A moldless mirror image filament laying method and a moldless mirror image filament laying system

Abstract

The present application belongs to the technical field of material processing, and relates to a mold-free mirror image filament laying method and a mold-free mirror image filament laying system. The method lays a filament ribbon in a mirror image symmetry, without a mold, thereby improving production flexibility and efficiency. The method acquires the temperature and pressure of the filament ribbon in real time, calculates the output power of a heater and the modification value of a filament laying end motion mechanism path by using temperature and pressure differences, and optimizes controller parameters through reinforcement learning, so as to realize stable control of the temperature and pressure. The system comprises a hardware part and a control system. The hardware part includes pressure and heat source components. The control system adjusts through real-time feedback to ensure the quality of filament laying. The present application solves the problem that the temperature and pressure are difficult to accurately control in the process of mold-free mirror image filament laying, and improves the performance of composite products.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of material processing, and relates to a mold-free mirror image fiber laying method and a mold-free mirror image fiber laying system. BACKGROUND

[0002] In the field of material processing, fiber laying technology, as a key process for composite material manufacturing, is widely used in the production of complex components such as aircraft wings. Traditional fiber laying technology relies on pre-manufactured metal molds, which need to be customized according to the curved surface shape of the component. Not only is it time-consuming and material-consuming, but each different component also requires an independent mold, greatly limiting the flexibility and efficiency of production. In addition, the design and manufacturing process of the mold is complex and requires high technical requirements, further increasing the technical threshold and cost investment.

[0003] In the field of material processing, mirror image processing technology generally refers to using two or more sets of identical or symmetrical processing equipment to simultaneously process the two sides of a workpiece to be processed, thereby achieving accurate and symmetrical processing of the workpiece. In theory, if mirror image processing technology can be combined with fiber laying technology, two sets of identical fiber laying equipment can be used, one on each side of the workpiece to be processed, and the two sets of equipment are in mirror image symmetrical positions. The two sets of equipment can provide necessary support to each other during the fiber laying process without the need for molds, achieving mold-free fiber laying. However, this idea is extremely difficult to implement because:

[0004] 1. In the fiber laying process, the curing heat is provided by the heaters on both sides, resulting in a complex and unstable heat transfer path. As the number of fiber layers increases and the structure of the intermediate layer changes, the heat transfer effect of the side heat becomes unpredictable, causing temperature fluctuations. This temperature fluctuation directly affects the curing quality of the thermoplastic resin, which may cause material performance degradation, brittle cracks, and even burning and other serious problems;

[0005] 2. If two driven compression rollers are used to generate compaction pressure by pressing against each other, due to the power source, small contact area, and high curvature characteristics of the compression rollers, the pressure is prone to fluctuate dramatically. Unstable pressure not only affects the filling effect of the resin on the fiber gap, which may form bubbles or voids and reduce the mechanical properties of the composite material, but also may cause excessive extrusion of the resin, resulting in uneven interlayer adhesion between the fibers and weakening the interlaminar peel strength of the composite material;

[0006] 3. In the fiber laying process, mechanical arms, gantries, and other motion mechanisms are usually used. The motion accuracy and rigidity of these motion mechanisms are relatively low, much lower than that of machine tools, which increases the difficulty of pressure and temperature control during the fiber laying process.

[0007] Therefore, to combine mirror image processing technology with fiber laying technology well, the problem that needs to be solved is how to accurately control the temperature and pressure during the fiber laying process.

[0008] Currently, there are some technologies for controlling temperature and pressure. For example, patent application US202117401138A discloses an in-situ monitoring method for a compaction roller in a composite material filament winding process, which uses an infrared thermal imager to collect temperature data for online monitoring and feedback control. However, this patent application is for traditional mold filament winding or flat filament winding scenarios, and the compaction roller temperature profile is analyzed to infer the composite material defect situation. This method is easily disturbed by various factors and is not suitable for moldless filament winding.

[0009] In addition, existing pressure control technologies are usually based on mold compaction force prediction (for example, documents "Modeling and experimental validation of compaction pressure distribution for automated fiber placement, Composite Structures, Volume 256, 2021, 113101, ISSN 0263-8223" and "Pressure distribution for automated fiber placement and design optimization of compaction rollers. Journal of Reinforced Plastics and Composites. 2019; 38(18): 860-870"). These technologies are also not suitable for moldless filament winding. SUMMARY

[0010] The purpose of the present application is to solve the problems existing in the prior art and provide a moldless mirror image filament winding method and a moldless mirror image filament winding system.

[0011] To achieve the above-mentioned purpose, the technical solutions adopted by the present application are as follows:

[0012] A moldless mirror image filament winding method does not use a mold, and two belts are laid by mirror symmetry, and the filament winding process is repeated multiple times.

[0013] During the filament winding process, the real belt temperature is obtained in real time, and the temperature difference is calculated in real time. The modification value of the heater output power is calculated in real time according to the temperature difference, and the heater output power is updated in real time.

[0014] The temperature difference is the difference between the real belt temperature and the target belt temperature.

[0015] The modification value of the heater output power is kp*×T + ki*×ΣT + kd*×ΔT.

[0016] In the formula, T is the temperature difference at the current time of the current filament laying process, ΣT is the sum of all temperature differences up to the current time of the current filament laying process, ΔT = temperature difference at the current time of the current filament laying process - temperature difference at the last time of the current filament laying process; kp*, ki*, kd* are the required controller parameter combinations found by continuously iterating and optimizing in the virtual environment through simulation using the reinforcement learning model;

[0017] Real-time acquisition of real tape pressure and real-time calculation of pressure difference in the filament laying process, real-time calculation of the modification value of the filament laying end motion mechanism (robotic arm or other motion mechanism, such as gantry or any single-axis or multi-axis motion execution mechanism) path according to the pressure difference, and real-time updating of the filament laying end motion mechanism path;

[0018] Pressure difference = difference between real tape pressure and target tape pressure;

[0019] Modification value of filament laying end motion mechanism path = kp' x P + ki' x ΣP + kd' x ΔP;

[0020] In the formula, P is the pressure difference at the current time of the current filament laying process, ΣP is the sum of all pressure differences up to the current time of the current filament laying process, ΔP = pressure difference at the current time of the current filament laying process - pressure difference at the last time of the current filament laying process; kp', ki', kd' are the required controller parameter combinations found by continuously iterating and optimizing in the virtual environment through simulation using the reinforcement learning model.

