Carbon fiber preform double-arm human-shaped acupuncture robot system and acupuncture method

Abstract

The invention discloses a carbon fiber preform double-arm human-shaped needling robot system and a needling method, the system is divided into a mechanical structure module, a decision module and an algorithm module, the whole process from preform laying to needling is automatically completed through automatic equipment and self-set algorithm participation, the dependence on manual operation is greatly reduced, and the needling efficiency is improved. And high-efficiency automatic operation of the carbon fiber preform is realized. According to the design and implementation, the robot system is controlled through a double-arm cooperative control and collision avoidance algorithm, a sensing and feedback mechanism and technological parameter self-adaptive optimization, multi-axis linkage control, intelligent sensing and a deep learning algorithm are fused, real-time optimization can be conducted when complex tasks are executed, full-automatic operation of the needling module is achieved, and the working efficiency is improved. And the method has high flexibility and adaptivity.

Description

Technical Field

[0001] The present invention relates to the field of automated manufacturing of carbon fiber preforms, and particularly to a double-arm humanoid needle punching robot system for carbon fiber preforms and a needle punching method. Background Art

[0002] Due to its excellent properties such as high strength, light weight, and high temperature resistance, carbon fiber composites are widely used in fields such as aerospace, automotive, and energy. In the field of new energy vehicles, carbon-ceramic brake discs, as important components for lightweight design, have the advantages of light weight, corrosion resistance, good wear resistance, and strong thermal stability. However, the manufacturing process of carbon-ceramic brake disc preforms is complex, and traditional manufacturing methods have problems such as low material utilization rate, low production efficiency, and unstable product quality, which seriously restrict the wide application of carbon-ceramic brake discs.

[0003] Chinese Patent with Publication No. CN221837210U discloses a needle punching robot, which designs a simple robotic arm structure to replace manual needle punching operation of preforms. However, in essence, it still belongs to a simple single-arm mechanical structure. On the one hand, the needle punching efficiency is limited and it relies on an additional workbench. On the other hand, there is a lack of automation improvement, and key steps are missing in the industrialization and mass production levels. Therefore, there is an urgent need for a system and method that can achieve high-efficiency automated carbon fiber preform operations based on the field of needle punching robots. Summary of the Invention

[0004] Object of the Invention: To solve the problems mentioned in the background art, the present invention discloses a double-arm humanoid needle punching robot system for carbon fiber preforms and a needle punching method. Through the participation of automated equipment and self-developed algorithms, the whole process from preform laying to needle punching is automatically completed, greatly reducing the dependence on manual operation and achieving high-efficiency automated carbon fiber preform operations.

[0005] Technical Solution:

[0006] The present invention discloses a double-arm humanoid needle punching robot system for carbon fiber preforms, which includes a mechanical structure module, a decision-making module, and an algorithm module.

[0007] The mechanical structure module includes a monitoring module, a robot with a movable 7-DOF double arm, and a needle punching module. The robot is controlled by a shaft servo motor.

[0008] The monitoring module is used to monitor and obtain in real time the state of the carbon fiber preform on the execution end of the fixed arm of the robot and the process parameters of the needle punching module, and send them to the algorithm module.

[0009] The needle punching module is arranged at the execution end of the working arm of the robot and is used to perform processing operations on the carbon fiber preform when the fixed arm reaches the target position.

[0010] The decision-making module is connected to the shaft servo motor and the needle punching module, and is used to receive the working strategy given by the algorithm module, convert it into a control instruction, and transmit it to the shaft servo motor and the needle punching module;

[0011] The algorithm module constructs and outputs a working strategy:

[0012] Construct a dual-arm collaborative control and collision avoidance algorithm: Based on the forward and inverse kinematic algorithms, obtain the position of the needle punching module and the angles of the robotic arm axes, and plan a collision-free working path for the dual arms through the RRT algorithm;

[0013] Construct a process parameter adaptive optimization model based on LSTM + XGBoost: By inputting process parameters, predict quality indicators in real time, and inject new data increments into the training set to fine-tune the model and optimize process parameters; The input end of the algorithm module is connected to the monitoring module, and the output end is connected to the decision-making module.

