A continuous fiber composite material 3D printing process parameter optimization method and system

Abstract

The application belongs to the technical field of composite material additive manufacturing, and discloses a continuous fiber composite material 3D printing process parameter optimization method and system, wherein the method comprises the following steps: constructing a first sample set; the samples in the first sample set are combinations of printing parameters and corresponding printing quality indexes and printing efficiency indexes; constructing a prediction model; the prediction model is trained by using the first sample set to obtain the trained prediction model; and parameter optimization: based on the trained prediction model, the combinations of printing parameters are optimized by using a multi-objective optimization algorithm for the purpose of comprehensive optimization of printing quality indexes and printing efficiency indexes, and the optimized printing parameter combinations are obtained. The application can realize real-time and dynamic parameter optimization in the printing process, is beneficial to obtaining the printing parameter combinations with the optimal comprehensive performance in the printing process, and effectively improves the optimization efficiency and optimization effect of the traditional optimization method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field related to additive manufacturing of composite materials, and more particularly, to a continuous fiber composite material 3D printing process parameter optimization method and system. BACKGROUND

[0002] Continuous fiber composite material (CFRP) is a composite material with continuous fibers as reinforcement and high molecular resin material as matrix. Compared with traditional polymer materials, its specific strength, specific stiffness and fatigue resistance have been greatly improved, so it has been widely used in military and civilian industries such as aerospace, automobiles and large ships, which have very high requirements on performance and weight. However, in the early stage of development, due to the limitations of traditional manufacturing processes, it is difficult to produce according to the designed specific complex structure, which greatly limits the application of continuous fiber composite materials. As a revolutionary manufacturing technology, the rapid development of 3D printing (also known as additive manufacturing) has greatly solved the problem of complex structure forming, and the application of continuous fiber composite materials has been greatly developed.

[0003] At the same time, there are many printing process parameters (such as layer angle, nozzle temperature, layer thickness and fiber filling density) in the 3D printing process, and it is difficult to find a comprehensive performance optimal printing parameter combination. In the absence of a large number of experiments and expert experience guidance, it is difficult to find a printing parameter combination that takes into account printing efficiency and printing quality, which greatly restricts the application of continuous fiber composite material 3D printed components in important fields such as aerospace. Therefore, it is very meaningful to predict and optimize the printing parameters in the 3D printing process of continuous fiber composite materials.

[0004] The 3D printing process parameter optimization technology of continuous fiber composite materials develops relatively slowly, and the mainstream method uses performance test results obtained by a large number of experiments as feedback, and then adjusts the parameters one by one. However, the current continuous fiber composite material 3D printing process involves a wide range of parameters and has a coupling effect, and the adjustment of a single parameter cannot meet the optimization requirements. In addition, the traditional optimization method is limited by the performance test of the sample, and the verification period of the optimized parameter combination is long, and feedback can only be obtained after printing and mechanical property testing; if the optimization effect is not up to standard, parameter optimization and sample printing need to be performed again, and real-time dynamic optimization cannot be achieved. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides a continuous fiber composite material 3D printing process parameter optimization method and system, which is used to solve the problem that the traditional 3D printing process parameter optimization method is difficult to realize multi-parameter coupling optimization and the verification period of the optimized parameter combination is long by adjusting the printing parameters to perform printing experiments, and then testing the performance of the sample to select the optimal printing parameter combination.

[0006] To achieve the above-mentioned purpose, according to one aspect of the present application, a continuous fiber composite material 3D printing process parameter optimization method is provided, comprising:

[0007] Constructing a first sample set: the samples in the first sample set are combinations of printing parameters and corresponding printing quality indicators and printing efficiency indicators;

[0008] Constructing a prediction model: the prediction model is used to predict the printing quality indicators and the printing efficiency indicators according to the combinations of the printing parameters, the first sample set is used to train the prediction model, and the trained prediction model is obtained;

[0009] Parameter optimization: based on the trained prediction model, a multi-objective optimization algorithm is used to optimize the combinations of the printing parameters for the purpose of comprehensive optimization of the printing quality indicators and the printing efficiency indicators, the optimized printing parameter combinations are obtained, and printing is performed according to the optimized printing parameter combinations.

[0010] According to the continuous fiber composite material 3D printing process parameter optimization method provided by the present application, it further comprises:

[0011] Constructing a second sample set: the samples in the second sample set are monitoring information in the printing process and corresponding defect information;

[0012] Constructing a printing defect monitoring model: the printing defect monitoring model is used to judge the defect information according to the monitoring information in the printing process, the second sample set is used to train the printing defect monitoring model, and the trained printing defect monitoring model is obtained;

[0013] Online defect monitoring: online monitoring information is obtained during the printing process, and the online defect information is obtained by using the online monitoring information and the trained printing defect monitoring model;

[0014] If the online defect information indicates that there is a defect, a parameter optimization operation is performed.