[0021] As a preferred technical solution:

[0022] The mirror image filament laying method as described above, kp*, ki*, kd* are obtained by the following process:

[0023] (a) generate a set of kp, ki, kd;

[0024] (b) let i = 1;

[0025] (c) perform the same simulation as the actual process in the virtual space, start the i-th filament laying process;

[0026] (d) let j = 1, let the virtual heater output power = the heater output power in the actual process;

[0027] (e) input the virtual heater output power and the corresponding heating time into the prediction model A, output the predicted tape temperature from it, calculate the difference between the predicted tape temperature and the target tape temperature, and get the j-th temperature difference of the i-th temperature difference sequence;

[0028] Prediction model A is a trained deep learning model. During training, the heater output power and the corresponding heating time are used as the input of the deep learning model, and the ribbon temperature is used as the theoretical output of the deep learning model. The parameters of the deep learning model are continuously adjusted.

[0029] (f) Determine whether the i-th fiber placement process is completed. If not, calculate the modified value of the virtual heater output power using the following formula, update the virtual heater output power, set j = j + 1, and return to step (e); otherwise, proceed to the next step;

[0030] Modified value of virtual heater output power = kp×T+ki×ΣT+kd×ΔT;

[0031] Where T is the jth temperature difference in the i-th temperature difference sequence, ΣT is the sum of all temperature differences in the i-th temperature difference sequence, ΔT = the jth temperature difference in the i-th temperature difference sequence - the j-1th temperature difference in the i-th temperature difference sequence. When j = 1, let ΔT = 0;

[0032] (g) Score the i-th temperature difference sequence and determine whether the score of the i-th temperature difference sequence reaches the set value. If so, output the last set of kp, ki, and kd as kp*, ki*, and kd*; otherwise, proceed to the next step;

[0033] (h) Determine whether all the laying processes are completed. If not, input the last set of kp, ki, kd and the i-th temperature difference sequence into the reinforcement learning model with a reward function, which outputs a new set of kp, ki, kd, and sets i=i+1, and then returns to step (c); otherwise, report an error, inform the technician that no suitable parameters have been found, and return to step (a).

[0034] For the aforementioned patternless mirror-image fiber placement method, the process for obtaining the prediction model A is as follows:

[0035] (i) Establishing a data set;

[0036] Collect the heater output power, corresponding heating time and actual ribbon temperature in the same historical process as the current process;

[0037] At the same time, the same simulation as the actual process is performed in the virtual space to obtain the virtual heater output power and the corresponding heating time and virtual ribbon temperature;

[0038] (ii) Data augmentation: increasing the size of the dataset to obtain a training set;

[0039] (iii) Establish and train a deep learning model to obtain prediction model A.

[0040] A moldless mirror image filament laying method as described above, kp', ki', kd' are obtained through the following process:

[0041] (A) Generate a set of kp, ki, kd;

[0042] (B) Let i = 1;

[0043] (C) Perform the same simulation as the actual process in the virtual space, start the i th laying process;

[0044] (D) Let j = 1, let the virtual filament laying end motion mechanism path = the filament laying end motion mechanism path in the actual process;

[0045] (E) Input the virtual filament laying end motion mechanism path into the prediction model B, output the predicted tape pressure from it, calculate the difference between the predicted tape pressure and the target tape pressure, and get the j th pressure difference of the i th pressure difference sequence;

[0046] The prediction model B is a trained deep learning model, and when training, the filament laying end motion mechanism path is taken as the input of the deep learning model, and the tape pressure is taken as the theoretical output of the deep learning model. The parameters of the deep learning model are adjusted constantly;

[0047] (F) Determine whether the i th laying process is completed, if not, calculate the modification value of the virtual filament laying end motion mechanism path using the following formula, update the virtual filament laying end motion mechanism path, and then let j = j + 1, return to step (E); otherwise, go to the next step;

[0048] The modification value of the virtual filament laying end motion mechanism path = kp × P + ki × ΣP + kd × ΔP;

[0049] In the formula, P is the j th pressure difference of the i th pressure difference sequence, ΣP is the sum of all pressure differences of the i th pressure difference sequence, and ΔP = the j th pressure difference of the i th pressure difference sequence - the j-1 th pressure difference of the i th pressure difference sequence. When j = 1, let ΔP = 0;

[0050] (G) Score the i th pressure difference sequence, determine whether the score of the i th pressure difference sequence reaches the set value, if yes, output the last set of kp, ki, kd, and take it as kp', ki', kd'; otherwise, go to the next step;

[0051] (H) Determine whether all laying processes are completed, if not, input the last set of kp, ki, kd and the i th pressure difference sequence into the reinforcement learning model with a reward function, output a new set of kp, ki, kd from it, let i = i + 1, and return to step (C); otherwise, report an error and inform the technician that no suitable parameters are found, and return to step (A).

[0052] The acquisition process of the prediction model B is as follows:

[0053] (I) Establish a data set;

[0054] Collect the tape laying end effector path and the real tape pressure in the historical process same as the current process;

[0055] At the same time, the same simulation as the actual process is performed in the virtual space to obtain the virtual tape laying end effector path and the virtual tape pressure;

[0056] (II) Data enhancement, improve the size of the data set, and obtain a training set;

[0057] (III) Establish and train a deep learning model, that is, obtain the prediction model B.

