Composite material automatic wire laying forming optimization method, system, medium and wire laying machine

Abstract

The present application relates to the field of wire laying control technology, and discloses a composite material automatic wire laying forming optimization method, system, medium and wire laying machine. The method comprises: inputting dynamic tension control data into a fuzzy neural network calculation engine to obtain a dynamic compensation model for the retraction and release speed; based on the dynamic compensation model for the retraction and release speed, using a reinforcement learning engine to calculate the shearing sequence of multiple bundles and generate an intelligent shearing instruction set; constructing a re-delivery trajectory mapping relationship based on the intelligent shearing instruction set, and forming a re-delivery path matrix after iterative calculation by a digital twin engine; establishing a temperature field distribution prediction model based on the re-delivery path matrix, and constructing a temperature field dynamic control sequence after finite element analysis and calculation; substituting the temperature field dynamic control sequence into a multi-objective constraint equation, and outputting a laying strategy data packet after deep reinforcement learning iterative optimization. The present application realizes the precise coordination of multiple process parameters such as tension control, speed compensation, shear control, and temperature control.

Description

Technical Field

[0001] The present application relates to the field of wire laying control, and in particular to an automatic wire laying forming optimization method, system, medium and wire laying machine for composite materials. Background Art

[0002] The automatic fiber placement technology for composite materials is widely used in aerospace, transportation and other fields. Traditional automatic fiber placement technology for composite materials mainly adopts manual operation, that is, the operator lays the carbon fiber cloth layer by layer on the mold surface, and then solidifies and forms it. With the improvement of industrial automation level, automatic fiber placement systems based on industrial robots have emerged, which control the fiber placement head through a robotic arm to realize the forming of carbon fiber composite structural parts. This technology controls the multi-degree-of-freedom robotic arm through a CNC system, lays the prepreg tow on the mold surface according to the preset trajectory, and completes the laying process through heating, compaction and other processes.

[0003] However, the existing automatic wire laying technology for composite materials has some obvious shortcomings. First, when encountering large curvature or special-shaped molds, traditional control strategies are difficult to ensure the laying quality, and are prone to problems such as wrinkles, overlaps, or excessive gaps. Secondly, the existing technology lacks synergy in key links such as tension control, shear positioning, and temperature control, making it difficult to achieve precise coordination of various process parameters, affecting the molding quality. In addition, traditional control methods mainly rely on experience settings for the optimization of process parameters, lack intelligent adaptive adjustment capabilities, and are difficult to meet the requirements of high-quality and high-efficiency production. Summary of the invention

[0004] This application solves the problem of insufficient precision of multi-parameter coordinated control in existing composite material automatic wire laying technology. By establishing a set of intelligent control systems based on deep learning, precise coordination of multiple process parameters such as tension control, speed compensation, shear control, and temperature control is achieved.

[0005] In the first aspect, the present application provides a method for optimizing automatic wire placement and forming of composite materials, which comprises: the prepreg wire bundle tension data input by the tension sensor is subjected to LSTM deep learning operation to output dynamic tension control data; the dynamic tension control data is input into a fuzzy neural network calculation engine, and after speed curve analysis operation, a dynamic compensation model of the retraction and release speed is obtained; based on the dynamic compensation model of the retraction and release speed, a reinforcement learning engine is used to calculate the multi-wire bundle shearing sequence and generate an intelligent shearing instruction set; a re-delivery trajectory mapping relationship is constructed based on the intelligent shearing instruction set, and a re-delivery path matrix is ​​formed after iterative calculation by a digital twin engine; a temperature field distribution prediction model is established according to the re-delivery path matrix, and a temperature field dynamic control sequence is constructed after finite element analysis and calculation; the temperature field dynamic control sequence is substituted into a multi-objective constraint equation, and after deep reinforcement learning iterative optimization, a placement strategy data packet is output.

[0006] In a second aspect, the present application provides a composite material automatic wire placement forming optimization system, the composite material automatic wire placement forming optimization system comprising:

[0007] An input module is used to process the prepreg tow tension data input by the tension sensor through LSTM deep learning calculation and output dynamic tension control data;

[0008] A calculation module, used for inputting the dynamic tension control data into a fuzzy neural network calculation engine, and obtaining a dynamic compensation model for the retraction and release speed after speed curve analysis and calculation;

[0009] A calculation module, for calculating the multi-tow cutting sequence based on the retraction and release speed dynamic compensation model using a reinforcement learning engine to generate an intelligent cutting instruction set;

[0010] A construction module is used to construct a re-delivery trajectory mapping relationship based on the intelligent cutting instruction set, and form a re-delivery path matrix after iterative calculation by the digital twin engine;

[0011] Establishing a module for establishing a temperature field distribution prediction model according to the retransmission path matrix, and constructing a temperature field dynamic control sequence after finite element analysis and calculation;

[0012] The output module is used to substitute the temperature field dynamic control sequence into the multi-objective constraint equation, and output the placement strategy data package after deep reinforcement learning iterative optimization.

[0013] A third aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned composite material automatic wire placement molding optimization method.

[0014] In a fourth aspect, the present invention further provides a wire laying machine, the wire laying machine comprising:

[0015] one or more processors;

[0016] A storage device for storing one or more programs;

[0017] When the one or more programs are executed by the one or more processors, the one or more processors implement the composite material automatic wire placement forming optimization method as described above.

[0018] In the technical solution provided by the present application, the prepreg tow tension data is processed by LSTM deep learning, and the fuzzy neural network calculation engine is combined to realize the accurate analysis of the speed curve, thereby improving the accuracy and stability of the tension control. The reinforcement learning engine optimizes and calculates the shearing sequence of multiple tows, and realizes the accurate mapping of the re-delivery trajectory through the iterative calculation of the digital twin engine, which greatly improves the shearing positioning accuracy and re-delivery reliability. The temperature field distribution prediction model combines the finite element analysis technology to realize the dynamic and accurate control of the temperature field, ensuring the temperature uniformity and stability during the laying process. Through deep reinforcement learning, the dynamic control sequence of the temperature field is optimized by multi-objective constraints, and the global optimization of the laying strategy is realized, which significantly improves the quality stability of the laying process. The whole scheme realizes the coordinated optimization of multiple key parameters such as tension control, speed compensation, shear positioning, temperature control, etc. in the laying process through multi-level data processing and intelligent algorithm optimization, effectively solving the problems of improper parameter matching and unstable process in traditional laying technology. At the same time, the intelligent optimization method based on deep learning and reinforcement learning enables the control strategy to be adaptively adjusted to adapt to the laying needs under different working conditions, thereby improving the flexibility and intelligence level of the production process. In addition, the introduction of digital twin technology makes laying trajectory planning more precise and temperature field prediction more accurate, providing reliable technical support for high-quality wire laying manufacturing. Through multi-objective constraint optimization, while ensuring product quality, it also achieves improved energy efficiency and reduced production costs, with significant economic benefits and practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0020] Figure 1 A schematic diagram of an embodiment of a composite material automatic wire placement forming optimization method in an embodiment of the present application;

[0021] Figure 2It is a schematic diagram of an embodiment of a composite material automatic wire placement forming optimization system in an embodiment of the present application.

[0022] Figure 3 A first structural schematic diagram of a wire laying machine provided in an embodiment of the present application;

[0023] Figure 4 The second structural schematic diagram of the wire laying machine provided in the embodiment of the present application. DETAILED DESCRIPTION

[0024] The embodiments of the present application provide a composite material automatic wire laying forming optimization method, system, medium and wire laying machine. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , one embodiment of the composite material automatic wire placement forming optimization method in the embodiment of the present application includes:

[0026] Step S101, the prepreg tow tension data input by the tension sensor is subjected to LSTM deep learning operation to output dynamic tension control data;

[0027] Step S102, input the dynamic tension control data into the fuzzy neural network calculation engine, and after speed curve analysis and calculation, obtain the retracting and releasing speed dynamic compensation model;

[0028] Step S103, based on the dynamic compensation model of the retracting and releasing speed, the multi-tow cutting sequence is calculated using the reinforcement learning engine to generate an intelligent cutting instruction set;

[0029] Step S104: construct a re-delivery trajectory mapping relationship based on the intelligent cutting instruction set, and form a re-delivery path matrix through iterative calculation by the digital twin engine;

[0030] Step S105, establishing a temperature field distribution prediction model based on the retransmission path matrix, and constructing a temperature field dynamic control sequence after finite element analysis and calculation;

[0031] Step S106: Substitute the temperature field dynamic control sequence into the multi-objective constraint equation, and after deep reinforcement learning iterative optimization, output the placement strategy data package.