[0015] According to the continuous fiber composite material 3D printing process parameter optimization method provided by the present application, constructing the first sample set and the second sample set comprises:

[0016] Pre-printing is performed with different combinations of printing parameters, and one layer of a sample part is printed under each combination of printing parameters; and monitoring information is acquired once for each layer of the sample part printed;

[0017] According to the monitoring information, corresponding defect information is acquired, a combination of printing parameters corresponding to no defect is selected, and a printing quality index and a printing efficiency index of a sample part printed under the combination of printing parameters are acquired, and a first sample set is constructed;

[0018] According to the monitoring information, corresponding defect information is acquired, and a second sample set is constructed.

[0019] The continuous fiber composite material 3D printing process parameter optimization method provided by the application further comprises: determining the parameter value space of each printing parameter;

[0020] The printing parameters include multiple of the following: a layer angle, a nozzle temperature, a fiber filling density, a filament feeding speed, a layer thickness, and a fiber printing speed;

[0021] The printing quality index includes at least one of the following: a tensile strength, a bending strength, and a porosity;

[0022] The printing efficiency index includes a printing time.

[0023] In the continuous fiber composite material 3D printing process parameter optimization method provided by the application, the NSGA-II algorithm is used to optimize the combination of printing parameters, the initial population is input into the trained prediction model, and the printing quality index and the printing efficiency index corresponding to each individual in the initial population are predicted; then, based on the predicted printing quality index and the printing efficiency index of each individual, the individuals in the initial population are evaluated through non-dominated sorting and congestion distance; the next generation population is generated based on the evaluation results; and the iteration is sequentially performed, and finally the Pareto optimal solution set is output.

[0024] In the continuous fiber composite material 3D printing process parameter optimization method provided by the application, the optimal objective function w min The distance between each point in the Pareto optimal solution set and the Utopia point is determined, and thus the optimal solution and the calculation expression of the optimal objective function w min are calculated.

[0025]

[0026] Wherein, n is the number of optimization objectives, f cτ (x) is the τth objective value in the cth Pareto solution, f τ (x) is the value of the Utopia point.

[0027] According to the continuous fiber composite material 3D printing process parameter optimization method provided by the application, non-dominated sorting is used to divide individuals into different levels according to the dominated relationship of individuals on multiple targets, and given two individuals A and B, A dominates B is denoted as The mathematical expression is:

[0028]

[0029] Wherein, f m (A) is the value of individual A on target m, and M is the total number of targets.

[0030] The crowding distance is used to measure the distance of an individual in the target space from its adjacent individuals; for target m, the calculation formula of the crowding distance l of individual i is:

[0031]

[0032] Wherein, f m (i+1) and f m (i-1) are the target values of the previous and next individuals of individual i after sorting of target m; And The maximum and minimum values of target m.

[0033] According to the continuous fiber composite material 3D printing process parameter optimization method provided by the application, the monitoring information in the printing process includes multiple of temperature information, surface image information, contact pressure information between the printing nozzle and the deposited layer, acoustic signal information generated by the friction between the printing nozzle and the continuous fiber, and point cloud information of the printed layer.

[0034] The parameter optimization specifically includes: performing parameter optimization at the beginning of printing;

[0035] And performing online defect monitoring once per layer, and selectively performing parameter optimization according to the online defect information.

[0036] According to another aspect of the application, a continuous fiber composite material 3D printing process parameter optimization system is provided, comprising a 3D printing platform; the 3D printing platform comprises a six-axis industrial robot, a printing nozzle, a printing forming platform and an upper computer; the printing nozzle is installed at the end of the six-axis industrial robot, and the upper computer is connected with the six-axis industrial robot, the printing nozzle and the printing forming platform respectively, and the six-axis industrial robot drives the printing nozzle to perform printing forming on the printing forming platform.

[0037] The upper computer comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the continuous fiber composite material 3D printing process parameter optimization method according to any one of the above.

[0038] The continuous fiber composite 3D printing process parameter optimization system provided by the application, the 3D printing platform is also integrated with a sensing component for acquiring monitoring information during printing, the sensing component includes at least one of an infrared camera, a visible light camera, a pressure sensor, an acoustic emission sensor and a laser scanner; the infrared camera and the visible light camera are used to collect temperature information and surface state information during printing; the pressure sensor is located below the printing forming platform and is used to collect contact pressure information between the printing nozzle and the deposited layer during printing; the acoustic emission sensor is located at the position of the upper end of the printing nozzle for feeding, and is used to collect acoustic signal information generated by the friction between the printing nozzle and the continuous fiber; the laser scanner is located below the printing nozzle and is used to collect point cloud information of the printed layer; the upper computer is connected with the sensing component.