[0058] The application also provides a moldless mirror image tape laying system, which comprises a hardware part and a control system;

[0059] The hardware part comprises symmetrical left and right parts, and the right part comprises a main plate, a pressure component, a cutting component, a re-feeding component, a tension component, a heat source component, a connecting component and a tape laying end effector;

[0060] The pressure component comprises a pressure sensor;

[0061] The heat source component comprises a heater and an infrared thermal imager, the heater is used for heating the tape, and the infrared thermal imager is used for real-time shooting of the tape laying area to obtain temperature field data of the tape;

[0062] The control system comprises a temperature stability control subsystem, a pressure stability control subsystem, a control instruction generation subsystem, a tape laying end effector control subsystem and a tape laying end function mechanism control subsystem;

[0063] The infrared thermal imager is used for real-time acquisition of the real tape temperature during the tape laying process, and the real tape temperature is sent to the temperature stability control subsystem in real time;

[0064] The temperature stability control subsystem is used for real-time calculation of a temperature difference, real-time calculation of a modification value of the heater output power according to the temperature difference, and real-time sending of the modification value of the heater output power to the control instruction generation subsystem;

[0065] The temperature difference is the difference between the real tape temperature and the target tape temperature;

[0066] The modification value of the heater output power is kp*×T+ki*×ΣT+kd*×ΔT;

[0067] In the formula, T is the temperature difference at the current time of the current fiber laying process, ΣT is the sum of all temperature differences up to the current time of the current fiber laying process, ΔT = temperature difference at the current time of the current fiber laying process - temperature difference at the last time of the current fiber laying process; kp*, ki*, and kd* are the required controller parameter combinations found by continuously iterating and optimizing the reinforcement learning model through simulation in a virtual environment;

[0068] The pressure sensor is used to obtain the real tape pressure in real time during the fiber laying process and send the real tape pressure to the pressure stabilization control subsystem in real time;

[0069] The pressure stabilization control subsystem is used to calculate the pressure difference in real time, calculate the modification value of the fiber laying end motion mechanism path according to the pressure difference in real time, and send the modification value of the fiber laying end motion mechanism path to the control instruction generation subsystem in real time;

[0070] The pressure difference = the difference between the real tape pressure and the target tape pressure;

[0071] The modification value of the fiber laying end motion mechanism path = kp' × P + ki' × ΣP + kd' × ΔP;

[0072] In the formula, P is the pressure difference at the current time of the current fiber laying process, ΣP is the sum of all pressure differences up to the current time of the current fiber laying process, ΔP = pressure difference at the current time of the current fiber laying process - pressure difference at the last time of the current fiber laying process; kp', ki', and kd' are the required controller parameter combinations found by continuously iterating and optimizing the reinforcement learning model through simulation in a virtual environment;

[0073] The control instruction generation subsystem is used to generate control instructions according to the modification value of the heater output power and the modification value of the fiber laying end motion mechanism path, and then send them to the fiber laying end function mechanism control subsystem and the fiber laying end motion mechanism control subsystem;

[0074] The fiber laying end function mechanism control subsystem is used to update the heater output power in real time after processing the control instructions;

[0075] The fiber laying end motion mechanism control subsystem is used to update the fiber laying end motion mechanism path in real time after processing the control instructions.

[0076] As a preferred technical solution:

[0077] The pressure component further comprises a compression roller, a supporting device, a heat insulation asbestos plate and a cylinder a, the central shaft of the compression roller is parallel to the front-rear direction, the compression roller, the supporting device, the heat insulation asbestos plate, the pressure sensor and the cylinder a are sequentially connected from left to right, the cylinder a is fixed on the main plate and is used for pushing the compression roller to the left to cooperate with the compression roller on the opposite side to extrude and generate compaction pressure; the heat insulation asbestos plate can insulate the heat on the compression roller to prevent damage to the pressure sensor;

[0078] The shearing component comprises a cylinder c and a blade, the cylinder c is used for driving the blade to shear the tape, and the cylinder c is fixed on the main plate;

[0079] The re-sending component comprises a driving wheel, a driven wheel, a motor and a cylinder b, the driving wheel and the driven wheel are vertically arranged and parallel to the left-right direction, the motor is used for driving the driving wheel to rotate, the cylinder b is used for pressing the driven wheel to the driving wheel, and the motor and the cylinder b are fixed on the main plate; before the tape is installed to start laying, the cylinder b is not started, and the driving wheel and the driven wheel are separated, when the tape is installed into the channel and passes between the driving wheel and the driven wheel, the cylinder b is started, the driven wheel is pressed to the driving wheel to realize the pressing of the tape, and the motor drives the driving wheel to rotate in the laying process, and the driven wheel is driven to extrude the tape together;

[0080] The tension component comprises a material roll and a brake, the material roll is used for winding the tape, and the material roll is forced to rotate under the driving of the re-sending component in the laying process to realize the feeding of the tape, the brake is used for providing torque for the material roll to maintain stable tension of the tape feeding, so that the tape feeding is not unstable under high-speed rotation of the material roll, and even the tape is not dropped, and the material roll and the brake are fixed on the main plate;

[0081] The heat source component further comprises a supporting plate, the heater is fixed on the main plate, and the infrared thermal imager is fixed on the main plate through the supporting plate;

[0082] The connecting piece is fixed on the main plate and connected with the tail end of the laying tail end movement mechanism.

[0083] Advantages:

[0084] (1) The mirror image laying technology is used, the dependence on the mold in the traditional laying technology is abandoned, the time and material cost required for manufacturing the mold are avoided, and the problem that different components need different molds is avoided, so that the production is more flexible, the diversified component shapes can be quickly adapted, and therefore the production efficiency is significantly improved.

[0085] (2)The present application obtains real ribbon temperature and pressure in real time, and uses the controller parameter combination optimized by the reinforcement learning model to calculate and adjust the heater output power and the path of the fiber laying end motion mechanism in real time, so as to ensure the stable control of temperature and pressure in the moldless mirror image fiber laying process, which effectively avoids the problems of uneven curing or too fast caused by temperature fluctuation, material performance decline and the like, and the defects of poor bonding performance, air bubbles and the like caused by pressure fluctuation, thereby significantly improving the final performance of the composite product. BRIEF DESCRIPTION OF DRAWINGS

[0086] Figure 1 It is a front view structural schematic diagram of the hardware part of the moldless mirror image fiber laying system of the present application.

[0087] Figure 2 It is a rear view structural schematic diagram of the left part of the hardware part of the moldless mirror image fiber laying system of the present application.

[0088] Figure 3 It is a front view structural schematic diagram of the right part of the hardware part of the moldless mirror image fiber laying system of the present application.

[0089] Figure 4 It is a schematic diagram of the fiber ribbon installation process.

[0090] Figure 5 It is a comparison of the fiber laying distance-temperature curves of the experimental group 1 and the control group 1.

[0091] Figure 6 It is a comparison of the fiber laying distance-temperature curves of the experimental group 2 and the control group 2.

