A nozzle shape adjustable continuous fiber reinforced composite 3D printing system and method

Abstract

The present application belongs to the technical field of composite material additive manufacturing, and discloses a continuous fiber reinforced composite material 3D printing system and method with adjustable nozzle shape, which comprises an adjustable nozzle structure, the adjustable nozzle structure comprising a nozzle head, a hose and a driving structure, the nozzle head being connected and communicated with a heating cavity for heating printing raw materials through the hose, and the driving structure being in transmission connection with the nozzle head and used for driving the nozzle head to move to adjust the nozzle shape and realize the adjustment of the discharge angle; printing is carried out in the nozzle shape obtained by pre-optimization. The present application proposes to connect the nozzle head and the heating cavity by the hose, so that the adjustment and flexible setting of the nozzle shape, i.e. the discharge angle, can be realized, the nozzle shape can be taken as a parameter to be optimized for pre-optimization, the probability of fiber breakage in the printing process can be effectively reduced, the deficiency of the fixed nozzle angle in the traditional continuous fiber composite material printing is made up, and it is beneficial to reduce the defects generated in the printing process and improve the printing quality.

Description

Technical Field

[0001] This invention belongs to the technical field of additive manufacturing of composite materials, and more specifically, relates to a continuous fiber reinforced composite material 3D printing system and method with adjustable nozzle shape. Background Technology

[0002] Continuous carbon fiber reinforced composites (CCFRCs) are widely used in high-precision fields such as aerospace and automotive manufacturing due to their excellent specific strength, specific stiffness, and fatigue resistance. However, traditional 3D printing technology suffers from two key bottlenecks: lagging quality inspection and passive process control. Regarding quality inspection, existing technologies mainly rely on non-destructive testing methods such as X-rays and ultrasound. These methods not only require expensive specialized equipment (such as industrial CT systems), but also involve complex and cumbersome inspection processes, demanding extremely high levels of expertise from operators. More importantly, they cannot achieve in-situ real-time monitoring during the manufacturing process, leading to a significant increase in the time and economic costs of subsequent part inspection and defect repair.

[0003] Current technical approaches for optimizing the parameters and processes in 3D printing of continuous fiber composite materials are relatively limited. Mainstream methods rely on performance test results obtained from numerous experiments as feedback, followed by individual parameter adjustments. However, this process involves many parameters with interdependent interactions, making adjustments to a single parameter insufficient to meet optimization requirements. Furthermore, traditional optimization methods are limited by prototype performance testing; feedback is only available after printing and mechanical property testing are completed. If the optimization results are unsatisfactory, parameter optimization and prototype printing must be repeated, making real-time dynamic optimization impossible.

[0004] Furthermore, the applicant discovered in their research that current 3D printing equipment commonly employs a fixed-angle nozzle design, which presents significant technological limitations. During printing, because the nozzle exit direction remains constant, when the printing path needs to change direction, the extruded continuous fiber bundle must undergo a considerable degree of forced bending deformation to reach the preset position. This unnatural bending state generates significant stress concentration within the fiber, especially when printing complex curved surfaces or sharp-angled transition areas, where the bending stress on the fiber is even more pronounced. Excessive mechanical bending not only leads to the breakage of individual fibers but also causes fiber bundle dispersion and orientation disorder, severely disrupting the uniform distribution of fibers in the resin matrix. This microstructural defect is directly reflected in macroscopic mechanical properties, significantly reducing key indicators such as interlaminar shear strength, tensile modulus, and impact resistance of the formed component. Simultaneously, the free ends generated by fiber breakage become sources of stress concentration, further weakening the fatigue life and service reliability of the formed component. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a continuous fiber reinforced composite material 3D printing system and method with adjustable nozzle shape. This system solves the problem that existing 3D printing commonly uses a fixed-angle nozzle design, which forces the extruded continuous fiber bundles to undergo significant forced bending deformation to reach the preset position, easily leading to fiber breakage and thus weakening part performance and affecting printing quality.

[0006] To achieve the above objectives, according to one aspect of the present invention, a continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape is provided, comprising an adjustable nozzle structure, the adjustable nozzle structure including a nozzle head, a hose and a drive structure, wherein the nozzle head is connected and communicates with a heating chamber for heating the printing material via the hose, and the drive structure is driven to the nozzle head for driving the nozzle head to move to adjust the nozzle shape to achieve adjustment of the material output angle; during printing, printing is performed using a pre-optimized nozzle shape.

[0007] According to the 3D printing system for continuous fiber-reinforced composite materials with adjustable nozzle shape provided by the present invention, the driving structure includes two driving motors, the output shaft of one driving motor is respectively connected to one end of two traction ropes, the other end of the two traction ropes is connected to the nozzle head and is symmetrically distributed along a first direction; the output shaft of the other driving motor is respectively connected to one end of two other traction ropes, the other end of the two other traction ropes is connected to the nozzle head and is symmetrically distributed along a second direction; the first direction and the second direction are perpendicular to each other.

[0008] The adjustable nozzle structure of the continuous fiber-reinforced composite material 3D printing system according to the present invention further includes at least one of the following structures: A return spring is fitted around the outside of the hose. One end of the return spring is connected to the heating chamber, and the other end is connected to the nozzle head. A guide wheel is fixedly provided near the nozzle head, and the guide wheel is provided in a one-to-one correspondence with the traction rope for passing through the traction rope; The output shaft of the drive motor is integrally fitted with a bushing, and the bushing is provided with a traction seat for connecting the traction rope; The nozzle head is provided with a traction lug for connecting a traction rope.