[0039] Overall, compared with the prior art, the continuous fiber composite 3D printing process parameter optimization method and system provided by the application have the following advantages:

[0040] 1. A trained prediction model is proposed, and a multi-objective optimization algorithm based on the trained prediction model can realize rapid optimization of the printing parameter combination. This optimization process does not require printing experiments and performance testing of printed samples, can realize real-time dynamic parameter optimization during printing, and can rapidly adjust the printing parameters to significantly improve the optimization efficiency and reduce waste samples to significantly reduce the printing cost. In addition, the coupling optimization of multiple printing parameters can be realized, which is beneficial to obtaining the printing parameter combination with the best comprehensive performance during printing. Multi-objective optimization can realize the simultaneous optimization of printing quality and printing efficiency, solve the problem of mutual restriction and difficulty in optimization of printing quality and printing efficiency in the continuous fiber composite 3D printing process, effectively improve the optimization efficiency and optimization effect of the traditional optimization method, and fill the gap in the field of continuous fiber composite 3D printing parameter optimization;

[0041] 2. The online defect monitoring process and the parameter optimization process are innovatively combined to realize a closed-loop dynamic optimization process of multi-sensor information acquisition, deep learning prediction model learning and training, prediction and optimization of the best printing parameter combination, layer-by-layer monitoring of defects during the printing process, and secondary optimization of the printing parameters according to the discovered defects;

[0042] 3. The proposed closed-loop dynamic optimization system can obtain the printing parameter combination with the best comprehensive performance in one printing process, without repeated parameter adjustment and optimization and sample printing, thereby improving the stability of the continuous fiber composite 3D printing parameter optimization method, realizing the cycle and dynamic optimization of the printing parameters, reducing the optimization cost, and improving the optimization efficiency and effect.

[0043] 4. By integrating sensors such as infrared cameras, visible light cameras, pressure sensors, acoustic emission sensors, and laser scanners on the print nozzle, a 3D printing platform with multi-sensing in-situ online monitoring function is formed, realizing in-situ data acquisition of the continuous fiber composite 3D printing process, and timely feedback to the upper computer to provide reference for subsequent printing parameter optimization related research. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a composition schematic diagram of a continuous fiber composite 3D printing platform with multi-sensing in-situ online monitoring function provided by the present application;

[0045] Figure 2 is a schematic diagram of a continuous fiber reinforced composite material 3D printing process parameter closed-loop dynamic optimization method based on deep learning provided by the present application;

[0046] Figure 3 is a network architecture schematic diagram of a CNN neural network in a specific embodiment provided by the present application;

[0047] Figure 4 is a flowchart of an NSGA-II algorithm in a specific embodiment provided by the present application;

[0048] Figure 5 is a Pareto optimal result diagram of one optimization in a specific embodiment provided by the present application;

[0049] Figure 6 is a comparison diagram used to verify the optimization result in a specific embodiment provided by the present application;

[0050] In all the drawings, the same reference signs are used to represent the same elements or structures, wherein:

[0051] 1-visible light camera; 2-infrared camera; 3-acoustic emission sensor; 4-pressure sensor; 5-laser scanner; 6-print nozzle; 7-upper computer; 8-six-axis industrial robot; 9-print forming platform. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0053] Please refer to Figure 1 and Figure 2The embodiment provides a continuous fiber composite material 3D printing process parameter optimization method.

[0054] A first sample set is constructed: samples in the first sample set are combinations of printing parameters and corresponding printing quality indicators and printing efficiency indicators; the combination of printing parameters is a combination of specific values of different printing parameters, and the printing quality indicator and the printing efficiency indicator are quality indicator values and efficiency indicator values of a printed sample under the printing parameter combination;

[0055] A prediction model is constructed: the prediction model is used for predicting the printing quality indicator and the printing efficiency indicator according to the combination of printing parameters, the first sample set is used for training the prediction model, and a trained prediction model is obtained; the trained prediction model can output predicted values of the printing quality indicator and the printing efficiency indicator according to the printing parameter combination;

[0056] Parameter optimization: based on the trained prediction model, a multi-objective optimization algorithm is used to optimize the combination of printing parameters for the purpose of comprehensive optimization of the printing quality indicator and the printing efficiency indicator, an optimized printing parameter combination is obtained, and printing is performed according to the optimized printing parameter combination. In the optimization process, the object of optimization is the combination of printing parameters, and the purpose of optimization is to make the combination of printing parameters correspond to the comprehensive optimal printing quality indicator and printing efficiency indicator, wherein the printing quality indicator and the printing efficiency indicator corresponding to the printing parameter combination are obtained by the trained prediction model. By using the multi-objective optimization algorithm, the optimal printing parameter combination can be obtained under multiple optimization objectives.

[0057] In the embodiment, the continuous fiber reinforced composite material includes but is not limited to a continuous fiber reinforced resin-based composite material such as carbon fiber, glass fiber, aramid fiber, plant fiber and basalt fiber.

[0058] The prediction model can be a forward prediction deep learning model, and the deep learning model is used to forward predict the mapping relationship between the printing process parameters and the mechanical properties and the manufacturing efficiency. The deep learning model includes but is not limited to a CNN neural network, a Bp neural network, an LSTM neural network, an RNN neural network and an N-BEATS deep learning model. The multi-objective optimization algorithm obtains the optimal printing parameter combination based on the mapping relationship between the printing process parameters and the mechanical properties and the manufacturing efficiency. The multi-objective optimization algorithm includes but is not limited to NSGA-II, MOGWO, MOPSO, SPEA2 and NSGA-III.