[0032] It is understandable that the execution subject of the present application may be a composite material automatic wire placement optimization system, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0033] Specifically, the prepreg tow tension data collected by the tension sensor is input into the system. The tension sensor is arranged on the prepreg tow conveying path to collect the tow tension change data in real time with a sampling frequency of 1000Hz. The collected tension data is decomposed at multiple scales by wavelet transform to separate the tension fluctuation characteristics of different frequency bands, and the noise interference is eliminated by singular value decomposition filtering. The processed tension data is input into a three-layer LSTM network structure containing 64 hidden neurons. The LSTM network learns the time series features of the tension data through the input gate, forget gate and output gate to capture the long-term dependency of the tension change. The processed dynamic tension control data is then input into the fuzzy neural network engine. The fuzzy neural network first reduces the dimension of the tension data through the autoencoder to reduce the data redundancy. The compressed feature data is fuzzy quantized by the triangular membership function to establish a fuzzy rule base. The fuzzy rule base is optimized by genetic algorithm to select the optimal rule set. The optimized rule set is input into the improved BP neural network for online learning to obtain the speed compensation feature vector. The speed compensation feature vector is used through a convolutional neural network to extract the speed curve features, and finally the retraction and release speed dynamic compensation model is output.

[0034] Based on the dynamic compensation model of the retracting and releasing speed, the state space of the shearing process is encoded by the PPO algorithm. The state space contains information such as the current position, tension, and speed of the tow, and the shearing action space representation is generated by multi-layer perceptron mapping. The shearing action space is evaluated by the dual Q network to evaluate the value function of different shearing actions, and the Monte Carlo tree search strategy is combined to output the shearing state transition probability matrix. The probability matrix is ​​learned by temporal difference to construct the shearing policy gradient vector. The policy gradient vector is input into the Actor-Critic network for optimization and update to form a multi-tow shearing decision sequence. Based on the intelligent shearing instruction set, the spatiotemporal features are extracted by the recursive neural tensor network to generate the feature point cloud of the re-delivery trajectory. The feature point cloud is compared and learned by the twin convolutional neural network, and the trajectory encoding sequence is output by combining the three-dimensional space transformation network. The encoding sequence is processed by the graph convolutional network to construct the re-delivery trajectory topology map. The trajectory topology map is input into the physical simulation engine to calculate the re-delivery dynamics parameters. The dynamics parameters are estimated by the self-correcting particle filter algorithm to finally form the re-delivery path matrix.

[0035] According to the retransmission path matrix, the temperature field data is layered according to the three spatial dimensions of X, Y, and Z to generate a temperature data cube. The temperature data is downsampled and discretized to construct a temperature field node data array. The node data is extracted through thermodynamic features, and energy conservation conversion is performed to output a temperature multi-dimensional feature data stream. The feature data stream is processed in time and space to generate a four-dimensional evolution surface of the temperature field. The temperature gradient information is extracted from the evolution surface, the temperature key control points are screened, and finally the temperature field dynamic control sequence is output. The temperature field dynamic control sequence is input into the multi-objective constraint optimization module. First, the control sequence is multi-dimensionally decoupled, and the feature vectors of the time dimension, space dimension, and temperature dimension are extracted respectively. The feature vector is regularized and the data is corrected according to the physical constraints. After multi-scale feature decomposition, the parameter change law is extracted and the optimization feature space is constructed. Through the quantification of the reward and punishment function, the weights of each parameter are allocated to form a multi-objective optimization data stream. Finally, after the strategy integration process, the placement strategy data packet is output.

[0036] Take the manufacturing of a certain aviation composite component as an example: During the manufacturing process, the initial tow tension data collected by the tension sensor ranged from 5 to 15N. After being processed by the LSTM network, the dynamic tension control accuracy was improved to ±0.2N. After these tension data were input into the fuzzy neural network, the generated speed compensation model controlled the retraction speed within the range of 10-50mm / s, and the compensation accuracy reached ±0.5mm / s. Based on this compensation model, the reinforcement learning algorithm planned the optimal shearing sequence for 8 bundles of fibers, and the shearing positioning accuracy was controlled within the range of ±0.15mm. The digital twin engine generated a re-feeding trajectory based on the shearing instruction, and the trajectory positioning accuracy reached ±0.1mm. The temperature field control sequence maintained the heating temperature within the range of 350±3℃ to ensure that the material was fully softened. The placement strategy data package achieved a placement accuracy of ±0.2mm, and the interlayer bonding strength reached more than 95% of the required value. Throughout the process, the coordinated optimization of various parameters significantly improved product quality and increased production efficiency by more than 3 times.

[0037] In the embodiment of the present application, the prepreg tow tension data is processed by LSTM deep learning, and the fuzzy neural network calculation engine is combined to realize the accurate analysis of the speed curve, thereby improving the accuracy and stability of the tension control. The reinforcement learning engine optimizes and calculates the shearing sequence of multiple tows, and realizes the accurate mapping of the re-delivery trajectory through the iterative calculation of the digital twin engine, which greatly improves the shearing positioning accuracy and re-delivery reliability. The temperature field distribution prediction model combines the finite element analysis technology to realize the dynamic and accurate control of the temperature field, ensuring the temperature uniformity and stability during the laying process. Through deep reinforcement learning, the dynamic control sequence of the temperature field is optimized by multi-objective constraints, and the global optimization of the laying strategy is realized, which significantly improves the quality stability of the laying process. The whole scheme realizes the coordinated optimization of multiple key parameters such as tension control, speed compensation, shear positioning, temperature control, etc. in the laying process through multi-level data processing and intelligent algorithm optimization, effectively solving the problems of improper parameter matching and unstable process in traditional laying technology. At the same time, the intelligent optimization method based on deep learning and reinforcement learning enables the control strategy to be adaptively adjusted to adapt to the laying needs under different working conditions, thereby improving the flexibility and intelligence level of the production process. In addition, the introduction of digital twin technology makes laying trajectory planning more precise and temperature field prediction more accurate, providing reliable technical support for high-quality wire laying manufacturing. Through multi-objective constraint optimization, while ensuring product quality, it also achieves improved energy efficiency and reduced production costs, with significant economic benefits and practical value.

[0038] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0039] (1) The prepreg tow tension data is decomposed into multi-scale time-frequency data by wavelet transform and Fourier analysis, and a tension multi-dimensional spectrum feature matrix is ​​generated after singular value decomposition filtering;

[0040] (2) The tension multi-dimensional spectrum feature matrix is ​​input into a three-layer LSTM network structure containing 64 hidden neurons. After forward recursive calculation and long short-term memory unit state update, the tension spatiotemporal state vector set is output;

[0041] (3) Based on the tension spatiotemporal state vector set, deep temporal correlation mining is performed through a bidirectional gated recurrent unit based on the Transformer architecture, and a deep feature map of tension prediction is formed by combining the residual connection mechanism;

[0042] (4) The tension prediction deep feature map is calculated through a multi-head attention mechanism, and the self-attention and cross-attention weight distributions are integrated to screen out a high-dimensional tension feature sequence;

[0043] (5) The high-dimensional tension feature sequence is iteratively optimized in batches through the improved stochastic gradient descent algorithm, and the tension prediction loss function surface is constructed by combining L1 and L2 regularization constraints;

[0044] (6) The tension prediction loss function surface is substituted into the Adam optimizer with momentum term, and the dynamic tension control data is output after batch normalization through adaptive learning rate adjustment and gradient clipping technology.

[0045] Specifically, multi-scale time-frequency decomposition analysis is performed on the prepreg tow tension data. Wavelet transform is used to decompose the tension signal at different scales, and Fourier analysis converts the signal into frequency domain space. After singular value decomposition filtering, a tension spectrum feature matrix containing multiple frequency band features is generated.

[0046] The calculation process of the multi-head attention mechanism can be expressed as:

[0047]

[0048] in:

[0049] Q, K, V represent the query matrix, key matrix, and value matrix respectively, h represents the number of attention heads, n represents the sequence length, and d k represents the key vector dimension, α ( represents the weight coefficient of the i-th attention head, ψ(V + ) represents the nonlinear transformation function of the value matrix, ω ( represents the output projection matrix of the i-th attention head, j represents the sequence position index currently processed, m represents the sum index during normalization, T represents the matrix transpose operation, Q i represents the query vector of the ith attention head, K j represents the key vector at the jth position, K m represents the key vector at the mth position, V j A vector representing the value at the j-th position.