[0009] The 3D printing system for continuous fiber reinforced composite materials with adjustable nozzle shape provided by the present invention further includes a parameter optimization module. The parameter optimization module optimizes the combination of printing parameters and nozzle shape based on a pre-trained input-output response surface model. The input of the input-output response surface model is the combination of printing parameters and nozzle shape, and the output is the printing quality index and / or printing efficiency index. The printing parameters include multiple parameters such as layup angle, nozzle temperature, fiber filling density, layer thickness, and fiber printing speed; the nozzle shape includes nozzle bending angle and / or hose curvature radius, where the nozzle bending angle is the angle between the tangential direction of the hose near the nozzle end and the vertical direction.

[0010] The nozzle shape adjustable continuous fiber reinforced composite material 3D printing system provided by the present invention has the following specific calculation formula for the nozzle shape: ; ; Wherein, the nozzle bending angle is α, the hose curvature radius is ρ, and the rotation angles of the two drive motors relative to their initial positions are γ1 and γ2, respectively. c This is the effective bending length of the hose. r This refers to the distance from the center of the output shaft to the part where the traction rope connects to the output shaft of the drive motor.

[0011] The 3D printing system for continuous fiber-reinforced composite materials with adjustable nozzle shape according to the present invention further includes a conductive forming platform, a conductivity measuring device, and a host computer. The nozzle head is a conductive nozzle head. The conductivity measuring device is connected to the nozzle head and the conductive forming platform respectively via wires, and is used to collect the conductivity data of the component in real time during the printing process. The host computer is connected to the conductivity measuring device and is used to perform real-time defect identification based on the conductivity data using a pre-trained defect identification model.

[0012] According to the 3D printing system for continuous fiber-reinforced composite materials with adjustable nozzle shape provided by the present invention, the parameter optimization module is integrated into the host computer, which is also connected to the adjustable nozzle structure. The system is used to optimize the printing parameters and nozzle shape combination at the beginning of printing, control the printing to be performed with the initially optimized printing parameters and nozzle shape combination, identify defects in the printing process in real time based on conductivity data, and optimize and update the printing parameters and nozzle shape combination again using the parameter optimization module when defects are identified.

[0013] According to another aspect of the present invention, a method for 3D printing continuous fiber reinforced composite materials with adjustable nozzle shape is provided, based on the 3D printing system for continuous fiber reinforced composite materials with adjustable nozzle shape as described in any one of the preceding claims, the method comprising: The pre-trained input-output response surface module optimizes the combination of printing parameters and nozzle shape. The input of the input-output response surface module is the combination of printing parameters and nozzle shape, and the output is the printing quality and / or printing efficiency index. The optimized combination of printing parameters and nozzle shape is obtained. Print using optimized printing parameters and nozzle configuration.

[0014] The 3D printing method for continuous fiber-reinforced composite materials with adjustable nozzle shape according to the present invention further includes: During the printing process, the electrical conductivity data between the nozzle head and the printing layer is acquired in real time, and the acquired electrical conductivity data is input into the pre-trained defect recognition model for real-time defect detection. If the number of defects reaches a preset threshold, the current printing parameters and nozzle shape combination will be input back into the training dataset of the input-output response surface model for retraining. Based on the retrained input-output response surface model, the parameters are optimized again to obtain updated printing parameters and nozzle shape combinations, and printing is performed using the updated printing parameters and nozzle shape combinations.

[0015] According to the 3D printing method for continuous fiber-reinforced composite materials with adjustable nozzle shape provided by the present invention, the training process of the defect recognition model is as follows: Pre-printing was performed with different combinations of printing parameters and nozzle shapes, and multiple samples were printed. The printing parameters and nozzle shape combinations were changed once for each preset number of layers printed in the same sample. The conductivity data obtained during the printing process was collected in real time, and the printing defect areas were marked after printing was completed, and the defect point labels corresponding to the conductivity data were obtained. The collected conductivity data sequence and the corresponding defect label are used as the dataset for the defect identification model. A historical conductivity data sequence in a sliding window is used as input and the identification result of whether it is a defect point is used as output to train the defect identification model, which is used to determine whether it is a defect point based on the conductivity data obtained in real time. The training process of the input-output response surface model is as follows: Multiple sets of data are randomly selected from printing data with different combinations of printing parameters and nozzle shapes. The printing parameters and nozzle shape combinations of each set, as well as the printing quality index and / or printing efficiency index of the corresponding sample, are used as the dataset for training the input-output response surface model. This model is used to train the input-output response surface model that reflects the mapping relationship between the printing parameters and nozzle shape combinations and the printing quality index and / or printing efficiency index.