[0059] Further, the continuous fiber composite material 3D printing process parameter optimization method provided by the embodiment further includes:

[0060] constructing a second sample set, wherein samples of the second sample set are monitoring information in a printing process and corresponding defect information;

[0061] constructing a printing defect monitoring model, wherein the printing defect monitoring model is used to judge defect information according to monitoring information in a printing process, the second sample set is used to train the printing defect monitoring model, and a trained printing defect monitoring model is obtained;

[0062] online defect monitoring, wherein online monitoring information is obtained in a printing process, the trained printing defect monitoring model and the online monitoring information are used to obtain online defect information;

[0063] If the online defect information indicates that there is a defect, a parameter optimization operation is performed.

[0064] In some embodiments, constructing the first sample set and the second sample set comprises:

[0065] pre-printing is performed with different combinations of printing parameters, wherein one layer of a sample is printed under each combination of printing parameters; and monitoring information is obtained once for each layer of the sample that is printed;

[0066] corresponding defect information is obtained according to the monitoring information, a combination of printing parameters corresponding to no defect is selected, and a printing quality index and a printing efficiency index of the sample printed under the combination of printing parameters are obtained, thereby constructing the first sample set; when the first sample set is constructed, a combination of printing parameters corresponding to no defect or no obvious defect and corresponding index efficiency index are required, and the monitoring information can be an image at this time, and the image can be used to quickly determine whether there is a defect, thereby quickly screening out images with no defect or no obvious defect, and the corresponding combination of printing parameters can be used to construct the first sample set. The index information in the first sample set can be obtained by measuring the printed sample.

[0067] corresponding defect information is obtained according to the monitoring information, thereby constructing the second sample set; when the second sample set is constructed, monitoring information needs to be used to judge printing defect information, such as whether there is a defect, and at this time, various monitoring information can be used to improve the comprehensiveness and accuracy of the judgment.

[0068] Further, the continuous fiber composite material 3D printing process parameter optimization method provided in the embodiment further comprises: determining a parameter value space of each printing parameter;

[0069] The printing parameters include multiple of a layering angle, a nozzle temperature, a fiber filling density, a filament feeding speed, a layer thickness, and a fiber printing speed;

[0070] The printing quality index includes at least one of tensile strength, bending strength and porosity; the printing quality index can be a relevant index capable of reflecting the quality of the printed sample, such as a strength index or a defect index, etc., and multiple indexes can be used, and the specific type and number are not limited.

[0071] The printing efficiency index includes printing time.

[0072] The monitoring information in the printing process includes multiple of temperature information, surface image information, contact pressure information between the printing nozzle and the deposited layer, acoustic signal information generated by the friction between the printing nozzle and the continuous fiber, and point cloud information of the printed layer.

[0073] The parameter optimization specifically includes: performing parameter optimization at the beginning of printing;

[0074] and performing online defect monitoring every time a layer is printed, and selectively performing parameter optimization according to the online defect information. The printing can be paused for a preset time every time a layer is printed to collect the online monitoring information of the printed layer, and to perform online defect monitoring. When the online defect information indicates that there is a defect, parameter optimization is performed to adjust the printing parameters, so that the printing process is performed with the optimized parameter combination.

[0075] Further, the embodiment provides a continuous fiber composite material 3D printing process parameter optimization system, which comprises a 3D printing platform; the 3D printing platform comprises a six-axis industrial robot 8, a printing nozzle 6, a printing forming platform 9 and an upper computer 7; the printing nozzle 6 is installed at the end of the six-axis industrial robot 8, the upper computer 7 is connected with the six-axis industrial robot 8, the printing nozzle 6 and the printing forming platform 9 respectively, and the six-axis industrial robot 8 drives the printing nozzle 6 to perform printing forming on the printing forming platform 9;

[0076] The upper computer 7 comprises a memory and a processor, the memory stores a computer program, and the processor executes the computer program to execute the continuous fiber composite material 3D printing process parameter optimization method of any one of the above embodiments.

[0077] Further, the 3D printing platform is also integrated with a sensing assembly for obtaining monitoring information in the printing process, the sensing assembly including at least one of an infrared camera 2, a visible light camera 1, a pressure sensor 4, an acoustic emission sensor 3 and a laser scanner 5; the infrared camera 2 and the visible light camera 1 are used to collect temperature information and surface state information in the printing process; the pressure sensor 4 is located below the printing forming platform 9 and is used to collect contact pressure information between the printing nozzle 6 and the deposited layer in the printing process; the acoustic emission sensor 3 is located at the position of the upper end feeding of the printing nozzle 6 and is used to collect acoustic signal information generated by the friction between the printing nozzle 6 and the continuous fibers; the laser scanner 5 is located below the printing nozzle 6 and is used to collect point cloud information of the printing layer; the upper computer 7 is connected with the sensing assembly.

[0078] In some specific embodiments, the present embodiment provides a deep learning-based continuous fiber reinforced composite 3D printing process parameter closed-loop dynamic optimization system and method, which includes a multi-sensing in-situ online monitoring system and a dynamic cycle optimization method. The two parts are mutually related and jointly constitute a closed-loop dynamic optimization system integrating prediction, optimization and monitoring, which makes up for the shortcomings of traditional continuous fiber composite 3D printing parameter optimization methods, effectively reduces the defects generated in the printing process, and reduces the time cost and material cost of optimization.