[0050] The time series features of tension data are extracted. When the sampling frequency is 1000Hz, 1000 raw tension data points are generated per second. The signal is decomposed into 4 scale levels through wavelet transform, and each level contains features of different frequency bands. The threshold of singular value decomposition filtering is set to 0.1 to filter out noise components with singular values ​​less than the threshold. Next, the obtained tension multi-dimensional spectrum feature matrix is ​​input into a three-layer LSTM network for processing. The first layer of the LSTM network contains 64 hidden neurons for preliminary feature extraction; the second layer also contains 64 neurons for feature fusion; the 64 neurons in the third layer are responsible for generating the final time series feature representation. Each LSTM unit contains an input gate, a forget gate, and an output gate. The information flow is controlled through a gating mechanism to capture long-term dependencies.

[0051] The processed tension spatiotemporal state vector set is then input into a bidirectional gated recurrent unit based on the Transformer architecture. The Transformer architecture performs deep feature extraction on tension data through a multi-layer self-attention mechanism and a feedforward neural network. The bidirectional gated recurrent unit processes sequence data in both the forward and reverse directions to capture bidirectional temporal dependencies. The residual connection mechanism retains the original feature information through jump connections, alleviating the gradient vanishing problem of deep networks. The tension prediction deep feature map is calculated by a multi-head attention mechanism to produce a richer feature representation. The multi-head attention mechanism projects the input features into multiple subspaces, calculates the attention scores separately, and then merges the results. The self-attention mechanism focuses on the correlation within the sequence, while the cross-attention mechanism focuses on the correlation between different features. By allocating attention weights, important features are highlighted and irrelevant information is suppressed.

[0052] The obtained high-dimensional tension feature sequence is optimized by an improved stochastic gradient descent algorithm. The algorithm adopts a mini-batch method, each batch contains 128 samples, and the learning rate is initially set to 0.001. The L1 regularization constraint is used to produce a sparse solution, and the L2 regularization constraint is used to prevent overfitting. The regularization coefficients are set to 0.01 and 0.001 respectively. The gradient is calculated by back propagation, the model parameters are iteratively updated, and the prediction loss function surface is constructed. Finally, the tension prediction loss function surface is input into the Adam optimizer. The Adam optimizer combines the advantages of the momentum method and RMSprop and has the ability to adjust the learning rate adaptively. The initial momentum decay rate is set to 0.9, and the second-order momentum decay rate is 0.999. The gradient clipping threshold is set to 5.0 to prevent gradient explosion. The batch normalization layer standardizes the features and accelerates the training convergence.

[0053] Taking the manufacturing of wing skin of a large passenger aircraft as an example, the original collected tension data fluctuates in the range of 5-15N, with a sampling frequency of 1000Hz, generating a lot of noise interference. After being decomposed into 4 scale levels by wavelet transform, the high-frequency noise is effectively separated. The singular value decomposition filter reduces the singular value from the original 127 to 43, retaining the main feature information. After LSTM network processing, the tension prediction error is reduced from the initial ±0.5N to ±0.08N. After further processing by the multi-head attention mechanism (8 attention heads), the prediction lead time is increased to 200ms when involving sudden working conditions. Finally, through Adam optimizer tuning, the dynamic tension control error is stabilized within the range of ±0.05N. The entire processing process realizes the denoising, feature extraction, timing prediction and precise control of tension data, providing a reliable data basis for subsequent speed planning and shear control.

[0054] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0055] (1) Performing noise reduction and feature extraction operations on dynamic tension control data through an autoencoder to output a tension feature dimension reduction data set;

[0056] (2) The tension feature dimension reduction data set is fuzzy quantized through the triangular membership function, and the fuzzy rule base matrix is ​​constructed through the Mamdani reasoning mechanism operation;

[0057] (3) The fuzzy rule base matrix is ​​optimized and screened by genetic algorithm, and the crossover and mutation operations are integrated to generate the optimal fuzzy inference rule set;

[0058] (4) The optimal fuzzy inference rule set is calculated through online learning of the improved BP neural network, combined with the Dropout random inactivation strategy to form a speed compensation feature vector;

[0059] (5) The speed compensation feature vector is input into the convolutional neural network to extract the speed curve features, and the speed control response sequence is constructed after multi-layer pooling and full connection operations;

[0060] (6) The speed control response sequence is optimized by back propagation algorithm based on particle swarm optimization, and the dynamic compensation model of the retraction and extension speed is output after error feedback iteration.

[0061] Specifically, when the autoencoder processes the dynamic tension control data, the data is input into the encoder. The encoder is composed of a multi-layer neural network, with 256 neurons in the first layer and 128 neurons in the second layer, and finally compresses the data into a 64-dimensional latent space. The decoder, on the contrary, reconstructs the data through a three-layer neural network and finally outputs data with the same dimension as the input. By minimizing the reconstruction error, the autoencoder learns the core features of the data, removes noise interference, and outputs a tension feature reduced dimension data set. The obtained tension feature reduced dimension data set is then fuzzy quantized by a triangular membership function. The triangular membership function maps continuous values ​​to fuzzy sets, and each variable is divided into three fuzzy subsets of low, medium, and high. The Mamdani reasoning mechanism constructs IF-THEN rules based on these fuzzy subsets to form an initial fuzzy rule base. The antecedent part of the rule describes the state of the input variable, and the consequent part corresponds to the control output. The entire rule base is stored in the form of a matrix. The fuzzy rule base matrix is ​​optimized by a genetic algorithm. The initial population size is set to 100, and each individual code represents a set of rule combinations. The crossover operation uses double-point crossover with a crossover probability of 0.8; the mutation operation uses uniform mutation with a mutation probability of 0.1. The performance of each rule combination is evaluated by the fitness function, and excellent individuals are selected to enter the next generation. After 500 generations of evolutionary optimization, the optimal fuzzy inference rule set is screened out. The improved BP neural network is used to process the optimal fuzzy inference rule set. The network structure includes an input layer (rule number node), two hidden layers (128 and 64 nodes respectively) and an output layer (control variable node). The Dropout random inactivation strategy randomly turns off 50% of the neurons during the training process to prevent overfitting. The network adopts an online learning mode, updates the weight parameters in real time, and outputs a speed compensation feature vector.

[0062] The speed compensation feature vector is input into the convolutional neural network for further processing. The network contains three convolutional layers, using 32, 64, and 128 convolution kernels, respectively, with a kernel size of 3×3. Each convolutional layer is followed by a maximum pooling layer with a pooling window size of 2×2. Finally, two fully connected layers (containing 512 and 256 nodes, respectively) are used for feature integration to output the speed control response sequence. In the final stage, the back propagation algorithm based on particle swarm optimization is used for parameter optimization. The particle swarm size is set to 200, and each particle represents a set of network parameters. The position update of the particle is based on the individual optimal position and the global optimal position. The inertia weight decreases linearly from 0.9 to 0.4, and the learning factor is set to 2.0. Through error feedback iterative optimization, the dynamic compensation model of the retraction and release speed is finally output.

[0063] Taking the manufacturing of a certain type of composite wind turbine blade as an example, the original dynamic tension control data contains 500 feature dimensions and the data volume is 1000 sample points per second. After being processed by the autoencoder, the feature dimension is reduced to 64 dimensions, and the reconstruction error is less than 3%. The tension value is divided into 7 fuzzy subsets in the fuzzy quantization stage, and the initial rule base contains 343 rules. After genetic algorithm optimization, the number of rules is reduced to 87, and the control accuracy is improved by 30%. In the online learning process of the improved BP neural network, the network parameters are updated every 500ms, and the prediction accuracy reaches 95%. After convolutional neural network processing, the time resolution of the speed control response reaches 10ms. Finally, after parameter adjustment by particle swarm optimization, the dynamic control accuracy of the retracting and releasing speed reaches ±0.2mm / s, and the adaptive compensation time is shortened to 50ms. In the actual production process, this processing flow effectively solves the problem of tension instability caused by speed fluctuations in the composite fiber laying process, ensuring the stability and consistency of the fiber laying quality. Each processing link is based on actual engineering needs, and the effectiveness of the algorithm is verified by quantitative optimization indicators.

[0064] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0065] (1) The retraction and extension speed dynamic compensation model is encoded in state space through the PPO algorithm and combined with multi-layer perceptron mapping to generate the shear action space representation;

[0066] (2) The shear action space representation is passed through a double Q network to estimate the value function, integrate the Monte Carlo tree search strategy, and output the shear state transition probability matrix;

[0067] (3) The shear state transition probability matrix is ​​learned through temporal difference learning through the TD-Lambda algorithm, and the shear strategy gradient vector is constructed by combining the experience replay buffer mechanism;

[0068] (4) The cutting strategy gradient vector is input into the Actor-Critic network architecture and optimized through the soft update strategy to form a multi-tow cutting decision sequence;

[0069] (5) The multi-tow cutting decision sequence is hierarchically decomposed and calculated using a hierarchical reinforcement learning algorithm, and the key cutting time series data is screened out by combining priority sampling technology;

[0070] (6) The key cutting time series data is explored and optimized through the maximum entropy reinforcement learning algorithm, and after being reshaped by the reward function, an intelligent cutting instruction set is output.