[0016] In summary, compared with the prior art, the 3D printing system and method for continuous fiber-reinforced composite materials with adjustable nozzle shape provided by the present invention offer the following advantages: 1. A method is proposed that uses a flexible hose to connect the nozzle head and the heating chamber. The hose can bend and deform, allowing the nozzle head to move under the drive of the drive structure. This enables the adjustment and flexible setting of the nozzle shape, i.e., the discharge angle. The nozzle shape can be used as a parameter to be optimized for pre-optimization, and printing can be performed based on the optimized nozzle shape. The use of a movable nozzle to optimize the discharge angle helps to reduce the bending stress on the fiber during printing, and avoids local large-angle bending of the fiber as it flows from the nozzle to the sample, which could cause misalignment or breakage inside the fiber. This effectively reduces the probability of fiber breakage during printing, and makes up for the shortcomings of the fixed nozzle angle in traditional continuous fiber composite material 3D printing. This helps to reduce defects generated during printing and improve print quality. 2. Using dual motors to drive the nozzle hose to bend via traction ropes, the nozzle moves throughout the entire space through the combined superposition of displacement in the first and second directions, allowing the nozzle to bend in different directions. For more complex printing structures, the dual-motor driven movable nozzle eliminates the need for planning more complex printing paths, maintaining fiber continuity and exhibiting greater adaptability and better printing results. The dual-motor driven movable nozzle design effectively solves the fiber bending and breakage problem caused by a fixed nozzle. This design uses two independently controlled motors to adjust the position and angle of the nozzle, allowing it to more flexibly adapt to complex printing paths and shape changes. This not only reduces the bending stress on the fibers during printing but also improves the fiber alignment accuracy and consistency, thereby ensuring high efficiency and high quality in continuous carbon fiber reinforced composite 3D printing. 3. In-situ conductivity sensing technology replaces traditional offline inspection, enabling real-time, in-situ monitoring of print quality. This technology continuously monitors changes in the material's conductivity during the printing process, allowing for timely detection and correction of potential defects, ensuring the quality of each layer. Compared to traditional offline inspection methods, in-situ conductivity sensing not only improves inspection efficiency but also provides more accurate and continuous data feedback, helping to optimize printing parameters and nozzle morphology, thereby improving overall print quality and production efficiency. 4. Innovatively, the online defect monitoring process is combined with the offline parameter optimization process, realizing a closed-loop dynamic optimization process: information is collected by conductivity sensors, deep learning models are trained, the best printing parameters and nozzle shapes are predicted and optimized, defects are monitored layer by layer during the printing process, and secondary optimization of printing parameters and nozzle shapes is performed after defects are found. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall composition of the continuous fiber reinforced composite material 3D printing system with adjustable nozzle shape provided by the present invention.

[0018] Figure 2 This is a schematic diagram of the adjustable nozzle structure provided by the present invention.

[0019] Figure 3 This is a partially enlarged view of the nozzle head installation provided by the present invention.

[0020] Figure 4 This is a parameter annotation diagram of the nozzle shape provided by the present invention.

[0021] Figure 5 This is a parameter annotation diagram of the traction seat provided by the present invention.

[0022] Figure 6 This is a diagram illustrating the direction of nozzle movement provided by the present invention.

[0023] Figure 7 This is an overall schematic diagram of the continuous fiber reinforced composite material 3D printing method with adjustable nozzle shape provided by the present invention.

[0024] Figure 8 This is a graph showing the change in conductivity when the printing speed remains constant but the printing temperature changes, according to a specific embodiment of the present invention.

[0025] Figure 9 This is a graph showing the change in conductivity when the printing temperature remains constant but the printing speed varies in a specific embodiment of the present invention.

[0026] In all the accompanying drawings, the same reference numerals are used to denote the same elements or structures, wherein: 1-Adjustable nozzle structure; 2-Six-axis industrial robot; 3-Conductive forming platform; 4-Host computer; 5-Conductivity measuring device; 6-Reset spring; 7-Nozzle head; 8-Heating rod; 9-Shaft sleeve; 10-Drive motor; 11-Traction ear; 12-Guide wheel. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0028] Please see Figure 1 and Figure 2This embodiment provides a continuous fiber reinforced composite material 3D printing system with adjustable nozzle shape. The system includes an adjustable nozzle structure 1, which includes a nozzle head 7, a hose, and a drive structure. The nozzle head 7 is connected to and communicates with a heating chamber for heating the printing material through the hose. The drive structure is connected to the nozzle head 7 and is used to drive the nozzle head 7 to move to adjust the nozzle shape and achieve the adjustment of the material output angle. During printing, the nozzle shape obtained through pre-optimization is used for printing.

[0029] refer to Figure 3 In this embodiment, a flexible hose connects the nozzle head 7 and the heating chamber. The hose can bend and deform, allowing the nozzle head 7 to move under the drive of the driving structure. This enables the adjustment and flexible setting of the nozzle shape, i.e., the discharge angle. The nozzle shape can be used as a parameter to be optimized for pre-optimization, and printing can be performed based on the optimized nozzle shape. The use of a movable nozzle to optimize the discharge angle helps reduce the bending stress on the fiber during printing and avoids large-angle bending of the fiber as it flows from the nozzle to the sample, which could cause misalignment or breakage inside the fiber. This effectively reduces the probability of fiber breakage during printing, making up for the shortcomings of fixed nozzle angles in traditional continuous fiber composite material 3D printing. It also helps reduce defects generated during printing and improves printing quality.

[0030] In some embodiments, the drive structure includes two drive motors 10, the output shaft of one drive motor 10 is connected to one end of two traction ropes respectively, the other ends of the two traction ropes are connected to the nozzle head 7 and are symmetrically distributed along a first direction; the output shaft of the other drive motor 10 is connected to one end of two other traction ropes respectively, the other ends of the two other traction ropes are connected to the nozzle head 7 and are symmetrically distributed along a second direction; the first direction and the second direction are perpendicular to each other.

[0031] refer to Figure 1 and Figure 2 The 3D printing system may also include a six-axis industrial robot 2, which is the foundation of the entire system. The adjustable nozzle structure 1 is installed at the end of the six-axis industrial robot 2 and moves under its drive. The tangential direction vector of its nozzle end is adjusted by pulling through the dual motor drive cable, thereby changing the nozzle shape and adjusting the discharge angle.

[0032] refer to Figure 3 In some embodiments, the adjustable nozzle structure 1 further includes at least one of the following structures: A return spring 6 is fitted around the outside of the flexible tube. One end of the return spring 6 is connected to the heating chamber, and the other end is connected to the nozzle head 7. The nozzle head 7 is connected to the heating chamber through the flexible tube. Therefore, rotating the motor pulls the cable to change the nozzle shape, and the return spring 6 around the flexible tube can reset the nozzle. During the printing initialization phase, both motors are in a disabled state. At this time, the flexible tube and nozzle head 7, under the action of the return spring 6, pull the cable to return the motor to the mechanical origin, i.e., the initial position. The heating chamber is equipped with a heating rod 8 to provide a heating source. A return spring 6 is set around the output pipe of the output nozzle to help the movable nozzle return to the mechanical origin and complete the posture initialization.