[0079] As shown in Figure 1 The 3D printing platform provided by the present embodiment with multi-sensing in-situ online monitoring function is composed of the following: a six-axis industrial robot 8, a printing nozzle 6, a printing forming platform 9, an upper computer 7, an infrared camera 2, a visible light camera 1, a pressure sensor 4, an acoustic emission sensor 3 and a laser scanner 5; the six-axis industrial robot 8 is the basis of the whole closed-loop dynamic optimization system; the printing nozzle 6 is installed at the end of the six-axis industrial robot 8 and moves under its drive; the infrared camera 2 and the visible light camera 1 are arranged around the printing nozzle 6 and are aligned with the position of the nozzle at the bottom of the printing nozzle 6 to collect temperature information and surface state information in the printing process; the pressure sensor 4 is located below the printing forming platform 9 and is used to collect contact pressure information between the nozzle and the deposited layer in the printing process; the acoustic emission sensor 3 is located at the position of the upper end feeding of the printing nozzle 6 and is used to collect acoustic signal information generated by the friction between the nozzle and the continuous fibers; the laser scanner 5 is located below the printing nozzle 6 and is aligned with the side of the printing layer to collect point cloud information of the printing layer; the upper computer 7 is connected with the six-axis industrial robot 8, the printing nozzle 6 and the printing forming platform 9, and is connected with various sensors, and features of various printing information collected by the above sensors are extracted and fused, and based on this, a deep learning model is used to predict and optimize the printing parameters, and the optimized printing process is monitored online, and real-time closed-loop dynamic optimization is performed.

[0080] The embodiment also provides a printing parameter dynamic cycle optimization method associated with the above-mentioned multi-sensor in-situ online monitoring system, and the method comprises the following steps:

[0081] (1) pre-printing is performed with different printing parameters, and a Python script is used to control the printing nozzle 6 to pause for 3-5 s after printing each layer, so as to facilitate information collection of multiple sensors;

[0082] (2) according to the visible light image information collected in step (1), 30 groups of printing parameter combinations without obvious printing defects are preliminarily screened out;

[0083] (3) the several groups of printing parameters in step (2) are divided into a training set and a test set, which are used to train a CNN neural network prediction model, so that the mapping relationship between the printing parameters and the printing quality and the printing efficiency can be predicted;

[0084] (4) the mapping relationship between the printing parameters and the printing quality and the printing efficiency obtained in step (3) is input into a NSGA-II multi-objective optimization algorithm as a target function, and the optimal printing parameter combination is predicted and optimized by the optimization algorithm;

[0085] (5) the information collected in step (1) is pre-processed to extract feature values and fuse them, and a correspondence between the printing defects is established, and then input into a YOLOv8 model, i.e. a printing defect monitoring model, for learning, so that it can judge whether a defect is generated according to the information fed back by the multiple sensors;

[0086] (6) according to the result judged in step (5), if no defect is generated, the printing is continued, and if a defect is generated, the printing is stopped, and the above-mentioned steps (4)-(5) are repeated, so that the closed-loop dynamic optimization of the 3D printing process parameters of the continuous fiber reinforced composite material can be realized.

[0087] Firstly, according to the monitoring results of the multiple sensors, several groups of printing parameter combinations without obvious printing defects are screened out and divided into a training set and a test set, which are used to train a forward prediction model, and then the mapping relationship between the printing process parameters and the mechanical properties and the manufacturing efficiency obtained by the forward prediction model is used as a target function of a multi-objective optimization algorithm to obtain an optimal solution; finally, the collected data is pre-processed to extract features and input into a defect detection deep learning model for feature fusion, and a corresponding relationship between the fusion results and the printing defects is established to realize defect recognition, which is used for layer-by-layer defect recognition in the printing process of the optimal solution, so that real-time cycle iteration optimization is realized.

[0088] The pretreatment method in step (5) includes but is not limited to image denoising, median filtering, 3D point cloud reconstruction and the like. The defect detection deep learning model performs feature-level fusion on the preprocessed data, and establishes a connection between the printing defects, so that whether a defect is generated can be judged according to the input sensor information in the printing process, including but not limited to YOLOv8, GAN network, GRU gate recurrent unit and the like. Feature-level fusion means that the features of each monitoring information are extracted first, and then feature fusion is performed, and defect monitoring and judgment are performed according to the fused features, and the feature fusion can be in the form of feature splicing, and the specific fusion mode is not limited.

[0089] The optimization method provided by the application can realize a closed-loop dynamic optimization process of multi-sensor information acquisition, CNN neural network learning and training, NSGA-Ⅱ multi-objective optimization algorithm prediction and optimization of the best printing parameters, YOLOv8 model for layer-by-layer monitoring of defects in the printing process, and secondary optimization of the printing parameters according to the discovered defects, as shown in Figure 2 , which can effectively reduce the defects in the printing process and significantly improve the mechanical properties of the formed sample.

[0090] Further, the present embodiment selects five printing parameters, i.e., the layer angle α (°), the nozzle temperature T (°C), the fiber filling density D (%), the layer thickness L (mm) and the fiber printing speed V (mm / s) as the optimization targets.