[0071] Specifically, in the process of optimizing the automatic fiber placement of composite materials, the dynamic compensation model of the retraction and release speed is first input into the PPO algorithm. The PPO algorithm updates the strategy by probability ratio truncation, encoding the continuous speed compensation data into a discrete state space. The state space contains key parameters such as the position, speed, and tension of the tow, and features are mapped through a three-layer multi-layer perceptron (each layer contains 256, 128, and 64 neurons, respectively), and finally generates a spatial representation of the shearing action.

[0072] The formula for hierarchical decomposition calculation is expressed as:

[0073]

[0074] Where: S represents the input state space, HDC represents the hierarchical decomposition calculation function, L represents the number of levels, n2 represents the number of nodes in the lth layer, β2 represents the weight coefficient of the lth layer, γ i represents the importance factor of node i, represents the state function of node i in layer l, represents the transfer function of node i, δ2 represents the inter-layer coupling coefficient, i represents the node index in the current layer, j represents the node index of the summation operation, represents the state vector of the i-th node in the l-th layer, represents the state vector of the jth node in the lth layer, represents the transition vector of the i-th node in the l-th layer.

[0075] The shear action space representation is then input into the dual Q network for processing. The dual Q network contains two networks with the same structure but independent parameters, which evaluate the action value respectively and suppress over-estimation by taking the minimum value. At the same time, the Monte Carlo tree search strategy is introduced to simulate the future state transition by sampling 1000 times, build a search tree with a depth of 10, and output the shear state transition probability matrix. The TD-Lambda algorithm is used for temporal difference learning of the shear state transition probability matrix. The Lambda parameter is set to 0.8 to balance the weights of current rewards and future rewards. The experience replay buffer size is set to 10000, and 256 samples are randomly sampled each time for batch learning to gradually build the shear policy gradient vector.

[0076] The cutting strategy gradient vector is input into the Actor-Critic network architecture. The Actor network is responsible for generating action strategies, and the Critic network evaluates the state value. The soft update strategy uses an update rate of 0.001 to smoothly update the network parameters through exponential moving average, and finally forms a multi-tow cutting decision sequence.

[0077] The multi-tow cutting decision sequence is hierarchically decomposed and calculated to decompose the complex cutting task into multiple subtasks. The priority sampling technology assigns weights to samples based on the TD error size, prioritizes learning experiences with high TD errors, and screens out key cutting timing data. Finally, the maximum entropy reinforcement learning algorithm is used to explore and optimize the key cutting timing data. The entropy weight coefficient is set to 0.01, which maximizes the expected return while maintaining a moderate policy randomness. The reward function comprehensively considers factors such as cutting accuracy, timing rationality, and energy consumption, and outputs the final intelligent cutting instruction set after reshaping and optimization.

[0078] Taking the manufacturing of a certain type of helicopter rotor blade as an example, the original velocity compensation model contains the placement data of 8 bundles of fibers, which are encoded by the PPO algorithm to obtain a 256-dimensional state space representation. After the dual Q network evaluation, 10 candidate actions are generated for each shearing position, and the dimension of the state transition probability matrix is ​​256×10. The TD-Lambda algorithm optimizes the dimension of the policy gradient vector to 64 dimensions through 1000 rounds of iterative learning. After the Actor-Critic network is processed, the generated multi-tow shearing decision sequence contains the optimal shearing parameters of each bundle of fibers at different times. The hierarchical decomposition divides the entire shearing process into three levels: preparation, execution, and verification, and selects 752 key timing points from the original 8000 decision points. Finally, after optimization by the maximum entropy reinforcement learning algorithm, the intelligent shearing instruction set achieves a shearing accuracy of ±0.15mm, and the shearing response time is reduced to 20ms, which greatly improves the manufacturing efficiency and quality stability of composite components.

[0079] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0080] (1) The intelligent cutting instruction set is used to extract spatiotemporal features through a recursive neural tensor network, and combined with the attention mechanism operation to generate a feature point cloud of the retransmission trajectory;

[0081] (2) The feature point cloud of the re-delivery trajectory is subjected to comparative learning calculation through a twin convolutional neural network, combined with a three-dimensional space transformation network, and the re-delivery trajectory coding sequence is output;

[0082] (3) The retransmission trajectory encoding sequence is processed by a graph convolutional network and integrated with edge feature extraction operations to construct a retransmission trajectory topology map;

[0083] (4) Inputting the re-delivery trajectory topology map into the simulation environment based on the physics engine and calculating it through the Verlet integration algorithm to form a re-delivery dynamics parameter set;

[0084] (5) The state of the re-delivery dynamics parameter set is estimated through the self-correcting particle filter algorithm, and the optimal solution of the re-delivery trajectory is screened out by combining the Kalman smoothing process;

[0085] (6) The uncertainty of the optimal solution of the re-delivery trajectory is quantified through the variational inference algorithm, and the re-delivery path matrix is ​​output after the Bayesian posterior distribution calculation.

[0086] Specifically, the intelligent cutting instruction set is input into the recursive neural tensor network for processing. The recursive neural tensor network adopts a three-dimensional tensor structure, and each neuron contains a weight tensor of 128×128×128 to capture high-order feature associations. Temporal features are extracted through time recursive connections, and spatial features are extracted through spatial convolution operations. The attention mechanism adopts a multi-head structure, with 8 attention heads, each with an attention dimension of 64. The features at different positions are weighted and aggregated, and finally a feature point cloud of the retransmitted trajectory containing spatial position and temporal information is generated. The feature point cloud of the retransmitted trajectory is then input into the twin convolutional neural network. The twin network consists of two convolution branches with shared weights, each of which contains 5 convolution layers, the convolution kernel size decreases from 7×7 to 3×3, and the number of channels increases from 32 to 256. Contrastive learning uses a triplet loss function to bring similar trajectories closer and push different trajectories farther away. The three-dimensional spatial transformation network contains three transformation modules: rotation, translation and scaling. Each module predicts the transformation parameters by a fully connected layer to achieve spatial alignment of features and finally outputs a standardized re-sent trajectory encoding sequence.

[0087] The re-delivery trajectory encoding sequence is subjected to deep feature extraction through a graph convolutional network. The graph convolutional network constructs the trajectory sequence into a directed graph structure, where nodes represent trajectory points and edges represent the connection relationship between points. The network contains three layers of graph convolutional layers, and the number of output channels of each layer is 64, 128, and 256, respectively. Edge feature extraction uses dynamic edge convolution operations to adaptively learn the connection weights between nodes and construct a complete re-delivery trajectory topology map. The re-delivery trajectory topology map is input into a simulation environment based on a physics engine. The physics engine considers physical constraints such as gravity, friction, and elasticity, and calculates the motion state of the trajectory points through the Verlet integration algorithm. The integration time step is set to 0.001 seconds. Each time step ensures numerical stability through iterative updates of position and velocity, and outputs a set of re-delivery dynamics parameters containing information such as position, velocity, and acceleration.

[0088] The re-delivery dynamics parameter set is estimated by the self-correcting particle filter algorithm. The particle filter algorithm uses 1000 particles to represent the state distribution, and each particle contains complete dynamics parameters. The self-correcting mechanism dynamically adjusts the particle weights according to the observed data to increase the sampling density in the high probability area. The Kalman smoother uses a forward-backward algorithm to perform bidirectional smoothing on the state sequence, eliminate noise fluctuations, and screen out the optimal solution for the re-delivery trajectory. The optimal solution for the re-delivery trajectory is quantified for uncertainty by the variational inference algorithm. Variational inference approximates the posterior distribution to a Gaussian distribution family and iteratively optimizes the distribution parameters by minimizing the KL divergence. The Bayesian posterior calculation uses the Monte Carlo sampling method with a sample size of 10,000 to calculate the uncertainty interval of the trajectory parameters and finally output the re-delivery path matrix with confidence information.

[0089] Taking the manufacturing of the fuselage skin of a large civil airliner as an example, the intelligent cutting instruction set contains the cutting information of 8 bundles of fibers, each of which has 500 key timing points. After processing by the recursive neural tensor network, point cloud data with 50,000 feature points is generated. After comparative learning of the twin network, the feature dimension is compressed to 1024 dimensions, and the spatial transformation network achieves an alignment accuracy of ±0.05mm. The topological map constructed by the graph convolutional network contains 2000 nodes and 8000 edges, and the edge feature dimension is 256. The physical engine simulation completes 10,000 steps of integral calculation in 1 second to generate a complete dynamic trajectory. After particle filtering processing, the average error of state estimation is reduced to 0.1mm, and Kalman smoothing controls the trajectory jitter amplitude within 0.05mm. The final output retransmission path matrix has a dimension of 8×500×6, containing the three-dimensional position and attitude information of each timing point, and is accompanied by a 95% confidence interval.