[0092] Figure 7 It is a comparison of the fiber laying distance-temperature curves of the experimental group 3 and the control group 3.

[0093] Figure 8 It is a comparison of the fiber laying distance-pressure curves of the experimental group 4 and the control group 4.

[0094] Figure 9 It is a comparison of the fiber laying distance-pressure curves of the experimental group 5 and the control group 5.

[0095] Figure 10 It is a comparison of the fiber laying distance-pressure curves of the experimental group 6 and the control group 6.

[0096] Figure 11 It is a whole flowchart of the present application, and the loop from generating the system control instruction file to judging whether the fiber laying is finished constitutes the fiber laying process, the reward calculation, the model training, the PID parameter (i.e. kp, ki, and kd in the following) adjustment, and the PID parameter decision flow.

[0097] Figure 12A logic diagram for reward calculation, model training and PID parameter adjustment of the application;

[0098] Wherein, 1-pressure sensor, 2-cylinder a, 3-connector, 4-tobacco roll, 5-supporting plate, 6-heater, 7-infrared thermal imager, 8-cylinder c, 9-pressing roller, 10-driving wheel, 11-cylinder b, 12-following wheel, 13-motor, 14-blade, 15-brake, 16-heat insulation asbestos plate, 17-main body plate, 19-silk ribbon, 20-bracket. DETAILED DESCRIPTION

[0099] The application will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the application and not used to limit the scope of the application. Furthermore, it should be understood that after reading the content taught by the application, those skilled in the art can make various modifications or changes to the application, and these equivalent forms also fall within the scope defined by the appended claims of the application.

[0100] A moldless mirror image fiber laying system, as shown in Figure 1 、 Figure 2 、 Figure 3 、 Figure 11 and Figure 12 , comprising a hardware part and a control system;

[0101] As shown in Figures 1 to 3 , the hardware part comprises a symmetrical left part and a right part, and the right part comprises a main body plate 17, a pressure component, a shearing component, a re-feeding component, a tension component, a heat source component, a connector 3 and a fiber laying end motion mechanism;

[0102] As shown in Figure 1 、 Figure 3 , the pressure component comprises a pressure sensor 1, a pressing roller 9, a supporting device, a heat insulation asbestos plate 16 and a cylinder a 2; the central axis of the pressing roller 9 is parallel to the front-back direction, and the pressing roller 9, the supporting device, the heat insulation asbestos plate 16, the pressure sensor 1 and the cylinder a 2 are connected in sequence from left to right, and the cylinder a 2 is fixed on the main body plate 17 and used to push the pressing roller 9 to the left;

[0103] The shearing component comprises a cylinder c 8 and a blade 14, the cylinder c 8 is used to drive the blade 14 to shear the silk ribbon 19, and the cylinder c 8 is fixed on the main body plate 17;

[0104] As shown in Figure 2 、 Figure 3As shown, the re-sending component includes a driving wheel 10, a driven wheel 12, a motor 13 and a cylinder b 11, the driving wheel 10 and the driven wheel 12 are both arranged vertically and parallel to the left-right direction, the motor 13 is used to drive the driving wheel 10 to rotate, the cylinder b 11 is used to press the driven wheel 12 to the driving wheel 10, and the motor 13 and the cylinder b 11 are fixed on the main body plate 17;

[0105] The tension component includes a material roll 4 and a brake 15, the material roll 4 is used to wind the tape 19, and the brake 15 is used to provide torque for the material roll 4 to maintain stable tension of the tape 19, and the material roll 4 and the brake 15 are both fixed on the main body plate 17;

[0106] The heat source component includes a heater 6, an infrared thermal imager 7 and a support plate 5, the heater 6 is used to heat the tape 19, the infrared thermal imager 7 is used to shoot the tape 19 in real time to obtain temperature field data of the tape 19, and the heater 6 is fixed on the main body plate 17, and the infrared thermal imager 7 is fixed on the main body plate 17 through the support plate 5;

[0107] The connecting piece 3 is fixed on the main body plate 17 and connected with the end of the tape laying end movement mechanism;

[0108] The control system includes a temperature stability control subsystem, a pressure stability control subsystem, a control instruction generation subsystem, a tape laying end movement mechanism control subsystem and a tape laying end function mechanism control subsystem;

[0109] The infrared thermal imager 7 is also used to obtain real tape temperature in real time during the tape laying process, and send the real tape temperature to the temperature stability control subsystem in real time; the temperature stability control subsystem is used to calculate the temperature difference in real time, calculate the modification value of the heater output power according to the temperature difference in real time, and send the modification value of the heater output power to the control instruction generation subsystem in real time;

[0110] The temperature difference is the difference between the real tape temperature and the target tape temperature;

[0111] The modification value of the heater output power is kp*×T+ki*×ΣT+kd*×ΔT;

[0112] In the formula, T is the temperature difference at the current time of the current tape laying process, ΣT is the sum of all temperature differences up to the current time of the current tape laying process, and ΔT is the temperature difference at the current time of the current tape laying process minus the temperature difference at the last time of the current tape laying process; kp*, ki* and kd* are obtained through the following process:

[0113] (a) generate a set of kp, ki and kd;

[0114] (b) let i = 1;

[0115] (c) In the virtual space, the same simulation as the actual process is performed, and the i-th laying process is started;

[0116] (d) Let j = 1, and let the virtual heater output power = the heater output power in the actual process;

[0117] (e) The virtual heater output power and the corresponding heating time are input into the prediction model A, and the predicted tape temperature is output therefrom, the difference between the predicted tape temperature and the target tape temperature is calculated, and the j-th temperature difference of the i-th temperature difference sequence is obtained;

[0118] The acquisition process of the prediction model A is as follows:

[0119] (i) A data set is established;

[0120] The heater output power and the corresponding heating time and the real tape temperature in the historical process same as the current process are collected;

[0121] At the same time, the same simulation as the actual process is performed in the virtual space, and the virtual heater output power and the corresponding heating time and the virtual tape temperature are obtained;

[0122] (ii) Data augmentation is performed to improve the size of the data set, and a training set is obtained;

[0123] (iii) A deep learning model is established and trained, and the prediction model A is obtained;