[0091] Further, the parameter value space of the optimization algorithm given by the present embodiment is shown as follows, and the parameter value space can be set according to the parameter implementation range of the printing equipment, or a better range can be selected according to the experience value, and the specific determination method is not limited.

[0092]

[0093] Further, the 30 groups of printing parameter combinations without obvious printing defects selected by the present embodiment are shown in Table 1.

[0094] Table 1

[0095]

[0096] Further, the tensile strength (σ / MPa) and the printing time (η / min) are selected as the prediction results of the CNN neural network, and are respectively taken as the representative values of the printing quality and the printing efficiency according to the present application.

[0097] Furthermore, CNN convolutional neural network is a feedforward neural network with deep structure and includes convolution calculation. It realizes end-to-end learning through a series of convolution, pooling, full connection and other operations. It has strong generalization and adaptive learning capabilities. The network architecture diagram is as follows Figure 3 This specific embodiment performs forward prediction based on the CNN convolutional neural network, converts the printing parameter data into a feature matrix, and then extracts the input data features through operations such as the constructed convolution layer, batch normalization layer, and activation layer. The network learns and the prediction results are output through the fully connected layer. Its input function is:

[0098] V=conv2(W,X,"valid")+b;

[0099] Where conv2() is the function used for convolution operation in Matlab R2024a, W is the convolution kernel matrix, X is the input matrix, valid is the type of convolution operation, and b is the bias.

[0100] The output function of the CNN convolutional neural network is:

[0101]

[0102] in is the activation function, and V is the above input function. For the last fully connected layer, set to the Lth layer, the output result is y in the form of a vector L , the expected output, that is, the true value, is d, then the total error formula is:

[0103]

[0104] It should be noted that the above input and output functions are for each convolutional layer. Each convolutional layer has a different weight matrix W, and W, X and Y are all in matrix form.

[0105] CNN convolutional neural network is trained by gradient descent and back propagation algorithm. The gradient formula of its convolution layer and pooling layer is:

[0106]

[0107] Among them, E is the total error, W is the convolution kernel matrix, and V is the input function.

[0108] Furthermore, 24 of the 30 data sets in Table 1 were randomly selected as training samples to complete the training of the CNN neural network, and the remaining 6 sets were used as validation sets to verify the model prediction effect. The root mean square error (RMSE) and mean absolute error (MAE) were used to quantitatively characterize the performance of the CNN neural network in this specific embodiment. The calculation expressions of RMSE and MAE are as follows:

[0109]

[0110] where y pred is the predicted value from the trained model, y true is the true value in the validation set, sum is the summation function, and n is the number of elements, is the average of n y true .

[0111] As shown in Table 2, the RMSE value and the MAE value of the CNN neural network in this embodiment on the tensile strength validation set are 0.889 and 1.168, respectively; and the RMSE value and the MAE value on the printing time validation set are 2.733 and 2.569, respectively.

[0112] Table 2

[0113]

[0114] Further, the NSGA-II algorithm is an excellent multi-objective optimization algorithm due to its fast non-dominated sorting, elitist strategy, and crowding distance calculation, which can ensure that the solutions obtained in the multi-objective optimization problem are uniformly distributed and close to the Pareto front. In this embodiment, the NSGA-II algorithm is used to optimize the printing parameters, and the flowchart is as shown in Figure 4 .

[0115] The purpose of non-dominated sorting is to divide individuals into different levels according to the dominance relationship of individuals in multiple objectives. Given two individuals A and B, if A is not worse than B in all objectives and is better than B in at least one objective, it is said that A dominates B (denoted as ), and its mathematical expression is:

[0116]

[0117] where f m (A) is the value of individual A in objective m, and M is the dimension of the objective.

[0118] The crowding distance is used to measure the distance of an individual from its neighboring individuals in the objective space, which is used to maintain the diversity of the population. For each individual, its crowding distance in the objective is calculated. The crowding distance is the difference between the distances of two adjacent individuals in the objective. For objective m, the calculation formula of the crowding distance l of individual i is:

[0119]

[0120] where f m (i+1) and f m(i-1) are the target values of the previous and next individuals of individual i after sorting target m respectively; and The maximum and minimum values of target m.

[0121] Further, the specific optimization steps of the CNN neural network and the NSGA-Ⅱ algorithm are as follows:

[0122] First, the tensile strength (σ / MPa) and the printing time (η / min) are selected as the representative values of the printing quality and the printing efficiency according to the present application. The objective functions of the tensile strength and the printing time and the constraint conditions are defined as follows:

[0123]

[0124] Wherein, minmize_F(α,T,D,L,V) is defined as the combination of 5 printing process parameters that can minimize f σ and f η , f σ and f η are the mapping relationships between the printing process parameters (laid angle α (°), nozzle temperature T (° C), fiber filling density D (%), layer thickness L (mm) and fiber printing speed V (mm / s)) obtained by the CNN neural network and the tensile strength and the printing time, and the printing process parameters are selected in the given parameter space.