[0090] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0091] (1) The retransmission path matrix is ​​layered in three-dimensional space to perform multi-dimensional segmentation on the temperature field data, and a temperature data cube is generated according to the X, Y, and Z axes. The temperature data cube is then downsampled and converted to output a basic temperature field data set;

[0092] (2) Perform high-dimensional numerical discretization on the basic data set of the temperature field, convert the continuous temperature surface into a node matrix, and then perform sparse processing on the node matrix to construct a multi-level node data array of the temperature field;

[0093] (3) The temperature field multi-level node data array is subjected to thermodynamic feature extraction operation, and after energy conservation conversion processing, a temperature multi-dimensional feature data stream is output;

[0094] (4) Perform spatiotemporal serialization processing on the temperature multidimensional characteristic data stream, rearrange the data according to the time axis, superimpose the spatial position information labels, and generate a four-dimensional evolution surface of the temperature field;

[0095] (5) Perform temperature gradient analysis on the four-dimensional evolution surface of the temperature field through core feature extraction and threshold screening to output the temperature key control point cloud;

[0096] (6) The temperature key control point cloud is converted into a piecewise continuous control signal sequence, and after smoothing and outlier correction, the temperature field dynamic control sequence is output.

[0097] Specifically, the temperature field information in the retransmission path matrix is ​​processed in three-dimensional space. Specifically, data segmentation is performed in three orthogonal directions, X, Y, and Z, and the resolution of each direction is set to 0.1 mm. A temperature data cube is constructed within the actual workpiece size range (such as 500 mm × 300 mm × 50 mm). For each spatial point, its corresponding temperature value is recorded to form a high-dimensional temperature distribution data set. In order to reduce the computational burden, the temperature data cube is downsampled, and the sampling interval is set to 1 mm. While maintaining the temperature field distribution characteristics, the data volume is greatly reduced, and the temperature field basic data set is finally output. The temperature field basic data set is then processed by high-dimensional numerical discretization. First, the continuous temperature distribution surface is converted into a discrete node matrix with a node spacing of 1 mm. Each node records the temperature value and its spatial coordinates. Considering the continuity and smoothness of the actual temperature field distribution, the node matrix is ​​sparsely processed, and the relative temperature gradient threshold method is used to delete the nodes with insignificant temperature changes (areas with temperature gradients less than 2 ° C / mm), and the dense sampling points in the area of ​​​​drastic temperature changes are retained, and finally a multi-level node data array of the temperature field is constructed. The multi-level node data array of the temperature field is processed by thermodynamic feature extraction operations. First, the heat flux and thermal diffusion coefficient of each node are calculated, taking into account the thermal conductivity of the material (such as the transverse thermal conductivity of carbon fiber is 0.8W / (m·K), and the longitudinal thermal conductivity is 8W / (m·K)). At the same time, the influence of thermal radiation and convection heat transfer is considered, and the energy balance equation under the boundary conditions is calculated. Under the premise of ensuring energy conservation, the temperature field is feature extracted to obtain a temperature multidimensional feature data stream containing information such as heat flux density, temperature gradient, and thermal resistance. The temperature multidimensional feature data stream is time-space serialized, with a sampling time interval of 0.1 seconds, and the complete three-dimensional temperature field distribution is recorded at each time point. The data is rearranged according to the time axis, and a spatial position label is added to each data point, including absolute coordinates and relative position information. Through the combination of time and space dimensions, a four-dimensional evolution surface of the temperature field is generated, which clearly shows the change law of the temperature field over time and space.

[0098] The four-dimensional evolution surface of the temperature field is processed through core feature extraction operations, focusing on analyzing the temperature gradient changes. The temperature gradient values ​​in all directions are calculated to identify areas with drastic temperature changes (gradient greater than 5°C / mm) and areas with stable temperature (gradient less than 1°C / mm). By setting the temperature gradient threshold (3°C / mm) for screening, the key temperature control points that have a significant impact on the quality of the wire laying are extracted to form a temperature key control point cloud.

[0099] Finally, the temperature key control point cloud is converted into a real-time control signal. The discrete control points are connected using the cubic spline interpolation algorithm to generate a smooth temperature control curve. Abnormal temperature points (deviating from the mean by more than 3 standard deviations) are corrected to ensure the continuity and reliability of the control signal. The final output temperature field dynamic control sequence contains complete timing control instructions to guide the real-time adjustment of the heating system.

[0100] For example: in the wire laying process of the barrel part (2 meters in diameter and 3 meters in length), the workpiece is first divided into 200,000 spatial grid points, and the temperature data is recorded at each point. After downsampling, 20,000 temperature sampling points are retained to form the basic data set of the temperature field. The discretization process converts the continuous temperature field into 4,000 key nodes, each of which contains temperature values ​​and three-dimensional coordinate information. In the process of thermodynamic feature extraction, the thermal conductivity (0.8-8W / (m·K)) and heat flux density (0-500W / m 2 ). The spatiotemporal serialization process recorded the temperature changes during the 8-hour fiber laying process, with a sampling interval of 0.1 seconds, generating 288,000 time series data points. Through temperature gradient analysis, it was found that the temperature gradient in the fiber crossing area reached 6°C / mm, while the temperature gradient in the flat area was only 0.5°C / mm. Finally, 500 key temperature control points were selected, and a smooth control sequence was generated through cubic spline interpolation, with a control accuracy of ±2°C.

[0101] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0102] (1) The temperature field dynamic control sequence is processed through multi-dimensional constraint decoupling operations, the time dimension, space dimension, and temperature dimension are layered and decomposed, and a multi-objective feature vector group is output;

[0103] (2) Regularize the multi-objective feature vector group, perform data correction according to physical constraints, and generate a constraint boundary matrix;

[0104] (3) Perform multi-scale feature decomposition on the constraint boundary matrix to extract the dynamic parameter change rules and construct the strategy optimization feature space;

[0105] (4) The strategy optimization feature space is quantified through the reward and punishment function, and the weights of each parameter are numerically assigned to form a multi-objective optimization data stream;

[0106] (5) Perform hierarchical processing on the multi-objective optimization data stream and output the placement optimization parameter set through a dynamic threshold screening mechanism;

[0107] (6) The placement optimization parameter set is strategically integrated, the control instructions are time-reorganized, and the placement strategy data packet is output.

[0108] Specifically, the temperature field dynamic control sequence is subjected to multi-dimensional constraint decoupling processing. In the time dimension, the continuous control sequence is sampled at a time interval of 0.1 seconds, and the temperature control value and change rate at each time point are recorded. In the spatial dimension, the three-dimensional space is divided into 10mm×10mm×1mm grid units, and each unit records its position coordinates and temperature distribution. In the temperature dimension, the temperature value is discretized at intervals of 5℃ to form a temperature grade sequence. Through data dimensionality reduction and feature extraction, the information of the three dimensions is integrated into a multi-objective feature vector group. Regularization processing is performed on the multi-objective feature vector group. First, the data is standardized and transformed so that the data of different dimensions are distributed in the same numerical range. The physical constraints include the upper temperature limit (350℃), the temperature change rate limit (±10℃ / s), and the spatial temperature gradient constraint (≤5℃ / mm). According to these constraints, the data points that are out of range are corrected to ensure that all data meet the physical feasibility. The corrected data is reorganized into a constraint boundary matrix, where each row of the matrix represents a set of constraints and each column corresponds to a control parameter.

[0109] The constraint boundary matrix is ​​decomposed into different frequency components by multi-scale feature decomposition operation using wavelet transform. For high-frequency components, the characteristics of rapid temperature changes are extracted; for low-frequency components, the long-term temperature change trend is focused on. By analyzing the change rules of parameters at different scales, the control rules are summarized, such as the heating power curve in the temperature rise stage and the temperature fluctuation range in the stable stage. These rules are encoded into the feature space and form the basis for strategy optimization. The strategy optimization feature space is then quantified by the reward and punishment function. The reward factor and the penalty factor are set. The reward factor is for the ideal control effect (such as good temperature stability and high bonding strength of adjacent layers), and the penalty factor is for the undesirable conditions (such as excessive temperature causing material damage and low temperature affecting interlayer bonding). According to the degree of influence of different parameters on the final quality, weight coefficients are assigned to generate multi-objective optimization data streams. The weight allocation uses the hierarchical analysis method to quantify the importance of each parameter into specific values.