[0014] Furthermore, the monitoring module includes a panoramic computer vision camera installed on the top of the 7-DOF dual-arm robot and a torque sensor configured on the needle punching module;

[0015] The panoramic computer vision camera is equipped with a high-resolution camera and multiple sensors, including an infrared sensor, a temperature and humidity sensor, and a force sensor. The camera scans and identifies the working environment to provide real-time image feedback. Combining with the built-in computer vision algorithm, it identifies the state parameters of the carbon fiber preform on the execution end of the fixed arm; The torque sensor is used to collect and feedback the process parameters of the needle punching module during operation.

[0016] Furthermore, the specific operation process of the dual-arm collaborative control and collision avoidance algorithm is as follows:

[0017] Taking the center of the robot base as the origin OA, the vertical downward direction as the positive x-axis direction, the horizontal rightward direction as the positive y-axis direction, and the needle punching module as the origin OB, define the direction of the local coordinate system. Assume the coordinates of the needle punching target point P in coordinate system B are (X, Y, Z), and convert the coordinates to the global coordinate system A. The coordinate transformation formula is:

[0018]

[0019] Where: is the rotation matrix from coordinate system B to coordinate system A, is the translation vector from coordinate system B to coordinate system A;

[0020] Let the length of the upper arm be L1, the length of the forearm be L2, and the angle between the upper arm and the forearm be α. According to the cosine theorem, obtain the rotation angles θ1 of the upper arm and θ2 of the forearm. The rotation angle of the wrist is a constant, representing the penetration depth of the acupuncture head, and its value is determined by factors such as the composite material type. The rotation angle of the acupuncture head is also a constant, representing the penetration depth of the acupuncture head;

[0021] The motion vector ξ of the robotic arm can be expressed as:

[0022] ξ = (θ1, θ2, θ3, θ4, θ5)

[0023] where θ3, θ4, and θ5 are the rotation angles of the other three degrees of freedom respectively;

[0024] For dual-arm collaborative control and path planning, use the Rapidly-exploring Random Tree (RRT) algorithm to select the initial point and the target point. Randomly select a node from the generated nodes, find the nearest node, generate a new node between them, and check if there is a collision with the obstacle. If not, add the new node to the tree; gradually expand the tree until a collision-free path from the starting point to the ending point is found.

[0025] Ensure that the minimum safe distance is maintained between the two arms. The collision avoidance condition is:

[0026]

[0027] where (X1, Y1, Z1) and (X2, Y2, Z2) are the positions of the end effectors of the two arms respectively.

[0028] Furthermore, the construction of the process parameter adaptive optimization model based on LSTM + XGBoost is specifically as follows:

[0029] Collect the initial data set, obtain the process parameters and the corresponding quality indicators to make the initial data set

[0030] Use the LSTM model to extract features from the time series data, and obtain the output of the hidden layer through the last layer of the LSTM model, that is, the feature vector;

[0031] Take the feature vector as the input, and train the XGBoost model to effectively process high-dimensional sparse data, and it has good interpretability and high prediction accuracy;

[0032] Construct the process parameter adaptive optimization model according to the stacking method model fusion strategy. The input is the output of the LSTM and XGBoost models, and the goal is to learn how to best combine the prediction results of these two models to generate the final process parameter adaptive optimization model;

[0033] Update the model parameters at intervals, collect the new data obtained by the real-time acquisition and monitoring module and add it to the training set, and fine-tune the model. When quality problems are detected, optimize the production process by adjusting the process parameters.