[0033] A guide wheel 12 is fixedly provided near the nozzle head 7. The guide wheel 12 is arranged in a one-to-one correspondence with the traction rope and is used to pass through the traction rope.

[0034] A bushing 9 is integrally fitted onto the output shaft of the drive motor 10. A traction seat is provided on the bushing 9 for connecting a traction rope. Specifically, the traction seat may be T-shaped, with the longer end connected to the bushing 9, for example, it may be an integrally connected structure. The traction rope, i.e., the cable, is connected to the traction seat, and the two shorter ends can limit the cable's movement. Figure 5 As shown.

[0035] The nozzle head 7 is provided with a traction lug 11 for connecting a traction rope. The nozzle head 7 is driven by two motors, each motor pulling two cables via a traction mechanism. One end of each cable is connected to the drive rod of the motor's traction mechanism, and the other end is connected to the nozzle head 7. The connection points between the cables and the nozzle head 7 are evenly distributed on the circumference of the nozzle head 7, such as... Figure 6 As shown. Cables can be selected with high strength, high modulus, good flexibility, and small cross-sectional area, including but not limited to: aramid fiber, carbon fiber, and liquid crystal polymer fiber.

[0036] In some embodiments, the continuous fiber reinforced composite material 3D printing system with adjustable nozzle shape further includes a parameter optimization module, which optimizes the combination of printing parameters and nozzle shape based on a pre-trained input-output response surface model, wherein the input of the input-output response surface model is the combination of printing parameters and nozzle shape, and the output is the printing quality index and / or printing efficiency index. The printing parameters include multiple parameters such as layup angle, nozzle temperature, fiber filling density, layer thickness, and fiber printing speed; the nozzle shape includes nozzle bending angle and / or hose curvature radius, where the nozzle bending angle is the angle between the tangential direction of the hose near the nozzle end and the vertical direction.

[0037] refer to Figure 4In this embodiment, the nozzle shape is defined using the bending angle of the nozzle tip and the radius of curvature of the hose. To prevent the input-output response surface model from failing due to the relative position of the motor and nozzle head 7, and to improve model robustness, a conversion calculation formula is used to transform the rotation angles γ1 (rad) and γ2 (rad) of the dual motors into the nozzle bending angle α (rad) and the hose radius of curvature ρ (mm), and these two dimensions of features are input into the model for training. The bending angle of the nozzle tip and the radius of curvature of the hose can be obtained from the motor rotation angle according to the conversion calculation formula. The specific calculation formula for the nozzle shape is as follows: ; ; Wherein, the nozzle bending angle is α, the hose curvature radius is ρ, and the rotation angles of the two drive motors 10 relative to their initial positions are γ1 and γ2, respectively. c (mm) is the effective bending length of the hose (specifically, the length from the point where the hose connects to the nozzle head 7 of the traction rope to the end of the hose that connects to the heating chamber). r (mm) represents the distance from the point where the traction rope connects to the output shaft of the drive motor 10 to the center of the output shaft, specifically the distance between the end of the traction seat furthest from the output shaft and the center of the output shaft. The initial position, i.e., the mechanical origin, serves as the origin for the angle values ​​of the two motors rotating during the input depth model training and subsequent printing process. It can also be the initial installation position of the nozzle head 7 and the hose when the motor is not enabled.

[0038] refer to Figure 1 In some embodiments, the continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape further includes a conductive forming platform 3, a conductivity measuring device 5, and a host computer 4. The nozzle head 7 is a conductive nozzle head 7. The conductivity measuring device 5 is connected to the nozzle head 7 and the conductive forming platform 3 respectively via wires, and is used to collect the conductivity data of the component in real time during the printing process. The host computer 4 is connected to the conductivity measuring device 5 and is used to perform real-time defect identification based on the conductivity data using a pre-trained defect identification model.

[0039] The positive electrode of the conductivity measuring device 5 is installed on the nozzle head 7, for example, attached to the conductive nozzle head 7 or clamped between the hose and the nozzle head 7, and the negative electrode is connected to the grounding terminal of the conductive forming platform 3 for real-time acquisition of the conductivity information of the component during the printing process; a conductive layer can be provided on the conductive forming platform 3 and a connection interface can be provided to connect to the conductive layer, and the connection interface is used to connect the negative electrode of the conductivity measuring device 5.

[0040] In some embodiments, the parameter optimization module is integrated into the host computer 4, which is also connected to the six-axis industrial robot 2 and the adjustable nozzle structure 1. It is used to optimize the printing parameters and nozzle shape combination at the beginning of printing, control printing with the initially optimized printing parameters and nozzle shape combination, identify defects in the printing process in real time based on conductivity data, and optimize and update the printing parameters and nozzle shape combination again when defects are identified.

[0041] Specifically, each conductivity data point corresponds to a defect point identification result. When the number of identified defect points reaches a preset threshold, the parameter optimization module is used to optimize and update the printing parameters and nozzle shape combination again. Specifically, the printing data corresponding to the current printing parameters and nozzle shape combination is input back into the dataset of the input-output response surface model for retraining, and the parameters are optimized again based on the retrained input-output response surface model to obtain the updated printing parameters and nozzle shape combination.