[0110] The multi-objective optimization data stream is processed hierarchically and divided into multiple levels according to the priority of the control objectives. The highest priority includes temperature stability and material safety, the second priority includes energy efficiency and process cycle, and finally the auxiliary optimization indicators. Each level sets a corresponding dynamic threshold, which is adaptively adjusted with the process stage. Through this multi-level screening mechanism, each parameter is gradually optimized, and finally a set of placement optimization parameter sets that meet all requirements is output. The placement optimization parameter set is finally processed by strategy integration. The discrete optimization parameters are rearranged in chronological order to construct a complete control strategy sequence. The control instructions at adjacent time points are smoothed to eliminate mutation points and ensure the continuity of control. At the same time, the coupling relationship between different control parameters is considered, the control instructions are coordinated and optimized, and finally the placement strategy data packet is output.

[0111] Take the manufacturing of a certain type of composite space cabin as an example: the original temperature field dynamic control sequence contains 8 hours of temperature data, with a sampling frequency of 10Hz, generating 288,000 temperature control points. After multi-dimensional decoupling, 2880 key time points are retained in the time dimension (sampling interval is 10 seconds), the space dimension is divided into 1000 control units, and the temperature dimension is divided into 70 levels (0-350℃, interval 5℃). Regularization processing standardizes all parameters to the [-1, 1] interval, and 85% of valid data points are retained through physical constraint screening. Multi-scale feature decomposition extracts typical control modes, such as preheating stage (0-50℃ / min), working stage (±2℃ / min), and cooling stage (-20℃ / min). The reward and punishment function quantifies the weight of 0.4 for temperature stability, 0.3 for material performance, 0.2 for energy efficiency, and 0.1 for other factors. Dynamic threshold screening optimizes 1000 control units into 200 key control points. The final generated placement strategy data package achieved a temperature control accuracy of ±2°C, the interlayer bonding strength reached 98% of the design requirements, and the energy utilization efficiency was improved by 15%.

[0112] The above describes the optimization method for automatic fiber placement molding of composite materials in the embodiment of the present application. The following describes the optimization system for automatic fiber placement molding of composite materials in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the composite material automatic wire placement forming optimization system includes:

[0113] An input module is used to process the prepreg tow tension data input by the tension sensor through LSTM deep learning calculation and output dynamic tension control data;

[0114] A calculation module, used for inputting the dynamic tension control data into a fuzzy neural network calculation engine, and obtaining a dynamic compensation model for the retraction and release speed after speed curve analysis and calculation;

[0115] A calculation module, for calculating the multi-tow cutting sequence based on the retraction and release speed dynamic compensation model using a reinforcement learning engine to generate an intelligent cutting instruction set;

[0116] A construction module is used to construct a re-delivery trajectory mapping relationship based on the intelligent cutting instruction set, and form a re-delivery path matrix after iterative calculation by the digital twin engine;

[0117] Establishing a module for establishing a temperature field distribution prediction model according to the retransmission path matrix, and constructing a temperature field dynamic control sequence after finite element analysis and calculation;

[0118] The output module is used to substitute the temperature field dynamic control sequence into the multi-objective constraint equation, and output the placement strategy data package after deep reinforcement learning iterative optimization.

[0119] Through the collaborative cooperation of the above components, the prepreg tow tension data is processed through LSTM deep learning, and the fuzzy neural network calculation engine is combined to realize the precise analysis of the speed curve, which improves the accuracy and stability of tension control. The reinforcement learning engine optimizes the shearing sequence of multiple tows, and realizes the precise mapping of the re-feeding trajectory through the iterative calculation of the digital twin engine, which greatly improves the shearing positioning accuracy and re-feeding reliability. The temperature field distribution prediction model combines finite element analysis technology to realize dynamic and precise control of the temperature field, ensuring the temperature uniformity and stability during the laying process. Through deep reinforcement learning, the dynamic control sequence of the temperature field is optimized with multi-objective constraints, and the global optimization of the laying strategy is realized, which significantly improves the quality stability of the laying process. The entire solution realizes the coordinated optimization of multiple key parameters such as tension control, speed compensation, shear positioning, and temperature control during the laying process through multi-level data processing and intelligent algorithm optimization, effectively solving the problems of improper parameter matching and process instability in traditional laying technology. At the same time, the intelligent optimization method based on deep learning and reinforcement learning enables the control strategy to be adaptively adjusted to meet the laying needs under different working conditions, improving the flexibility and intelligence level of the production process. In addition, the introduction of digital twin technology makes laying trajectory planning more precise and temperature field prediction more accurate, providing reliable technical support for high-quality wire laying manufacturing. Through multi-objective constraint optimization, while ensuring product quality, it also achieves improved energy efficiency and reduced production costs, with significant economic benefits and practical value.

[0120] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer, and when the instructions are executed on a computer, the computer executes the steps of the composite material automatic wire placement molding optimization method.

[0121] The present application also provides a wire placing machine, which includes: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement an optimization method for automatic wire placing and forming of composite materials.

[0122] The wire laying machine provided in this application is as follows: Figure 3 and Figure 4 As shown, it is a first schematic diagram and a second schematic diagram of the wire laying machine provided in this application, Figure 3 The invention comprises: a wire guide wheel 31, a clamping device 84, a re-feeding device 75, and an infrared heating lamp 96. Figure 4 It includes: a connecting flange 153, a control system and a tension motor device 166, a carbon fiber prepreg 129, and a fiber placement device 53.

[0123] The fiber laying machine can only lay thermosetting and thermoplastic carbon fiber prepregs. 8 rolls of thermosetting fiber prepreg are installed on the storage roll. Each bundle of fibers is drawn out from the storage roll. After passing through the film collection device, the isolation liner on the surface of the prepreg bundle is recovered by a separate film collection roll. At the same time, each bundle enters the tension control mechanism. The bundle drawn out from the tension control mechanism has a certain tension. Then the bundle passes through an independent clamping mechanism, a separate re-feeding mechanism and a separate shearing mechanism and is finally laid on the mold surface under the action of the heating device and the compacting mechanism to achieve heating, compaction and shaping, and complete the laying process.

[0124] The fiber prepreg itself has a certain degree of stickiness. During the storage of the tow, in order to prevent the tows from sticking to each other, the tows are covered with a layer of isolation liner made of polyethylene material on one side. If the fiber tow is not separated from the isolation liner in time before entering the laying process, the isolation liner rolled into the laying process will affect the tight adhesion between the tow layers, causing quality defects such as pores, seriously reducing the molding quality. Therefore, the isolation liner needs to be recycled. The difficulty of the design is that the feeding of the fiber tow and the recycling of the isolation liner are a dynamic and synchronous process, and the feeding speed of the fiber tow and the recycling speed of the isolation liner should be consistent. As the laying process proceeds, the shaft diameter of the storage roller becomes smaller and smaller, and the shaft diameter of the film winding shaft becomes larger and larger. Therefore, how to maintain the real-time matching of the feeding speed and the film winding speed is a design difficulty.

[0125] The film winding action is a real-time dynamic matching action. The unwinding speed of the fiber bundle and the film winding speed should be basically consistent. If the film winding speed cannot keep up with the unwinding speed, the release liner that has not been separated from the fiber bundle in time will be rolled into the laying process, which will affect the fiber shearing performance at the least and cause serious product quality defects at the worst. If the unwinding speed cannot keep up with the film winding speed, the fiber bundle will accumulate before entering the laying process, affecting the fiber laying efficiency and quality. Our solution for unwinding-film winding is as follows: friction is used to achieve real-time dynamic matching of the film winding speed and the unwinding speed. The fiber bundle is stored on the winding roller in the form of winding, the release liner adheres to the upper surface of the fiber bundle, and the friction roller and the film winding roller have an interference fit. Under the action of tension, the fiber bundle covered with the release liner is drawn out from the unwinding roller and divided into two paths. One path is the fiber bundle passing through the friction roller, and the other path is the release liner wound on the film winding roller. The fiber bundle passing through the friction roller drives the film collecting roller to rotate in the form of friction, so that the isolation liner separated from the fiber bundle can be gradually wound on the film collecting roller to complete the film collecting action. The invention has the advantages of simple structure, easy disassembly and maintenance; it can achieve real-time speed matching of yarn release and film collecting with small fluctuations.