[0124] (f) It is determined whether the i-th laying process is completed, if not, the modification value of the virtual heater output power is calculated using the following formula, the virtual heater output power is updated, j = j + 1 is simultaneously performed, and then step (e) is returned; otherwise, the next step is entered;

[0125] The modification value of the virtual heater output power = kp x T + ki x ΣT + kd x ΔT;

[0126] In the formula, T is the j-th temperature difference of the i-th temperature difference sequence, ΣT is the sum of all temperature differences of the i-th temperature difference sequence, ΔT = the j-th temperature difference of the i-th temperature difference sequence - the j-1-th temperature difference of the i-th temperature difference sequence, and when j = 1, ΔT = 0;

[0127] (g) The i-th temperature difference sequence is scored, and it is determined whether the score of the i-th temperature difference sequence reaches a set value, if yes, the last set of kp, ki and kd is output as kp*, ki* and kd*; otherwise, the next step is entered;

[0128] (h) judging whether all the filament laying processes are completed, if not, inputting the last set of kp, ki, kd and the i-th temperature difference sequence into the reinforcement learning model provided with the reward function, outputting a new set of kp, ki, kd from the reinforcement learning model, and then returning to step (c) with i = i + 1; otherwise, returning to step (a);

[0129] The pressure sensor 1 is used to acquire the real tape pressure in real time during the filament laying process and send the real tape pressure to the pressure stabilization control subsystem in real time; the pressure stabilization control subsystem is used to calculate the pressure difference in real time, calculate the modification value of the filament laying end motion mechanism path according to the pressure difference in real time, and send the modification value of the filament laying end motion mechanism path to the control instruction generation subsystem in real time;

[0130] The pressure difference = the difference between the real tape pressure and the target tape pressure;

[0131] The modification value of the filament laying end motion mechanism path = kp' x P + ki' x ΣP + kd' x ΔP;

[0132] In the formula, P is the pressure difference at the current time of the current filament laying process, ΣP is the sum of all pressure differences up to the current time of the current filament laying process, ΔP = the pressure difference at the current time of the current filament laying process - the pressure difference at the last time of the current filament laying process; kp', ki', and kd' are obtained through the following process:

[0133] (A) generating a set of kp, ki, and kd;

[0134] (B) setting i = 1;

[0135] (C) starting the i-th filament laying process in the virtual space by performing the same simulation as the actual process;

[0136] (D) setting j = 1, and setting the virtual filament laying end motion mechanism path = the filament laying end motion mechanism path in the actual process;

[0137] (E) inputting the virtual filament laying end motion mechanism path into the prediction model B, outputting the predicted tape pressure from the prediction model B, calculating the difference between the predicted tape pressure and the target tape pressure to obtain the j-th pressure difference of the i-th pressure difference sequence;

[0138] The acquisition process of the prediction model B is as follows:

[0139] (I) establishing a data set;

[0140] Collecting the filament laying end motion mechanism path and the real tape pressure in the historical process which is the same as the current process;

[0141] At the same time, the same simulation as the actual process is performed in the virtual space to obtain the virtual filament laying end motion mechanism path and the virtual tape pressure;

[0142] (II) Data augmentation, increase the size of the data set, get the training set;

[0143] (III) Establish and train deep learning model, that is, get prediction model B;

[0144] (F) Determine whether the i-th laying process is completed, if not, calculate the modification value of the virtual laying end motion mechanism path using the following formula, update the virtual laying end motion mechanism path, and then let j=j+1, return to step (E); otherwise, proceed to the next step;

[0145] The modification value of the virtual laying end motion mechanism path = kp*P + ki*Sigma P + kd*Delta P;

[0146] In the formula, P is the j-th pressure difference of the i-th pressure difference sequence, Sigma P is the sum of all pressure differences of the i-th pressure difference sequence, Delta P = the j-th pressure difference of the i-th pressure difference sequence - the j-1-th pressure difference of the i-th pressure difference sequence, when j=1, let Delta P=0;

[0147] (G) Score the i-th pressure difference sequence, and determine whether the score of the i-th pressure difference sequence reaches the set value, if yes, output the last set of kp, ki and kd as kp', ki' and kd'; otherwise, proceed to the next step;

[0148] (H) Determine whether all laying processes are completed, if not, input the last set of kp, ki and kd and the i-th pressure difference sequence into the reinforcement learning model with the reward function, output a new set of kp, ki and kd from the reinforcement learning model, let i=i+1, and return to step (C); otherwise, return to step (A);

[0149] The control instruction generation subsystem is used to generate control instructions according to the modification value of the heater output power and the modification value of the laying end motion mechanism path, and then send the control instructions to the laying end functional mechanism control subsystem and the laying end motion mechanism control subsystem;

[0150] The laying end functional mechanism control subsystem is used to update the heater output power in real time after processing the control instructions;

[0151] The laying end motion mechanism control subsystem is used to update the laying end motion mechanism path in real time after processing the control instructions.

[0152] Now taking two laying processes as an example, the steps of using the system of the present application to lay a wire are described as follows:

[0153] (1) Install the wire belt;

[0154] For example Figure 4After the ribbon is wound on the roll, the ribbon is passed between the driving wheel and the driven wheel, through the blade to the compression roller, and fixed on the support 20 which is a vertical frame or a truss fixed on the ground. Then, the connecting piece is connected with the end of the ribbon laying end movement mechanism.

[0155] When the ribbon passes between the driven wheel and the driving wheel, the air cylinder b is started to press the ribbon between the driven wheel and the driving wheel.

[0156] When the ribbon reaches the compression roller, the air cylinder a is started to push out the compression rollers on both sides to cooperate with the compression rollers on the opposite side to realize the compression of the ribbon.

[0157] (2) First time of laying the ribbon;

[0158] Firstly, the ribbon laying end movement mechanism starts to move along the ribbon laying path, at the same time, the heater is started to heat the ribbon, the infrared thermal imager starts to take the temperature field data of the ribbon laying area in real time and monitor the temperature of the ribbon, the pressure sensor obtains the pressure of the ribbon in real time, and the motor drives the driving wheel and the driven wheel to rotate to drive the ribbon to advance. In this process, the temperature stable control subsystem and the pressure stable control subsystem calculate the modification value of the output power of the heater and the modification value of the path of the ribbon laying end movement mechanism respectively according to the real-time data, and send them to the corresponding control subsystem (i.e. the ribbon laying end function mechanism control subsystem or the ribbon laying end movement mechanism control subsystem) through the control instruction generation subsystem, so as to adjust the heating power and the ribbon laying path in real time to ensure that the temperature and pressure of the ribbon are kept in the target range.