[0125] Further, the minimum value of the objective function is obtained by the NSGA-Ⅱ algorithm. The trained CNN neural network prediction model is used to construct the objective functions f σ and f η . In the inverse optimization process of the process parameters, the NSGA-Ⅱ algorithm will automatically generate many initial filled process parameter combinations, including the laid angle, the nozzle temperature, the fiber filling density, the layer thickness and the fiber printing speed. The initial population is input into the objective functions f σ and f η to predict the tensile strength and the printing time; then, based on the objective functions f σ and f η , the initial population is evaluated by non-dominated sorting and crowding distance calculation. According to the non-dominated level and the crowding degree, the excellent part of the initial population is selected to enter the next generation population; the remaining part generates a new population by crossover and mutation, and enters the next generation population; when the size of the next generation population reaches the predetermined size, the inverse model will enter the next iteration. After the iteration process is completed, the Pareto optimal solution set is output. The optimization process established in this embodiment is completed on Matlab R2024a, and the population initialization of the NSGA-Ⅱ algorithm is searched in the given parameter space.

[0126] That is, the parameter optimization adopts the NSGA-II algorithm to optimize the combination of the printing parameters, the NSGA-II algorithm inputs the initial population into the trained prediction model, and is used to predict the printing quality index and the printing efficiency index corresponding to each individual in the initial population; then, based on the predicted printing quality index and the printing efficiency index of each individual, the individuals in the initial population are evaluated by non-dominated sorting and congestion distance; the next generation population is generated based on the evaluation results; and the iteration is performed in sequence, and finally the Pareto optimal solution set is output.

[0127] Further, in the embodiment, the initial population size is set to 50, the crossover probability is 0.85, the variance probability is 0.01, the mutation probability is 0.2, the maximum iteration number is 120, the mapping relationship between the printing parameters and the printing quality and the printing efficiency obtained by the CNN neural network is used as the objective function of the NSGA-II algorithm, and the finally obtained Pareto optimal boundary is as shown in Figure 5 The optimal objective function w min is determined based on the optimal objective function w min , and the calculation expression of the optimal objective function w cτ is as follows:

[0128]

[0129] wherein n is the number of objectives, f cτ (x) is the τth objective value in the c Pareto solutions, f τ (x) is the value of the Utopia point, that is, the optimal point in the ideal; the point in the Pareto optimal solution set that makes the optimal objective function reach the minimum value can be taken as the optimal solution.

[0130] Further, the YOLOv8 model is one of the excellent deep learning models in the field of feature fusion and defect detection, and the YOLOv8 model is selected as the deep learning model for feature fusion and establishing the relationship between the feature fusion result and the printing defect in the embodiment, and is used for real-time monitoring of the entire optimization process. In the verification printing process of the above-mentioned Pareto optimal solution, whether secondary optimization is needed is judged layer by layer according to the established relationship between the feature fusion result and the printing defect.

[0131] Further, in order to prove that each optimization in the embodiment is effective, the optimized process parameters are subjected to actual experiments and corresponding mechanical property tests, and the test results are compared with No. 21 in Table 1 with the best comprehensive performance. The experimental results show that, as the optimized parameters, 0° layup angle, 260℃ nozzle temperature, 80% filling rate, 0.6mm layer thickness and 10mm / s printing speed, the maximum tensile strength of the sample is 144.7MPa, and the printing time is only 77.8min. Compared with No. 21 in Table 1 with the best comprehensive performance, the manufacturing efficiency is improved by about 27% while the mechanical property is improved by about 53%, as shown in Figure 6 .

[0132] The application provides a continuous fiber reinforced composite material 3D printing process parameter closed-loop dynamic optimization system and method based on deep learning. The system is based on a 3D printing platform integrated with an infrared camera 2, a visible light camera 1, a pressure sensor 4, an acoustic emission sensor 3 and a laser scanner 5 and the like multi-sensor, and incorporates a forward prediction model, a multi-objective optimization algorithm and a target detection model and the like deep learning model. The application innovatively combines the online defect monitoring process with the parameter optimization process to form a closed-loop dynamic optimization system integrating prediction, optimization and monitoring. The system solves the problem of mutual restriction between printing quality and printing efficiency in the process of continuous fiber composite material 3D printing, and cooperates with the multi-sensor in-situ monitoring device to detect defects layer by layer, realizes real-time and dynamic cyclic iteration optimization of printing parameters, and can realize real-time and layer-by-layer verification of the obtained optimal parameter combination and give feedback through real-time monitoring, cyclic iteration optimization of printing parameters, and finally realize the printing parameter combination with the best comprehensive performance in one printing process without repeated parameter optimization and sample printing.

[0133] Those skilled in the art will readily understand that the above description is only the preferred embodiment of the application and is not intended to limit the application. Any modification, equivalent replacement and improvement within the spirit and principle of the application shall be included in the protection scope of the application.