[0042] This system dynamically adjusts the end-point direction vector of the print nozzle head 7 using dual motors, effectively reducing fiber breakage and improving print quality. It also establishes an integrated prediction-optimization-monitoring framework, enabling real-time, layer-by-layer verification of the optimal parameter combination. Through monitoring feedback, it achieves iterative optimization of print parameters. Specifically, it extracts features from conductivity data collected by conductivity measurement device 5, integrates print parameters and nozzle shape, and inputs them into a deep learning-driven forward prediction-dynamic optimization-defect identification model to obtain the optimal nozzle shape and other print parameters. The optimized printing process is then monitored online to achieve closed-loop dynamic optimization.

[0043] First, a defect identification model based on conductivity data to predict printing defects is constructed using neural networks such as RNNs, LSTMs, and Elman's algorithm. Second, an input-output response surface model based on MLPs is constructed to predict the relationship between print quality and print efficiency based on combinations of print parameters and nozzle shapes. This model serves as a predictor, generating multiple sets of print parameter and nozzle shape combinations within a reasonable range, along with their corresponding print quality and print efficiency. Finally, these combinations are input into a multi-objective optimization algorithm to obtain the optimal combination of print parameters and nozzle shapes. The multi-objective optimization algorithm uses the mapping relationship between the print parameter and nozzle shape combinations derived from the above input-output response surface model and print quality and / or print efficiency as the objective function to derive the optimal combination of print parameters and nozzle shapes. Multi-objective optimization algorithms include, but are not limited to, NSGA-II, MOGWO, MOPSO, SPEA2, and NSGA-III.

[0044] The continuous fiber reinforced composite material can be selected as a continuous conductive fiber reinforced composite material, including but not limited to continuous carbon fiber reinforced thermoplastic resin matrix composite material, continuous carbon fiber reinforced thermosetting resin matrix composite material, continuous graphite fiber reinforced resin matrix composite material and continuous conductive fiber hybrid reinforced composite material.

[0045] Furthermore, this embodiment also provides a 3D printing method for continuous fiber reinforced composite materials with adjustable nozzle shape, based on the 3D printing system for continuous fiber reinforced composite materials with adjustable nozzle shape described in any of the above embodiments, the method comprising: The pre-trained input-output response surface module optimizes the combination of printing parameters and nozzle shape. The input of the input-output response surface module is the combination of printing parameters and nozzle shape, and the output is the printing quality and / or printing efficiency index. The optimized combination of printing parameters and nozzle shape is obtained. Print using optimized printing parameters and nozzle configuration.

[0046] In some embodiments, the continuous fiber-reinforced composite material 3D printing method with adjustable nozzle shape further includes: During the printing process, the electrical conductivity data between the nozzle head 7 and the printing layer is acquired in real time, and the acquired electrical conductivity data is input into the pre-trained defect recognition model for real-time defect detection. If the number of defects does not reach the preset threshold, printing continues. If the number of defects reaches the preset threshold, the printing data corresponding to the current printing parameters and nozzle shape combination is input back into the training dataset of the input-output response surface model for retraining. Based on the retrained input-output response surface model, the parameters are optimized again to obtain the updated printing parameters and nozzle shape combination. Printing is then performed using the updated printing parameters and nozzle shape combination, thereby achieving closed-loop dynamic optimization of the process parameters for continuous fiber reinforced composite material 3D printing.

[0047] In this embodiment, a dataset of printing parameters and nozzle shape combinations, along with corresponding printing quality and / or printing efficiency, is first constructed and divided into training and testing sets to train an input-output response surface model. Then, the mapping relationship between printing process parameters and printing quality and / or printing efficiency derived from the input-output response surface model is used as the objective function of a multi-objective optimization algorithm to obtain the optimal solution. Finally, printing is performed according to the optimal printing parameters and nozzle shape. During the printing process, defects are identified in real time for feasibility analysis of the optimal solution. If a printing defect is detected, the current printing parameters and nozzle shape combination are input back into the dataset of the input-output response surface model, and the dataset is retrained to achieve real-time iterative optimization.

[0048] This embodiment innovatively achieves real-time quality monitoring through in-situ conductivity sensing, enabling real-time monitoring of print quality. This allows for in-situ inspection of continuous fiber reinforced composite 3D printed components, freeing them from reliance on large specialized equipment and cumbersome inspection methods. Secondly, the innovative movable nozzle design allows for flexible adjustment of the filament exit angle, significantly reducing fiber bending and breakage, and improving interlayer bonding strength, effectively solving the fiber bending and breakage problem caused by fixed nozzles during printing. Simultaneously, it combines deep learning algorithms to dynamically optimize process parameters, linking conductivity measurements with various printing parameters and nozzle morphology. This allows for real-time analysis and adjustment of printing parameters and nozzle morphology during printing, reducing optimization time and material costs and solving the problem of low efficiency in process optimization caused by traditional trial-and-error methods. The closed-loop control system organically combines monitoring, prediction, and optimization, achieving adaptive adjustment of printing parameters and nozzle morphology. While ensuring print quality, it significantly improves forming efficiency, fundamentally changing the traditional post-printing process's reactive defect handling mode of detection and passive adjustment. It achieves a technological leap from "problem detection - shutdown for repair" to "prediction - real-time adjustment," solving the core problems of lagging detection and passive process control in traditional methods.

[0049] In some embodiments, the training process of the defect identification model is as follows: Pre-printing is performed with different combinations of printing parameters and nozzle shapes. Multiple samples are printed. In the same sample, the printing parameters and nozzle shape combination are changed once for each preset number of layers (e.g., 5 layers). The conductivity data obtained during the printing process is collected in real time. After printing, the printing defective parts are marked and the defect point labels corresponding to the conductivity data are obtained. Specifically, after printing, the conductivity values ​​corresponding to the printing defects are annotated on the visualization model using a printing path visualization program. The visualization program associates and displays the conductivity values ​​of various locations on the printed sample based on G-code. The program's marker brush can be used to annotate the conductivity at the location of the defect in the sample. This allows for the annotation of defective areas on the printed sample after a single print run, determining the location and size of the defects. The defects are then mapped onto the model, and the printing path visualization program determines the corresponding conductivity data. The label corresponding to this conductivity data is used to identify the defect point.