[0159] Then, when the laying is about to be completed, the air cylinder c is started to drive the blade to cut the ribbon and cut off the excess part. After cutting, the air cylinder c is reset.

[0160] Finally, the reserved ribbon between the blade and the compression roller is continued to be laid until the first time of laying is completed.

[0161] (3) Second time of laying the ribbon;

[0162] Firstly, the ribbon laying end movement mechanism moves to the initial position.

[0163] Then, the ribbon laying end movement mechanism starts to move along the ribbon laying path, and when the laying is about to be completed, the air cylinder c is started to drive the blade to cut the ribbon and cut off the excess part. After cutting, the air cylinder c is reset.

[0164] Finally, the reserved ribbon between the blade and the compression roller is continued to be laid until the second time of laying is completed, and the heater, the motor, the infrared thermal imager, the air cylinder a and the air cylinder b are turned off, and the connecting piece is disconnected with the end of the ribbon laying end movement mechanism.

[0165] In order to prove that the control system in the moldless mirror image filament laying system of the present application can accurately control temperature and pressure, the following experiments are carried out for verification:

[0166] Experiment group 1: using the moldless mirror image filament laying system of the present application, setting the target tape temperature to 230℃, the initial speed of the filament laying end motion mechanism to 10mm / s, the initial output power of the heater to 90W, and laying the filament for 300mm;

[0167] Control group 1: basically the same as experiment group 1, the only difference being that there is no control system in the moldless mirror image filament laying system;

[0168] Experiment group 2: using the moldless mirror image filament laying system of the present application, setting the target tape temperature to 380℃, the initial speed of the filament laying end motion mechanism to 50mm / s, the initial output power of the heater to 210W, and laying the filament for 300mm;

[0169] Control group 2: basically the same as experiment group 2, the only difference being that there is no control system in the moldless mirror image filament laying system;

[0170] Experiment group 3: using the moldless mirror image filament laying system of the present application, setting the target tape temperature to 500℃, the initial speed of the filament laying end motion mechanism to 10mm / s, the initial output power of the heater to 135W, and laying the filament for 300mm;

[0171] Control group 3: basically the same as experiment group 3, the only difference being that there is no control system in the moldless mirror image filament laying system;

[0172] Experiment group 4: using the moldless mirror image filament laying system of the present application, setting the target tape pressure to 15N, the spacing of the filament laying end motion mechanism in the vertical direction to 4mm, and laying the filament for 300mm;

[0173] Control group 4: basically the same as experiment group 4, the only difference being that there is no control system in the moldless mirror image filament laying system;

[0174] Experiment group 5: using the moldless mirror image filament laying system of the present application, setting the target tape pressure to 30N, the spacing of the filament laying end motion mechanism in the vertical direction to 4mm, and laying the filament for 300mm;

[0175] Control group 5: basically the same as experiment group 5, the only difference being that there is no control system in the moldless mirror image filament laying system;

[0176] Experiment group 6: using the moldless mirror image filament laying system of the present application, setting the target tape pressure to 45N, the spacing of the filament laying end motion mechanism in the vertical direction to 4mm, and laying the filament for 300mm;

[0177] Control group 6: basically the same as experiment group 6, the only difference being that there is no control system in the moldless mirror image filament laying system;

[0178] The comparison of the laying distance-temperature curves in the laying process of the experimental group 1 and the control group 1, the experimental group 2 and the control group 2, and the experimental group 3 and the control group 3 is shown in Figure 5 , Figure 6 and Figure 7 It can be seen from Figures 5 to 7 that, compared with the control groups 1-3, the experimental groups 1-3 can stably control the real ribbon temperature in a range close to the target ribbon temperature;

[0179] The comparison of the laying distance-pressure curves in the laying process of the experimental group 4 and the control group 4, the experimental group 5 and the control group 5, and the experimental group 6 and the control group 6 is shown in Figure 8 , Figure 9 and Figure 10 It can be seen from Figures 8 to 10 that, compared with the control groups 4-6, the experimental groups 4-6 can stably control the real ribbon pressure in a range close to the target ribbon pressure;

[0180] The above verification process and verification results are sufficient to prove that the control system in the moldless mirror image laying system can accurately control the temperature and pressure.

Claims

1. A mirror-image fiber placement method without a mold, characterized in that: Without using a mold, two ribbons are laid in a mirror-symmetrical manner and the laying process is repeated multiple times; During the laying process, the actual ribbon temperature is obtained in real time and the temperature difference is calculated in real time. The modified value of the heater output power is calculated in real time based on the temperature difference and the heater output power is updated in real time. Temperature difference = actual ribbon temperature - target ribbon temperature difference; Modified value of heater output power = kp*×T + ki*×ΣT + kd*×ΔT; Wherein, T is the temperature difference at the current moment of the current laying process, ΣT is the sum of all temperature differences of the current laying process up to the current moment, ΔT = the temperature difference at the current moment of the current laying process - the temperature difference at the previous moment of the current laying process; kp*, ki*, and kd* are controller parameter combinations that meet the requirements found by simulation in a virtual environment and continuous iterative optimization using a reinforcement learning model. The acquisition process is as follows: virtual laying is started with initial kp, ki, and kd, and the predicted ribbon temperature is obtained using prediction model A that predicts the ribbon temperature based on the virtual heater output power and the corresponding heating time. The difference between the predicted ribbon temperature and the target ribbon temperature is calculated, and closed-loop control is performed by dynamically adjusting the virtual heater output power. After a single simulation, if the temperature difference meets the standard, the current parameters kp*, ki*, and kd* are output; otherwise, the last set of kp, ki, kd and temperature difference data are input into the reinforcement learning model to generate new parameters for re-simulation and continuous iteration; During the laying process, the actual ribbon pressure is obtained in real time and the pressure difference is calculated in real time. The modified value of the movement mechanism path of the laying terminal is calculated in real time based on the pressure difference, and the movement mechanism path of the laying terminal is updated in real time; Pressure difference = actual ribbon pressure - target ribbon pressure difference; Modified value of the path of the fiber placement end motion mechanism = kp'×P + ki'×ΣP + kd'×ΔP; Wherein, P is the pressure difference at the current moment of the current fiber placement process, ΣP is the sum of all pressure differences in the current fiber placement process up to the current moment, ΔP = the pressure difference at the current moment of the current fiber placement process - the pressure difference at the previous moment of the current fiber placement process. kp', ki', and kd' are controller parameter combinations that meet the requirements and are found by simulation in a virtual environment and continuous iterative optimization using a reinforcement learning model. The acquisition process is as follows: virtual fiber placement is started with initial kp, ki, and kd. The predicted ribbon pressure is obtained using prediction model B based on the path of the fiber placement end motion mechanism. The difference between the predicted ribbon pressure and the target ribbon pressure is calculated. Closed-loop control is performed by dynamically adjusting the path of the virtual fiber placement end motion mechanism. After a single simulation, if the pressure difference meets the standard, the current parameters kp', ki', and kd' are output. Otherwise, the last set of kp, ki, kd and pressure difference data are input into the reinforcement learning model to generate new parameters for re-simulation and continuous iteration.