Claims

1. A method for optimizing process parameters of 3D printing of continuous fiber composite materials, characterized in that: include: Constructing a first sample set: the samples in the first sample set are combinations of printing parameters and corresponding printing quality indicators and printing efficiency indicators; Constructing a prediction model: the prediction model is used to predict the printing quality index and the printing efficiency index according to the combination of printing parameters, and the prediction model is trained using the first sample set to obtain the trained prediction model; Parameter optimization: With the goal of achieving the best combination of print quality and print efficiency indicators, the combination of printing parameters is optimized using a multi-objective optimization algorithm based on the trained prediction model, the optimized printing parameter combination is obtained, and printing is performed according to the optimized printing parameter combination; Also includes: Constructing a second sample set: the samples of the second sample set are monitoring information during the printing process and corresponding defect information; Constructing the first sample set and the second sample set includes: Perform pre-printing with different printing parameter combinations, where one layer of sample is printed under each printing parameter combination; and obtain monitoring information once each time a layer of sample is printed; Obtaining corresponding defect information based on the monitoring information, selecting a printing parameter combination corresponding to a defect-free state, and obtaining a print quality index and a print efficiency index of a sample printed under the printing parameter combination to construct a first sample set; Obtain corresponding defect information based on the monitoring information and construct a second sample set; The monitoring information during the printing process includes multiple ones of temperature information, surface image information, contact pressure information between the printing nozzle and the deposited layer, acoustic signal information generated by friction between the printing nozzle and the continuous fiber, and point cloud information of the printed layer; Parameter optimization is specifically as follows: parameter optimization is performed at the beginning of printing; And perform online defect monitoring for each printed layer, and selectively optimize parameters based on the online defect information.

2. The method for optimizing process parameters of continuous fiber composite 3D printing according to claim 1, wherein: Also includes: Constructing a printing defect monitoring model: the printing defect monitoring model is used to judge defect information based on monitoring information during the printing process, and the printing defect monitoring model is trained using the second sample set to obtain the trained printing defect monitoring model; Online defect monitoring: obtaining online monitoring information during the printing process, and using the online monitoring information and the trained printing defect monitoring model to obtain online defect information; If the online defect information indicates that a defect exists, a parameter optimization operation is performed.

3. The method for optimizing process parameters for 3D printing of continuous fiber composite materials according to claim 1, wherein: Also includes: Determine the parameter value space of each printing parameter; Printing parameters include multiple ones of layer angle, nozzle temperature, fiber filling density, wire feed speed, layer thickness, and fiber printing speed; The print quality index includes at least one of tensile strength, flexural strength, and porosity; Printing efficiency indicators include printing time.

4. The method for optimizing process parameters for 3D printing of continuous fiber composite materials according to claim 1, wherein: The NSGA-II algorithm is used for parameter optimization to optimize the combination of printing parameters. The NSGA-II algorithm inputs the initial population into the trained prediction model to predict the print quality index and print efficiency index corresponding to each individual in the initial population. Then, based on the print quality index and print efficiency index predicted by each individual, the individuals in the initial population are evaluated through non-dominated sorting and crowding distance. The next generation of population is generated based on the evaluation results. The algorithm is iterated sequentially and finally outputs the Pareto optimal solution set.

5. The method for optimizing process parameters for 3D printing of continuous fiber composite materials according to claim 4, wherein: Based on the optimal objective function Determine each point in the Pareto optimal solution set and Utopia The distance between the points, thereby calculating the optimal solution, the optimal objective function The calculation expression is: ； in, n is the number of optimization objectives, It is c The first of the Pareto solutions target values, yes Utopia The value of the point.

6. The method for optimizing process parameters for 3D printing of continuous fiber composite materials according to claim 4, wherein: Non-dominated sorting is to divide individuals into different levels according to their dominance relationship on multiple targets. Given two individuals A and B ,but A Dominate B Recorded as A B The mathematical expression is: ； in, is an individual A On target m The value on M is the total number of targets; The crowding distance is used to measure the distance between an individual and its neighboring individuals in the target space; m ,individual i The crowding distance l The calculation formula is: ； in, and The targets are m After sorting, individuals i The target values ​​of the previous and next individuals; and Target m The maximum and minimum values ​​of .

7. A continuous fiber composite material 3D printing process parameter optimization system, characterized in that: The 3D printing platform includes a six-axis industrial robot, a print head, a printing platform, and a host computer. The print head is mounted at the end of the six-axis industrial robot. The host computer is connected to the six-axis industrial robot, the print head, and the printing platform, respectively. The six-axis industrial robot drives the print head to perform printing on the printing platform. The host computer includes a memory and a processor, the memory stores a computer program, and the processor executes the continuous fiber composite material 3D printing process parameter optimization method according to any one of claims 1 to 6 when executing the computer program.

8. The continuous fiber composite material 3D printing process parameter optimization system according to claim 7, characterized in that: The 3D printing platform is further integrated with a sensor component for obtaining monitoring information during the printing process, and the sensor component includes at least one of an infrared camera, a visible light camera, a pressure sensor, an acoustic emission sensor, and a laser scanner; the infrared camera and the visible light camera are used to collect temperature information and surface condition information during the printing process; the pressure sensor is located below the printing platform and is used to collect contact pressure information between the print head and the deposited layer during the printing process; The acoustic emission sensor is located at the feeding position at the upper end of the print nozzle, and is used to collect the acoustic signal information generated by the friction between the print nozzle and the continuous fiber; the laser scanner is located below the print nozzle, and is used to collect point cloud information of the printed layer; the host computer is connected to the sensor component.

Images