[0050] The collected conductivity data sequence and corresponding defect label are used as the dataset for the defect identification model to train the defect identification model based on RNN. A historical conductivity data sequence in a sliding window is used as input and the identification result of whether it is a defect point is used as output to train the defect identification model, which is used to determine whether it is a defect point based on the conductivity data obtained in real time.

[0051] The training process of the input-output response surface model is as follows: Multiple sets of data (e.g., 50 sets of printing data) are randomly selected from printing data with different combinations of printing parameters and nozzle shapes. The printing parameters and nozzle shape combinations of each set, as well as the printing quality index and / or printing efficiency index of the corresponding sample, are used as the dataset for training the input-output response surface model (e.g., a model built based on MLP). This dataset is used to train the input-output response surface model that reflects the mapping relationship between the printing parameters and nozzle shape combinations and the printing quality index and / or printing efficiency index.

[0052] Meanwhile, different combinations of printing parameters and nozzle shapes are used as the parameter space of the optimization algorithm. Using the input-output response surface model obtained in the above steps, a large number of input combinations of printing parameters and nozzle shapes are randomly generated within the parameter space to predict the corresponding output combinations of printing quality and printing efficiency. Then, the output combinations are input into the multi-objective optimization algorithm to obtain the optimal printing parameters and nozzle shapes.

[0053] The optimization method provided in this embodiment achieves a closed-loop dynamic optimization process through the above steps: multi-sensor information acquisition—predicting the mapping relationship between printing parameters, nozzle shape, printing quality, and printing efficiency using an input-output response surface model based on MLP—predicting and optimizing the optimal printing parameters and nozzle shape using the NSGA-II multi-objective optimization algorithm—monitoring defects layer by layer during the printing process using an RNN model—and further optimizing the printing parameters and nozzle shape based on defect detection. Figure 7 As shown.

[0054] In this embodiment, the printing quality index reflects the presence or absence of printing defects, while the printing efficiency index (e.g., printing speed in mm / s) reflects the printing speed. The printing quality index aims to reflect printing quality and can be, for example, a mechanical performance index, without specific limitations. The printing quality index can also be a label indicating whether the printed sample is defective. When constructing the dataset, each set of printing data corresponds to one printed sample. Whether the printed sample is defective can be determined based on the number of defect points identified by the defect identification model. A first threshold can be set; if the number of defect points reaches this first threshold, a defect is determined to exist; otherwise, it is considered defect-free. During the actual printing process, a second threshold can be set. When the number of identified defect points reaches the second threshold, the current printing parameters and nozzle shape, combined with a defective printed label, need to be input back into the training dataset of the input-output response surface model for retraining and parameter optimization updates. The first and second thresholds can be equal or unequal, and can be flexibly set according to the actual situation without specific limitations.

[0055] Furthermore, in this specific embodiment, the vertically downward direction of the nozzle head 7 is taken as the positive direction of the mechanical origin, and the angle between the tangential direction of the nozzle end of the printing hose and the positive direction of the mechanical origin is the nozzle bending angle α (rad). The radius of curvature of the hose is ρ (mm), the layup angle is β (rad), the nozzle temperature is T (°C), the fiber filling density is D (%), the layer thickness is L (mm), and the fiber printing speed is... V (mm / s), seven printing parameters and print nozzle shape were used as optimization targets.

[0056] To improve the robustness of the model, the parameter space range and the deep learning dataset are both trained using the nozzle bending angle α and the hose curvature radius ρ. The nozzle bending angle α and the hose curvature radius ρ can be calculated from the angles γ1 (rad) and γ2 (rad) rotated by the two motors respectively, the radius of the traction mechanism drive rod is r (mm), and the effective bending length c (mm) of the nozzle and hose.

[0057] Cable length change ΔL: ; like Figure 6 As shown, the nozzle head 7 moves in the x and y planes respectively. The geometric parameters of the movable nozzle can be derived as follows: ; ; Since α is small, the above equation can be simplified to: ; ; Since the length of the hose remains constant, we can conclude that: ; ; Substituting the formula into the equation simplifies it to: ; ; Since α is small, Taylor expansion is used here, that is: ; Ultimately, we can obtain: ; ; ; .

[0058] Furthermore, the optimization algorithm parameter space given in this specific embodiment is as follows: ; In this specific embodiment, 10 samples are printed. For each sample, the printing parameters and nozzle shape are changed every 5 layers, forming 50 combinations of printing parameters and nozzle shapes.

[0059] First, the resistance values ​​and conductivity data read in real time from 50 groups are calibrated. The resistance values ​​are used as input values, and the calibration status of the resistance values, i.e., whether there are defects or not, is used as the output result to construct a dataset for training the defect recognition model.

[0060] Furthermore, in this specific embodiment, a defect identification model is built based on an RNN recurrent neural network.