2. A patternless mirror-image fiber placement method according to claim 1, characterized in that: kp*, ki*, and kd* are obtained by the following process: (a) Generate a set of kp, ki, kd; (b) Let i = 1; (c) The same simulation as the actual process is performed in the virtual space and the i-th wire laying process begins; (d) Let j = 1, and let the virtual heater output power = the actual heater output power; (e) Inputting the virtual heater output power and the corresponding heating time into the prediction model A, which outputs the predicted ribbon temperature. The difference between the predicted ribbon temperature and the target ribbon temperature is calculated to obtain the jth temperature difference in the i-th temperature difference sequence; Prediction model A is a trained deep learning model. During training, the heater output power and the corresponding heating time are used as the input of the deep learning model, and the ribbon temperature is used as the theoretical output of the deep learning model. The parameters of the deep learning model are continuously adjusted. (f) Determine whether the i-th fiber placement process is completed. If not, calculate the modified value of the virtual heater output power using the following formula, update the virtual heater output power, set j = j + 1, and return to step (e). Otherwise, proceed to the next step. The modified value of the virtual heater output power = kp×T + ki×ΣT + kd×ΔT; Where T is the jth temperature difference in the i-th temperature difference sequence, ΣT is the sum of all temperature differences in the i-th temperature difference sequence, ΔT = the jth temperature difference in the i-th temperature difference sequence - the j-1th temperature difference in the i-th temperature difference sequence. When j = 1, let ΔT = 0; (g) Score the i-th temperature difference sequence and determine whether the score of the i-th temperature difference sequence reaches the set value. If so, output the last set of kp, ki, and kd as kp*, ki*, and kd*; otherwise, proceed to the next step; (h) Determine whether all the laying processes are completed. If not, input the last set of kp, ki, kd and the i-th temperature difference sequence into the reinforcement learning model with a reward function, which outputs a new set of kp, ki, kd. After setting i = i + 1, return to step (c); otherwise, return to step (a).

3. The method for patternless mirror-image fiber placement according to claim 2, characterized in that: The process of obtaining prediction model A is as follows: (i) Establishing a dataset; Collect the heater output power, corresponding heating time and actual ribbon temperature in the same historical process as the current process; At the same time, the same simulation as the actual process is performed in the virtual space to obtain the virtual heater output power and the corresponding heating time and virtual ribbon temperature; (ii) Data augmentation: increasing the size of the dataset to obtain a training set; (iii) Build and train a deep learning model to obtain prediction model A.

4. The method for patternless mirror-image fiber placement according to claim 1, characterized in that: kp', ki', and kd' are obtained by the following process: (A) Generate a set of kp, ki, kd; (B) Let i = 1; (C) The same simulation as the actual process is performed in the virtual space, and the i-th fiber placement process begins; (D) Let j = 1, and let the virtual fiber placement end motion mechanism path = the fiber placement end motion mechanism path in the actual process; (E) The virtual path of the fiber placement end motion mechanism is input into prediction model B, which outputs the predicted ribbon pressure. The difference between the predicted ribbon pressure and the target ribbon pressure is calculated to obtain the jth pressure difference in the i-th pressure difference sequence. Prediction model B is a trained deep learning model. During training, the path of the fiber placement end motion mechanism is used as the input of the deep learning model, and the ribbon pressure is used as the theoretical output of the deep learning model. The parameters of the deep learning model are continuously adjusted. (F) Determine whether the i-th fiber placement process is complete. If not, calculate the modified value of the virtual fiber placement end motion mechanism path using the following formula, update the virtual fiber placement end motion mechanism path, set j = j + 1, and return to step (E). Otherwise, proceed to the next step. The modified value of the virtual fiber placement end motion mechanism path = kp×P + ki×ΣP + kd×ΔP; Where P is the jth pressure difference in the i-th pressure difference sequence, ΣP is the sum of all pressure differences in the i-th pressure difference sequence, ΔP = the jth pressure difference in the i-th pressure difference sequence - the j-1th pressure difference in the i-th pressure difference sequence. When j = 1, let ΔP = 0; (G) Score the i-th pressure difference sequence and determine whether the score of the i-th pressure difference sequence reaches the set value. If so, output the last set of kp, ki, and kd as kp', ki', and kd'; Otherwise, proceed to the next step; (H) Determine whether all the laying processes are completed. If not, input the last set of kp, ki, kd and the i-th pressure difference sequence into the reinforcement learning model with a reward function, which outputs a new set of kp, ki, kd. After setting i = i + 1, return to step (C); otherwise, return to step (A).

5. The method for patternless mirror-image fiber placement according to claim 4, characterized in that: The process of obtaining prediction model B is as follows: (I) Establishing a data set; Collect the path of the end-of-line movement mechanism and the actual ribbon pressure in the historical process that is the same as the current process; At the same time, the same simulation as the actual process is performed in the virtual space to obtain the virtual fiber placement end motion mechanism path and virtual ribbon pressure; (II) Data augmentation: increasing the size of the dataset to obtain a training set; (III) Establish and train a deep learning model to obtain prediction model B.

