to weight penalties for spacing violations, out-of-bounds placement, and height order errors. A stochastic gradient descent optimizer was used with a learning rate of 0.001 and a batch size of 32. 1600 datasets were used as the training set, and 400 datasets as the validation set (8:2 ratio). Early stopping was employed during training to prevent overfitting. Training lasted for 200 epochs, and the best-performing model checkpoints were periodically saved.

[0036] 5) Model Validation: The model performance is evaluated using 400 sets of reserved independent test data. The average Euclidean distance error between the predicted coordinates and the manually optimized coordinates is calculated. The satisfaction rate of the output scheme with various process constraints is statistically analyzed, and the model placement accuracy and reliability are comprehensively evaluated.

[0037] 6) Model Application: The trained model can quickly output recommended placement coordinates by inputting actual tooling and platform parameters, significantly improving layout efficiency and consistency. The system can record feedback from actual applications for subsequent model iteration and optimization.

[0038] Example 2: 1) Data Acquisition and Preprocessing: 1500 sets of historical tooling layout data were collected. Each set of data includes: tooling characteristics (length, width, height, windward orientation), autoclave platform dimensions (length, width), and manually optimized placement coordinates. The raw data was cleaned and normalized to form a standardized dataset. The tooling characteristic diagram in this embodiment is shown below. Figure 5 For example.

[0039] 2) Data Normalization and Constraint Processing: Tooling attributes and platform dimensions are integrated into a structured feature vector. Each sample contains complete tooling-platform parameters and corresponding coordinate labels. All data strictly meet the following process constraints: the tooling's windward side faces the same direction as the autoclave door; the spacing between tooling fixtures is no less than 50cm; tooling near the door is no less than 50cm from the front end of the platform, and tooling near the rear end is no less than 50cm from the rear end of the platform; tooling is arranged in ascending order of height from the door to the rear end; all tooling positions are within the platform's physical boundaries. A schematic diagram of tooling placement on the platform can be found here. Figure 4 .

[0040] 3) Model Construction: A deep learning model based on a multilayer perceptron (MLP) was built and implemented using the PyTorch framework. The network structure includes: an input layer (receiving normalized feature vectors), three fully connected hidden layers (each containing 128 neurons and using the ReLU activation function), and an output layer (regressing the planar coordinate values ​​(x, y) of each tool).

[0041] 4) Model Training: A composite loss function was employed, including mean squared error coordinate loss and constraint loss, to jointly optimize for insufficient spacing, out-of-bounds placement, and incorrect height order. The Adam optimizer was used, with an initial learning rate of 0.0005 and a batch size of 64. 1200 datasets were used as the training set, and 300 datasets as the validation set (8:2 ratio). Learning rate decay and early stopping strategies were employed during training, for a total of 300 epochs. The optimal model parameters were periodically saved.

[0042] 5) Model Validation: The model is quantitatively evaluated using 300 independent sets of test data. The average position error between the predicted coordinates and the manually optimized coordinates is calculated, and the satisfaction rate of the output scheme with various process constraints is statistically analyzed to comprehensively evaluate the model performance.

[0043] 6) Model Application: The trained model can quickly output recommended placement coordinates by inputting actual tooling and platform parameters, significantly improving layout efficiency and consistency. The system can record feedback from actual applications for subsequent model iteration and optimization.

[0044] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0045] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for optimizing the arrangement of a composite material component autoclave curing tool, characterized in that, The method comprises the following steps: Step 1) data collection and preprocessing: collect multi-source historical data including tooling geometric features, autoclave platform size and verified placement position, and normalize the data to obtain standardized input parameters required by the model, that is, form a structured data set in text form; Step 2) model construction: a feedforward neural network model based on multilayer perceptron is established, which receives numerical feature vectors of tooling and platform, realizes feature fusion through multilayer nonlinear transformation, and outputs two-dimensional coordinate prediction values of each tooling on the autoclave platform, that is, outputs the position coordinates (x, y) of each tooling through the regression layer; Step 3) model training: the back propagation algorithm and the stochastic gradient descent method are used to optimize the model parameters, and a composite loss function containing coordinate prediction error and physical constraint violation items is used in the training process to make the model accurately predict the position while meeting the actual process constraints; the training set and the validation set are divided in the ratio of 8:2, and the model performance is continuously optimized through iterative training; Step 4) model verification: the performance of the trained model is evaluated on an independent test set, and the effectiveness of the model is evaluated comprehensively by comparing the placement accuracy, constraint satisfaction rate and calculation efficiency of the model output scheme and the artificial optimization scheme.