[0061] Furthermore, the Recurrent Neural Network (RNN) is a recursive structure with a context layer. It captures temporal dynamics by remembering historical states, making it suitable for sequence modeling of conductivity data. It takes a window-length segment of conductivity data as input, and the RNN network uses this historical sequence to determine whether a defect has occurred. The RNN network consists of an input layer, hidden layers, a context layer, and an output layer. The context layer stores the previous hidden layer state, implementing short-term memory. This specific embodiment uses an RNN network for forward prediction, converting conductivity feature data into a temporal input vector. The current state is calculated using a weight matrix and an activation function. The prediction result is generated by the output layer, whose hidden layer state function is: ; in x t for t Input feature vector (dimension 1×30) at each time step. W ih This is the weight matrix from the input layer to the hidden layer. W hc This is the weight matrix from the context layer to the hidden layer. b h For the hidden layer bias, f The activation function is ReLU. The output layer function is: ; in W ho The weight matrix from the hidden layer to the output layer. b o For output layer bias, g Let be a linear activation function. For the training process, the total error function is defined as: ; in d (i) For the desired output vector,n The number of samples is given. The RNN network updates its weights using the Backpropagation-Time (BPTT) algorithm, and its gradient calculation formula is: ; in W weight matrix W ih , W hc , W ho , T The sequence length is given.

[0062] Furthermore, 40 sets of the 50 sets of conductivity time-series data were randomly selected as training samples to complete the training of the RNN network, and the remaining 10 sets were used as the test set to evaluate the model performance. The root mean square error (RMSE) and mean absolute error (MAE) were used to quantitatively characterize the prediction accuracy.

[0063] Furthermore, this specific embodiment predicts the corresponding print quality and print efficiency for 50 combinations of print parameters and print nozzle shapes.

[0064] Furthermore, this specific embodiment selects a Multilayer Perceptron (MLP) neural network as the basis to construct an input-output response surface model, which is used to map the input combination of printing parameters and print nozzle shape to the output combination of whether the print is defective and the printing time. MLP is a feedforward neural network that approximates complex functional relationships through multi-layer nonlinear transformations, and is suitable for static regression problems with multiple inputs and multiple outputs. Its network consists of an input layer, several hidden layers, and an output layer, with fully connected neurons in each layer. Predicted values ​​are calculated through forward propagation, and weights are updated through backpropagation.

[0065] In this specific embodiment, the input nodes of the MLP network correspond to seven dimensions of printing parameters and nozzle morphology: layup angle, nozzle temperature, fiber filler density, layer thickness, fiber printing speed, nozzle bending angle, and hose curvature radius. These are normalized and used as the input vector X. The hidden layer employs a two-layer structure with nodes h1 and h2 respectively, and ReLU is selected as the activation function to mitigate gradient vanishing. The output nodes are two, corresponding to tensile strength and printing time respectively. The forward propagation calculation is as follows: ; ; ; in W （l） , b （l） These are the weight matrix and bias vector of the l-th layer, respectively. fThe hidden layer activation function is ReLU, and the output layer has linear activation. For the portion of the output used for printing efficiency during regression, the total loss function during training is defined as mean squared error (MSE). To account for the difference in dimensions between printing quality and printing efficiency, tensile strength and printing time are standardized separately. The loss function for printing efficiency is as follows: ; For the print quality judgment step, which belongs to binary classification, it is first converted into a probability using the Sigmoid function: ; Where P is the probability of printing a defect, and σ() is the sigmoid activation function. Then, a binary cross-entropy loss function is constructed based on this probability: ; in y true The true label is 0 or 1, where 0 indicates no defects and 1 indicates defects.

[0066] Similarly, 40 sets of the 50 sets of printing parameters and print nozzle shape data were randomly selected as training samples to complete the input-output response surface model training, and the remaining 10 sets were used as the test set to evaluate the model performance.

[0067] Furthermore, the NSGA-II algorithm solves multi-objective optimization problems through fast non-dominated sorting and an elitist strategy, ensuring a uniform distribution of the Pareto solution set. This specific embodiment is based on NSGA-II dynamic optimization process parameters: In non-dominated sorting, for individual A to dominate individual B, the following must be satisfied: ; in The objective function value is denoted as . The crowding distance is calculated as follows: ; Used to maintain population diversity.

[0068] Furthermore, the optimization steps for the input-output response surface model based on NSGA-II are as follows: First, establish the objective function: ; Among the optimization variables u =[β,T,D,L,V,α,ρ]. After the NSGA-II generates the initial population, the input-output response surface model predicts ƒ. k and ƒ ξ Elite individuals are selected through non-dominated sorting and crowding, and offspring are generated through simulated binary crossover (crossover probability 0.9) and polynomial mutation (mutation probability 0.1); the process iterates until the Pareto front converges, and the optimal parameter u is output. This embodiment is implemented in PyCharm 2023. The RNN network is built using the PyTorch framework, and the multi-objective optimization part uses the NSGA-II algorithm implemented in the pymoo library.

[0069] In this specific example, a continuous fiber 3D printer was used to print continuous carbon fiber reinforced thermosetting and thermoplastic composite materials at different printing temperatures and speeds, and the electrical conductivity of the parts was measured. Figure 8 The figure shows that when the printing speed remains constant but the printing temperature changes, the conductivity value is approximately 459 S / m when the printing temperature is 190℃ and the printing speed is determined by the printing speed coefficient Vα(1) on the printer used; the conductivity value is approximately 384 S / m when the printing temperature is 210℃ and the printing speed coefficient Vα(1); and the conductivity value is approximately 312 S / m when the printing temperature is 230℃ and the printing speed coefficient Vα(1). Figure 9 The figure shows that when the printing temperature remains constant but the printing speed varies, the conductivity value is approximately 384 S / m at a printing temperature of 210℃ and a printing speed coefficient Vα(1); approximately 575 S / m at a printing temperature of 210℃ and a printing speed coefficient Vα(2); and 972 S / m at a printing temperature of 210℃ and a printing speed coefficient Vα(3). This result verifies the accuracy and feasibility of the conductivity measurement.

[0070] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape, characterized in that, The device includes an adjustable nozzle structure, which comprises a nozzle head, a hose, and a drive structure. The nozzle head is connected to and communicates with a heating chamber for heating the printing material via the hose. The drive structure is connected to the nozzle head and drives the nozzle head to move to adjust the nozzle shape and thus adjust the discharge angle. During printing, the nozzle shape is pre-optimized.