6. A moldless mirror-image fiber placement system, characterized in that: Including hardware part and control system; The hardware part includes a symmetrical left part and a right part, the right part includes a main plate (17), a pressure component, a shearing component, a re-feeding component, a tension component, a heat source component, a connecting piece (3), and a wire laying end motion mechanism, the pressure component, the shearing component, the re-feeding component, and the tension component are arranged in sequence from front to back along the wire running direction; The pressure component includes a pressure sensor (1); The heat source component includes a heater (6) and an infrared thermal imager (7), wherein the heater (6) is used to heat the ribbon (19), and the infrared thermal imager (7) is used to photograph the ribbon laying area in real time to obtain temperature field data of the ribbon (19); The control system includes a temperature stabilization control subsystem, a pressure stabilization control subsystem, a control instruction generation subsystem, a wire placement terminal motion mechanism control subsystem, and a wire placement terminal functional mechanism control subsystem; The infrared thermal imager (7) is used to obtain the real ribbon temperature in real time during the laying process, and send the real ribbon temperature to the temperature stabilization control subsystem in real time; The temperature stabilization control subsystem is used to calculate the temperature difference in real time, calculate the modified value of the heater output power in real time based on the temperature difference, and send the modified value of the heater output power to the control instruction generation subsystem in real time; Temperature difference = actual ribbon temperature - target ribbon temperature difference; Modified value of heater output power = kp*×T + ki*×ΣT + kd*×ΔT; Wherein, T is the temperature difference at the current moment of the current laying process, ΣT is the sum of all temperature differences of the current laying process up to the current moment, ΔT = the temperature difference at the current moment of the current laying process - the temperature difference at the previous moment of the current laying process; kp*, ki*, and kd* are controller parameter combinations that meet the requirements found by simulation in a virtual environment and continuous iterative optimization using a reinforcement learning model. The acquisition process is as follows: virtual laying is started with initial kp, ki, and kd, and the predicted ribbon temperature is obtained using prediction model A that predicts the ribbon temperature based on the virtual heater output power and the corresponding heating time. The difference between the predicted ribbon temperature and the target ribbon temperature is calculated, and closed-loop control is performed by dynamically adjusting the virtual heater output power. After a single simulation, if the temperature difference meets the standard, the current parameters kp*, ki*, and kd* are output; otherwise, the last set of kp, ki, kd and temperature difference data are input into the reinforcement learning model to generate new parameters for re-simulation and continuous iteration; The pressure sensor (1) is used to obtain the real ribbon pressure in real time during the laying process, and send the real ribbon pressure to the pressure stabilization control subsystem in real time; The pressure stabilization control subsystem is used to calculate the pressure difference in real time, calculate the modified value of the fiber placement terminal motion mechanism path based on the pressure difference in real time, and send the modified value of the fiber placement terminal motion mechanism path to the control instruction generation subsystem in real time; Pressure difference = actual ribbon pressure - target ribbon pressure difference; Modified value of the path of the fiber placement end motion mechanism = kp'×P + ki'×ΣP + kd'×ΔP; Wherein, P is the pressure difference at the current moment of the current laying process, ΣP is the sum of all pressure differences in the current laying process up to the current moment, ΔP = the pressure difference at the current moment of the current laying process - the pressure difference at the previous moment of the current laying process; kp', ki', and kd' are controller parameter combinations that meet the requirements and are found by simulation in a virtual environment and continuous iterative optimization using a reinforcement learning model. The acquisition process is as follows: virtual laying is started with initial kp, ki, and kd, and the predicted ribbon pressure is obtained using prediction model B based on the path of the laying end motion mechanism to predict the ribbon pressure. The difference between the predicted ribbon pressure and the target ribbon pressure is calculated, and closed-loop control is performed by dynamically adjusting the path of the virtual laying end motion mechanism. After a single simulation, if the pressure difference meets the standard, the current parameters kp', ki', and kd' are output; otherwise, the last set of kp, ki, kd and pressure difference data are input into the reinforcement learning model to generate new parameters for re-simulation and continuous iteration; The control instruction generation subsystem is used to generate a control instruction according to the modified value of the heater output power and the modified value of the wire placement terminal motion mechanism path, and then send the control instruction to the wire placement terminal functional mechanism control subsystem and the wire placement terminal motion mechanism control subsystem; The fiber placement terminal functional mechanism control subsystem is used to process control instructions and update the heater output power in real time; The wire placement terminal motion mechanism control subsystem is used to process the control instructions and update the wire placement terminal motion mechanism path in real time.

7. The patternless mirror-image fiber placement system according to claim 6, characterized in that: The pressure component further comprises a pressure roller (9), a supporting device, an insulating asbestos board (16) and a cylinder a (2). The central axis of the pressure roller (9) is parallel to the front-back direction. The pressure roller (9), the supporting device, the insulating asbestos board (16), the pressure sensor (1) and the cylinder a (2) are connected in sequence from left to right. The cylinder a (2) is fixed on the main plate (17) and is used to push the pressure roller (9) to the left. The shearing component includes a cylinder c (8) and a blade (14), wherein the cylinder c (8) is used to drive the blade (14) to shear the ribbon (19), and the cylinder c (8) is fixed on the main plate (17); The re-feeding component includes a driving wheel (10), a driven wheel (12), a motor (13) and a cylinder b (11). The driving wheel (10) and the driven wheel (12) are arranged vertically and parallel to the left and right directions. The motor (13) is used to drive the driving wheel (10) to rotate. The cylinder b (11) is used to press the driven wheel (12) onto the driving wheel (10). The motor (13) and the cylinder b (11) are fixed to the main body plate (17). The tension component includes a material reel (4) and a brake (15), the material reel (4) is used to wind the ribbon (19), and the brake (15) is used to provide torque to the material reel (4) to maintain a stable tension force for wire feeding, and the material reel (4) and the brake (15) are both fixed on the main plate (17); The heat source component further includes a support plate (5), a heater (6) is fixed on the main plate (17), and an infrared thermal imager (7) is fixed on the main plate (17) via the support plate (5); The connecting piece (3) is fixed on the main plate (17) and connected to the end of the wire laying end motion mechanism.

Images