[0126] During the fiber placement process, there are mainly two tension stages: the re-feeding tension in the tow re-feeding stage and the laying tension in the wire laying stage. If the re-feeding tension is too small, the tow will relax during the re-feeding process and easily detach from the guide device; if the re-feeding tension is too large, the tow will easily slip between the roller, reducing the re-feeding accuracy, and a large wire feeding tension means a greater clamping force, which will cause more serious fiber wear. If the laying tension is too small, the product lacks prestress and the mechanical properties are reduced; if the laying tension is too large, it is easy to cause bridging when laying concave surfaces, and even worse, the fiber will be torn. At the same time, if the laying tension fluctuates violently, it will cause different prestresses in various parts of the product, affecting the overall mechanical properties. Therefore, maintaining an appropriate and sufficiently stable tension has an important impact on the quality of wire laying.

[0127] The function of the tension control mechanism is to keep the fiber bundle at a certain tension during the automatic laying process, in order to reduce the tension fluctuation during the laying process. The tension control method can be a combination of active control and passive regulation: the first active method is to provide damping for the laying roller during the fiber laying process, and the resistance is provided by a torque motor or an electronic damper. The closed-loop control of the laying roller resistance is achieved through the tension sensor combined with the system preset value; the second passive method is to apply tension to the tension shaft through the tension cylinder. The cylinder tension is compensated for the tension accuracy of the tension shaft by the electronic control ratio combined with the system preset value. At the same time, precise closed-loop control is implemented for the application and reduction of tension in conjunction with shearing and re-feeding.

[0128] The fiber laying device bundles multiple prepreg bundles into different adjustable lengths. The length adjustment is achieved by controlling the cutting action of any fiber bundle. Therefore, the shearing mechanism must have the function of independently cutting any fiber bundle. The function of the shearing mechanism is to receive the command of the control system to cut the fiber. When cutting, the fiber bundle remains taut under the coordination of tension. To cut the fiber bundle quickly and neatly, the shearing mechanism should meet the following requirements: fast response, quick action, and sufficient shearing force.

[0129] The cutter is the core part of the shearing mechanism. There are two cutting methods: ultrasonic cutting and mechanical cutting. The ultrasonic cutter uses ultrasonic energy to locally heat and melt the material to be cut to achieve the purpose of cutting the material. Ultrasonic cutting can seal the cutting edge, which better avoids the shortcomings of fuzzing and wire drawing at the incision, but it is large in size, difficult to install, and expensive. Mechanical cutting uses a sharp-edged tool to cut the fiber bundle. It is low-cost, easy to install, and can operate in a narrow space. At the same time, the cutting performance can be improved by the reasonable design of the tool and the control of the cutting speed. Considering the strict requirements of the wire laying head for compact structure, pneumatic mechanical cutting is used to minimize the pipeline setting value.

[0130] The shape of the cutting edge has an important influence on the cutting performance. Common cutting edge design schemes include single-sided cutting and double-sided cutting. Single-sided cutting is to grind a bevel on one side of the blade tip and keep the other side flat. The biggest advantage of single-sided cutting is that it can grind a very small bevel angle and the cutting edge is sharp. However, the blade surface is thin and very easy to wear. Double-sided cutting is to grind a bevel on both sides of the blade tip to form a cutting edge. The blade surface has higher strength than a single blade and is more suitable for straight cutting occasions. The shearing of fiber filaments is a process dominated by straight cutting, so a single-sided cutting design is adopted. At the same time, the cutting edge is designed to have a certain slope. Based on a large number of experiments and experience, we set this angle to 5°. It effectively avoids the phenomenon of knife collision and plays a role in protecting the cutter. When 8 tows are laid, the width of each bundle is 6.35mm. If single-filament cutting is to be achieved, a highly integrated cutting device is required. We use cylinders as power and customize double-force cylinders to achieve smooth cutting of tows with a thickness of less than 0.2mm. At the same time, each bundle of fibers is driven by an independent cylinder for shearing.

[0131] The re-feeding mechanism can send the cut ends of the fiber bundles back to the compacting mechanism when laying is required, and then enter the re-laying process. The reliability of the re-feeding mechanism directly affects the re-feeding accuracy. The function of the re-feeding mechanism is to re-transmit the cut fiber ends to the compacting mechanism when the width of the laid fiber bundle needs to be increased. The reliability of the re-feeding mechanism has an important impact on the laying accuracy. In order to meet the requirements of multi-channel re-feeding in a narrow space: ensure that the re-feeding force is appropriate, the fiber bundles neither slip nor break; at the same time, the action response is fast and the reliability is high. The fiber re-feeding scheme is as follows:

[0132] The fiber is dragged in a coaxial multi-position form to complete re-delivery. In this invention, every time the laying head completes a fiber shearing and needs to re-deliver the fiber to the compaction mechanism, the re-delivery motor drives the re-delivery roller by driving the internal and external meshing gears, and drags the cut fiber end to the pressing roller mechanism in cooperation with the wire pressing wheel to complete a re-delivery, and can ensure a re-delivery accuracy of ±0.15mm. The re-delivery pressure wheel keeps the fiber ribbon and the re-delivery wheel in a tightened state in real time under pneumatic drive, and cooperates with the system to analyze the laid model to transport and stop the fiber ribbon. The present invention uses roller extrusion to re-deliver, which can meet the fiber re-delivery requirements in a limited space, and by selecting a reasonable re-delivery wheel material, the fiber re-delivery performance can be effectively enhanced. The fiber re-delivery speed and re-delivery length can be controlled by adjusting the speed and time of the driving wheel. The control is simple and the structure is compact.

[0133] The invention is different from the traditional compaction roller in that it uses an integral pressure roller to replace the independent pressure block, and two sets of micro pressure sensors are installed on the fixed base plate and the moving plate of the pressure roller. The range of each sensor is: 0-400N, the accuracy is 0.02%, and the final pressure is 800N. The sensor can provide real-time feedback on the pressure of the pressure roller, and accurately adjust the compaction force required by the equipment through the gap between the pressure roller and the mold or the layer combined with the system preset value.

[0134] When the laying device is laying, each bundle of wire has a certain tension. After the single wire is cut by the cutting device, the bundle of wires at the rear end needs to be sent back to the pressure roller mechanism as required, but the bundle of wires always maintains a certain tension. When the re-feeding pressure wheel is lifted, the bundle of wires will be retracted. In this way, the next re-feeding will fail, so we added a clamping device at the upstream end of the re-feeding mechanism. After the bundle of wires passes through the clamping system, no matter how the downstream re-feeding mechanism and the shearing mechanism are used together, the bundle of wires cannot be automatically retracted, thus ensuring the next re-feeding. The clamping system consists of a set of fixed wheels and a set of moving wheels. We install a one-way bearing inside the fixed wheel, which causes the bundle of wires to move only in one direction. When the moving wheel group clamps the bundle of wires, the one-way bearing acts as a stop, thus ensuring that the cut fibers will not shrink.

[0135] Heating carbon fiber prepreg during laying is an indispensable part of the process. Without heating, the prepregs cannot fit tightly together, there are protrusions on the laying surface, there are gaps between layers, and the laying surface is very uneven. After heating, the laying quality is significantly improved and the laying surface is flat. Infrared heating refers to the process in which the heated body absorbs radiation energy and converts it into thermal energy after infrared radiation. Infrared heating is actually a special form of resistance heating. In the present invention, three infrared heating lamps are used as light sources. Due to limited installation space, we use a double-tube infrared lamp with a length of 100mm and a single lamp power of 700W. The heating temperature can reach 350℃ at a fixed position, with a temperature error of ±3℃. At the same time, the carbon fiber heats up significantly under the irradiation of the infrared heater.

[0136] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A composite material automatic wire placement molding optimization method, characterized in that: The composite material automatic wire placement forming optimization method comprises: The prepreg tow tension data input by the tension sensor is processed through LSTM deep learning to output dynamic tension control data; The dynamic tension control data is input into a fuzzy neural network calculation engine, and after speed curve analysis and calculation, a dynamic compensation model for the retraction and release speed is obtained; Based on the retracting and releasing speed dynamic compensation model, a reinforcement learning engine is used to calculate the multi-tow cutting sequence and generate an intelligent cutting instruction set; A re-delivery trajectory mapping relationship is constructed based on the intelligent cutting instruction set, and a re-delivery path matrix is ​​formed after iterative calculation by the digital twin engine; A temperature field distribution prediction model is established according to the retransmission path matrix, and a temperature field dynamic control sequence is constructed after finite element analysis and calculation; The temperature field dynamic control sequence is substituted into the multi-objective constraint equation, and after deep reinforcement learning iterative optimization, a placement strategy data package is output.