[0034] Furthermore, the algorithm module further includes a sensor fusion algorithm, which obtains the data transmitted by the monitoring module in real time, and comprehensively adjusts the motion strategy PID control algorithm, which is used to accurately control the force of the acupuncture module during acupuncture operations, and corrects the acupuncture operations based on the error signal. The formula of the PID controller is:

[0035]

[0036] where u(t) is the control signal, e(t) is the error signal (the difference between the actual value and the expected value), K p is the proportional coefficient, K i is the integral coefficient, K d is the differential coefficient;

[0037] Let the force measured by the force sensor be F m , and the expected force be F e , then the error signal is:

[0038] e(t) = F e - F m .

[0039] Furthermore, to ensure that multiple tasks can be executed in parallel and avoid task conflicts, the algorithm module also introduces a multi-threaded scheduling system and a priority task scheduling algorithm, intelligently allocates computing resources, changes the job priority according to the task requirements and sends it to the decision-making module to form a priority instruction, and controls the multi-process operation of the robot or operates according to the task priority.

[0040] Furthermore, the present invention discloses a method for acupuncture of a carbon fiber preform double-arm humanoid acupuncture robot, and the method includes the following steps:

[0041] S1: Start the robot and complete the system self-check. Use the monitoring module to scan and identify the position and state of the carbon fiber preform and send it to the algorithm module;

[0042] S2: The algorithm module constructs and runs a double-arm cooperative control and collision avoidance algorithm: Based on the forward kinematics algorithm, calculate the position and posture of the hand according to the axial angles of the robot arms of the movable 7-DOF double arms, use the inverse kinematics algorithm to determine the angles that each joint needs to reach to reach the predetermined position, and apply the RRT algorithm to plan a collision-free path for the two arms to form a working strategy;

[0043] S3: The algorithm module sends the working strategy to the decision-making module to be converted into control instructions. The fixed arm is moved to a predetermined position by controlling the shaft servo motor, and the carbon fiber preform is fixed at the execution end. The operating arm starts the needle punching operation through the needle punching module according to the preset process parameters of the needle punching module. The needle punching module feeds back the needle punching data to the algorithm module in real time through the torque sensor;

[0044] S4: The algorithm module inputs the preset process parameters to form a data set, trains a process parameter adaptive optimization model based on LSTM+XGBoost to optimize the process parameters and feeds back the working strategy to the decision-making module. The decision-making module is converted into control instructions and sent to the needle punching module to adjust the process parameters of the needle punching module in real time for operation;

[0045] S5: After the needle punching operation is completed, the operator checks or replaces the preform, and the fixed arm releases the carbon fiber preform to prepare for the next round of operation.

[0046] Advantageous effects:

[0047] 1. Through the replaceable design of the needle punching module and the combination with a movable 7-DOF dual-arm robot, the present invention enables more complex needle punching tasks of carbon fiber preforms to be realized under the premise of high precision and multi-degree-of-freedom movement. At the same time, it can more flexibly handle various tasks including needle punching, winding, and cutting, improving the applicability of the system and method and perfecting the operation chain.

[0048] 2. While maintaining the advantages of high-flexibility needle punching operation of the dual-arm robot, the present invention overcomes the possible collision of the two arms of the robot during the working process through the design of the dual-arm cooperative control and collision avoidance algorithm, which may lead to failures and errors in the needle punching operation, prolongs the equipment life and reduces the production cost, and improves the overall fluency of the needle punching operation and the system stability.

[0049] 3. By constructing a process parameter adaptive optimization model based on LSTM+XGBoost, the present invention feeds back and optimizes the process parameters of the needle punching module in real time, thereby further reducing errors, improving accuracy, and reducing the probability of defective products during the needle punching operation. Description of the drawings

[0050] Figure 1 is the hardware framework flowchart of the present invention;

[0051] Figure 2 is the operation flowchart of the present invention;

[0052] Figure 3 is the specific structural schematic diagram of the embodiment of the present invention. Detailed implementation manners

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0054] As Figures 1-3 shown, the present invention discloses a carbon fiber preform double-arm humanoid needle-punching robot system, which includes a mechanical structure module, a decision-making module, and an algorithm module.

[0055] The mechanical structure module includes a monitoring module, a robot with a movable 7-DOF double arm, and a needle-punching module. The movable 7-DOF double-arm robot is controlled by a shaft servo motor.