2. The continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape as described in claim 1, characterized in that, The drive structure includes two drive motors. The output shaft of one drive motor is connected to one end of two traction ropes, and the other ends of the two traction ropes are connected to the nozzle head and symmetrically distributed along a first direction. The output shaft of the other drive motor is connected to one end of two other traction ropes, and the other ends of the two other traction ropes are connected to the nozzle head and symmetrically distributed along a second direction. The first direction and the second direction are perpendicular to each other.

3. The continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape as described in claim 2, characterized in that, The adjustable nozzle structure also includes at least one of the following structures: A return spring is fitted around the outside of the hose. One end of the return spring is connected to the heating chamber, and the other end is connected to the nozzle head. A guide wheel is fixedly provided near the nozzle head, and the guide wheel is provided in a one-to-one correspondence with the traction rope for passing through the traction rope; The output shaft of the drive motor is integrally fitted with a bushing, and the bushing is provided with a traction seat for connecting the traction rope; The nozzle head is provided with a traction lug for connecting a traction rope.

4. The continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape as described in claim 2, characterized in that, It also includes a parameter optimization module, which optimizes the combination of printing parameters and nozzle shape based on a pre-trained input-output response surface model. The input of the input-output response surface model is the combination of printing parameters and nozzle shape, and the output is the printing quality index and / or printing efficiency index. The printing parameters include multiple parameters such as layup angle, nozzle temperature, fiber filling density, layer thickness, and fiber printing speed; the nozzle shape includes nozzle bending angle and / or hose curvature radius, where the nozzle bending angle is the angle between the tangential direction of the hose near the nozzle end and the vertical direction.

5. The 3D printing system for continuous fiber-reinforced composite materials with adjustable nozzle shape as described in claim 4, characterized in that, The specific calculation formula for nozzle shape is as follows: ； ； Wherein, the nozzle bending angle is α, the hose curvature radius is ρ, and the rotation angles of the two drive motors relative to their initial positions are γ1 and γ2, respectively. c This is the effective bending length of the hose. r This refers to the distance from the center of the output shaft to the part where the traction rope connects to the output shaft of the drive motor.

6. The 3D printing system for continuous fiber-reinforced composite materials with adjustable nozzle shape as described in claim 4, characterized in that, It also includes a conductive forming platform, a conductivity measuring device, and a host computer. The nozzle head is a conductive nozzle head. The conductivity measuring device is connected to the nozzle head and the conductive forming platform respectively through wires, and is used to collect the conductivity data of the component in real time during the printing process. The host computer is connected to the conductivity measuring device and is used to perform real-time defect identification based on conductivity data using a pre-trained defect identification model.

7. The continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape as described in claim 6, characterized in that, The parameter optimization module is integrated into the host computer, which is also connected to the adjustable nozzle structure. It is used to optimize the printing parameters and nozzle shape combination at the beginning of printing, control the printing to be performed with the initially optimized printing parameters and nozzle shape combination, identify defects in the printing process in real time based on conductivity data, and optimize and update the printing parameters and nozzle shape combination again when defects are identified.

8. A method for 3D printing continuous fiber-reinforced composite materials with adjustable nozzle shape, characterized in that, The continuous fiber-reinforced composite material 3D printing system with adjustable nozzle shape according to any one of claims 1-7, the method comprising: The pre-trained input-output response surface module optimizes the combination of printing parameters and nozzle shape. The input of the input-output response surface module is the combination of printing parameters and nozzle shape, and the output is the printing quality and / or printing efficiency index. The optimized combination of printing parameters and nozzle shape is obtained. Print using optimized printing parameters and nozzle configuration.

9. The 3D printing method for continuous fiber-reinforced composite materials with adjustable nozzle shape as described in claim 8, characterized in that, Also includes: During the printing process, the electrical conductivity data between the nozzle head and the printing layer is acquired in real time, and the acquired electrical conductivity data is input into the pre-trained defect recognition model for real-time defect detection. If the number of defects reaches a preset threshold, the current printing parameters and nozzle shape combination will be input back into the training dataset of the input-output response surface model for retraining. Based on the retrained input-output response surface model, the parameters are optimized again to obtain updated printing parameters and nozzle shape combinations, and printing is performed using the updated printing parameters and nozzle shape combinations.

10. The 3D printing method for continuous fiber-reinforced composite materials with adjustable nozzle shape as described in claim 9, characterized in that, The training process of the defect identification model is as follows: Pre-printing was performed with different combinations of printing parameters and nozzle shapes, and multiple samples were printed. The printing parameters and nozzle shape combinations were changed once for each preset number of layers printed in the same sample. The conductivity data obtained during the printing process was collected in real time, and the printing defect areas were marked after printing was completed, and the defect point labels corresponding to the conductivity data were obtained. The collected conductivity data sequence and the corresponding defect label are used as the dataset for the defect identification model. A historical conductivity data sequence in a sliding window is used as input and the identification result of whether it is a defect point is used as output to train the defect identification model, which is used to determine whether it is a defect point based on the conductivity data obtained in real time. The training process of the input-output response surface model is as follows: Multiple sets of data are randomly selected from printing data with different combinations of printing parameters and nozzle shapes. The printing parameters and nozzle shape combinations of each set, as well as the printing quality index and / or printing efficiency index of the corresponding sample, are used as the dataset for training the input-output response surface model. This model is used to train the input-output response surface model that reflects the mapping relationship between the printing parameters and nozzle shape combinations and the printing quality index and / or printing efficiency index.

Images