2. The method of claim 1, wherein the method further comprises: The tooling geometric features in step 1) include length, width, height and windward surface orientation information, the autoclave platform size includes platform length and width, and the verified placement position refers to the historical data of the position coordinates (x, y) of the tooling after artificial optimization on the platform.

3. The method of claim 1, wherein the method further comprises: The normalization processing in step 1) refers to data cleaning and constraint compliance verification to ensure that the data meets the process requirements; The normalization formula is: x_norm = (x - x_min) / (x_max - x_min), wherein x_norm is the normalized value, x is the original data value, x_min is the minimum value of the data set, and x_max is the maximum value of the data set. It is applied to tooling length, width, height, autoclave platform length, width and artificial optimized coordinates.

4. The method of claim 1, wherein, In step 1), the windward surface orientation of the tooling is processed periodically: θ_sin = sin(2π*θ / 360), wherein θ_sin is the sine value corresponding to the angle, and θ is the input angle in degrees.

5. The method of claim 1, wherein, The constraint conditions in step 2) include: the windward surface orientation of the tooling is consistent with the direction of the autoclave door; the minimum tooling spacing is within 50 cm; the minimum distance between the tooling and the front end of the platform is 50 cm, the minimum distance between the tooling and the rear end of the platform is 50 cm; the tooling is arranged in the order of increasing height along the direction from the door to the tail of the autoclave; and all toolings are located inside the platform boundary.

6. The method of claim 1, wherein, The neural network model in step 2) includes an input layer, multiple hidden layers and an output layer, wherein the output layer corresponds to the position information of each tooling; The input feature vector of the model includes 9 nodes, which are tooling length, tooling width, tooling height, θ_sin, autoclave platform length, autoclave platform width, minimum tooling spacing, minimum distance between tooling and platform front end, and minimum distance between tooling and platform rear end.

7. The method of claim 1, wherein, The hidden layer in step 2) comprises 3 fully connected layers, each layer having 128 neurons, and the activation function is ReLU; the output layer has 2 nodes (predicted X, Y coordinates), and the activation function is Sigmoid.

8. The method of claim 1, wherein, The composite loss function in step 3) is composed of a mean square error loss and a constraint loss; during training, the loss function uses labels (verified placement positions by humans) to calculate MSE; during application, the model only relies on the above ontology and platform features for prediction. The mean square error loss function In the formula L MSE is the mean squared error loss function, and Σ{i=1}^N is the summation over all samples, ( the prediction error squared for the variable i of the first X sample, the prediction error squared for the variable i of the first Y sample, N is averaged over the total error; The constraint loss comprises a tool spacing constraint, a tank door / tank tail distance constraint, a height sorting constraint, and a windward surface orientation constraint. Tool spacing constraint where i, j are the index of fixtures, i, j ∈ {1, 2,..., N}, X j ', Y j ' are the coordinate positions of the jth fixture on the platform predicted by the model, d min is the minimum fixture distance required by the process, is the Euclidean distance between two fixtures, max(0, d min distance) is the penalty when the actual distance is less than the minimum distance, otherwise 0. Platform boundary constraint loss where Y'i is the coordinate position of the i-th tooling in the platform longitudinal direction (Y-axis), d front is the minimum safety distance of the tooling from the front end of the platform, d ᵣ ea ᵣis the minimum safety distance of the tooling from the rear end of the platform, L p is the total length of the autoclave platform, L p - d ᵣ ea ᵣ) is the safety boundary position of the rear end of the platform; Height sorting constraint loss In the formula, H sorted is the tool height array sorted by Y coordinate from small to large, H sorted[i] is the tool height at the i-th position after sorting, H sorted[i+1] is the tool height at the i+1-th position after sorting (the tool further away from the Y axis), max(0, H sorted[i] - H sorted[i+1] is the penalty when the tool height in front is greater than the tool height behind. Windward surface orientation constraint wherein θ sin,i sin(i) is the sine encoding value for the i-th tool facing the wind, θ target sin(t) is the sine encoding value for the target facing. Therefore, the composite loss function is In the formula, λ1, λ2, λ3, and λ4 are weight coefficients of the mean square error, tool spacing, platform boundary, and height sorting constraints, respectively.

9. The method of claim 1, wherein, In step 3), the training process adopts an early stopping strategy to prevent overfitting and regularly saves model checkpoints; the early stopping method: terminate training when the training set loss does not improve within 10 epochs.

Images