2. The composite material automatic wire placement molding optimization method according to claim 1, characterized in that: The prepreg tow tension data input by the tension sensor is processed by LSTM deep learning operation to output dynamic tension control data, including: Performing multi-scale time-frequency decomposition on the prepreg tow tension data by wavelet transform and Fourier analysis, and generating a tension multi-dimensional frequency spectrum feature matrix after singular value decomposition filtering; The tension multi-dimensional spectrum feature matrix is ​​input into a three-layer LSTM network structure containing 64 hidden neurons, and after forward recursive calculation and long short-term memory unit state update, a tension spatiotemporal state vector set is output; According to the tension spatiotemporal state vector set, deep temporal correlation mining is performed through a bidirectional gated recurrent unit based on the Transformer architecture, and a tension prediction deep feature map is formed in combination with a residual connection mechanism; The tension prediction deep feature map is calculated by a multi-head attention mechanism, and the self-attention and cross-attention weight distributions are integrated to screen out a high-dimensional tension feature sequence; The high-dimensional tension feature sequence is iteratively optimized in batches by using an improved stochastic gradient descent algorithm, and a tension prediction loss function surface is constructed by combining L1 and L2 regularization constraints; The tension prediction loss function surface is substituted into the Adam optimizer with momentum term, and the dynamic tension control data is output after batch normalization processing through adaptive learning rate adjustment and gradient clipping technology.

3. The composite material automatic wire placement molding optimization method according to claim 1, characterized in that: The dynamic tension control data is input into the fuzzy neural network calculation engine, and after speed curve analysis and calculation, a dynamic compensation model for the retracting and releasing speed is obtained, including: Performing noise reduction and feature extraction operations on the dynamic tension control data through an autoencoder to output a tension feature dimension reduction data set; The tension feature dimension reduction data set is fuzzy quantized by a triangular membership function, and a fuzzy rule base matrix is ​​constructed by Mamdani reasoning mechanism operation; The fuzzy rule base matrix is ​​optimized and screened by a genetic algorithm, and crossover and mutation operations are integrated to generate an optimal fuzzy reasoning rule set; The optimal fuzzy inference rule set is subjected to online learning calculation by an improved BP neural network, combined with a Dropout random deactivation strategy, to form a speed compensation feature vector; The speed compensation feature vector is input into a convolutional neural network to extract speed curve features, and a speed control response sequence is constructed through multi-layer pooling and full connection operations; The speed control response sequence is subjected to parameter optimization calculation by a back propagation algorithm based on particle swarm optimization, and the retraction and extension speed dynamic compensation model is output after error feedback iteration.

4. The composite material automatic wire placement molding optimization method according to claim 1, characterized in that: The method uses a reinforcement learning engine to calculate the multi-tow cutting sequence based on the retracting and releasing speed dynamic compensation model and generate an intelligent cutting instruction set, including: The retracting and releasing speed dynamic compensation model is encoded in state space by PPO algorithm, combined with multi-layer perceptron mapping, to generate a shearing action space representation; The shear action space representation is passed through a double Q network to estimate the value function, and a Monte Carlo tree search strategy is integrated to output a shear state transition probability matrix; The shear state transition probability matrix is ​​subjected to temporal difference learning through the TD-Lambda algorithm, and a shear strategy gradient vector is constructed in combination with an experience replay buffer mechanism; The cutting strategy gradient vector is input into the Actor-Critic network architecture, and a multi-tow cutting decision sequence is formed through soft update strategy optimization; The multi-tow cutting decision sequence is subjected to hierarchical decomposition calculation by a hierarchical reinforcement learning algorithm, and the key cutting time series data is screened out by combining the priority sampling technology; The key cutting time series data is explored and optimized through the maximum entropy reinforcement learning algorithm, and reshaped through the reward function to output the intelligent cutting instruction set.

5. The composite material automatic wire placement molding optimization method according to claim 1, characterized in that: The re-delivery trajectory mapping relationship is constructed based on the intelligent cutting instruction set, and a re-delivery path matrix is ​​formed after iterative calculation by the digital twin engine, including: The intelligent cutting instruction set is subjected to spatiotemporal feature extraction through a recursive neural tensor network, and combined with attention mechanism operation to generate a feature point cloud of the retransmission trajectory; The feature point cloud of the re-delivery trajectory is subjected to comparative learning calculation by a twin convolutional neural network, and combined with a three-dimensional space transformation network, a re-delivery trajectory coding sequence is output; The re-delivery trajectory coding sequence is processed by a graph convolutional network, and an edge feature extraction operation is integrated to construct a re-delivery trajectory topology map; The re-delivery trajectory topology map is input into a simulation environment based on a physical engine, and is calculated by a Verlet integration algorithm to form a re-delivery dynamics parameter set; The state of the re-delivery dynamics parameter set is estimated by a self-correcting particle filter algorithm, and the optimal solution of the re-delivery trajectory is screened out by combining Kalman smoothing. The uncertainty of the optimal solution of the re-delivery trajectory is quantified through a variational inference algorithm, and the re-delivery path matrix is ​​output after Bayesian posterior distribution calculation.

6. The composite material automatic wire placement molding optimization method according to claim 1, characterized in that: The temperature field distribution prediction model is established based on the retransmission path matrix, and after finite element analysis and calculation, a temperature field dynamic control sequence is constructed, including: The retransmission path matrix is ​​multi-dimensionally segmented into temperature field data through three-dimensional space stratification, a temperature data cube is generated according to the X, Y, and Z axes, and then the temperature data cube is downsampled and converted to output a temperature field basic data set; Performing high-dimensional numerical discretization processing on the basic data set of the temperature field, converting the continuous temperature surface into a node matrix, and then performing sparse processing on the node matrix to construct a multi-level node data array of the temperature field; The temperature field multi-level node data array is subjected to thermodynamic feature extraction operation, and energy conservation conversion processing to output a temperature multi-dimensional feature data stream; Performing spatiotemporal serialization processing on the temperature multidimensional characteristic data stream, rearranging the data according to the time axis, superimposing spatial position information labels, and generating a four-dimensional evolution surface of the temperature field; The temperature field four-dimensional evolution surface is subjected to core feature extraction operation, temperature gradient analysis is performed, and then threshold screening is performed to output a temperature key control point cloud; The temperature key control point cloud is converted into a segmented continuous control signal sequence, and after smoothing and outlier correction, the temperature field dynamic control sequence is output.

7. The composite material automatic wire placement molding optimization method according to claim 1, characterized in that: Substituting the temperature field dynamic control sequence into the multi-objective constraint equation, and after deep reinforcement learning iterative optimization, outputting a placement strategy data package, including: The temperature field dynamic control sequence is processed by multi-dimensional constraint decoupling operation, the time dimension, space dimension and temperature dimension are layered and disassembled, and a multi-objective feature vector group is output; Regularizing the multi-objective feature vector group, performing data correction according to physical constraints, and generating a constraint boundary matrix; Performing multi-scale feature decomposition operation on the constraint condition boundary matrix, extracting the dynamic parameter change law, and constructing a strategy optimization feature space; The strategy optimization feature space is quantized through a reward and punishment function, and numerical values ​​are assigned to the weights of each parameter to form a multi-objective optimization data stream; hierarchically processing the multi-objective optimization data stream, and outputting a set of placement optimization parameters through a dynamic threshold screening mechanism; The placement optimization parameter set is subjected to strategic integration processing, the control instructions are reorganized in time sequence, and the placement strategy data packet is output.

8. A composite material automatic wire placement forming optimization system, used to implement the composite material automatic wire placement forming optimization method according to any one of claims 1 to 7, characterized in that: The composite material automatic wire placement forming optimization system comprises: An input module is used to process the prepreg tow tension data input by the tension sensor through LSTM deep learning calculation and output dynamic tension control data; A calculation module, used for inputting the dynamic tension control data into a fuzzy neural network calculation engine, and obtaining a dynamic compensation model for the retraction and release speed after speed curve analysis and calculation; A calculation module, for calculating the multi-tow cutting sequence based on the retraction and release speed dynamic compensation model using a reinforcement learning engine to generate an intelligent cutting instruction set; A construction module is used to construct a re-delivery trajectory mapping relationship based on the intelligent cutting instruction set, and form a re-delivery path matrix after iterative calculation by the digital twin engine; Establishing a module for establishing a temperature field distribution prediction model according to the retransmission path matrix, and constructing a temperature field dynamic control sequence after finite element analysis and calculation; The output module is used to substitute the temperature field dynamic control sequence into the multi-objective constraint equation, and output the placement strategy data package after deep reinforcement learning iterative optimization.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the composite material automatic wire placement forming optimization method according to any one of claims 1 to 7 is implemented.

10. A wire laying machine, characterized in that: The wire laying machine comprises: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the composite material automatic wire placement optimization method as described in any one of claims 1-7.

Images