[0056] Each mechanical arm of the movable 7-DOF double-arm robot is configured with 7 degrees of freedom (7-DOF), including shoulder rotation (2-DOF), elbow pitch (2-DOF), wrist flip (2-DOF), and end effector rotation (1-DOF), supporting complex space trajectory planning; the drive module uses a harmonic reducer (accuracy ≤ 1 arc minute) and a high-torque servo motor (peak torque 30 N·m) to ensure high speed (maximum movement speed 2 m / s) and high-precision coordination; lightweight design, with a carbon fiber composite material mechanical arm skeleton (density 1.6 g / cm 3 ), reducing the weight by 40% compared with traditional aluminum alloy.

[0057] The torso and rotating shaft design allows for "360-degree rotation" through the rotating shaft. However, to avoid collisions and increase stability, the maximum rotation angle is set to 180 degrees, which can ensure that the robot can quickly adjust its posture when facing different working positions, enabling the arm to accurately align with the working target.

[0058] The monitoring module is used to monitor and obtain in real time the state of the carbon fiber preform on the execution end of the fixed arm of the robot and the process parameters of the needle-punching module, and send them to the algorithm module. The monitoring module includes a panoramic computer vision camera installed on the top of the movable 7-DOF double-arm robot and a torque sensor configured in the needle-punching module. The panoramic computer vision camera is equipped with a high-resolution camera and a variety of sensors, including infrared sensors, temperature and humidity sensors, and force sensors. The camera scans and identifies the working environment to provide real-time image feedback, and combines the built-in computer vision algorithm to identify the state parameters of the carbon fiber preform on the execution end of the fixed arm. The torque sensor is used to collect and feedback the process parameters of the needle-punching module during its operation.

[0059] The needle punching module is arranged at the execution end of the robotic arm. When the fixed arm reaches the target position, the needle punching module performs processing operations on the carbon fiber preform.

[0060] The needle punching module is a replaceable part, and the replacement modules are the winding module and the cutting module. After the system controls the needle punching module to complete the needle punching operation, the winding module and the cutting module are replaced according to the operation requirements to perform winding and cutting operations on the carbon fiber preform.

[0061] The winding module includes carbon fiber tow tension control and laying path optimization. Among them, the carbon fiber tow tension control is based on a magnetic particle brake and a PID algorithm, and the tension fluctuation ≤ ±5% (standard value 50 - 200N). The laying path optimization is achieved through a B-spline curve interpolation algorithm to realize uniform winding of complex surfaces.

[0062] The cutting module includes a laser cutting head and a cooling system. Among them, the fiber laser of the laser cutting head (wavelength 1070nm, power 500W), the cutting accuracy is ±0.02mm, and it supports synchronous cutting of carbon fiber and ceramic matrix. The cooling system ensures that the flatness of the cutting surface Ra ≤ 1.6μm through circulating water cooling (flow rate 10L / min).

[0063] The decision-making module is connected to the shaft servo motor and the needle punching module, and is used to receive the working strategy given by the algorithm module and convert it into control instructions, which are transmitted to the shaft servo motor and the needle punching module.

[0064] The input end of the algorithm module is connected to the monitoring module, and the output end is connected to the decision-making module. The algorithm module constructs and outputs a working strategy:

[0065] Construct a dual-arm collaborative control and collision avoidance algorithm: Based on the forward and inverse kinematic algorithms, obtain the position of the needle punching module and the 7-DOF angle, and plan a collision-free working path for the two arms through the RRT algorithm.

[0066] The specific operation process of the dual-arm collaborative control and collision avoidance algorithm is as follows:

[0067] Taking the center of the robot base as the origin OA, the vertical downward direction as the positive x-axis direction, and the horizontal right direction as the positive y-axis direction, and taking the needle punching module as the origin OB, define the direction of the local coordinate system. Assume that the coordinates of the needle punching target point P in the coordinate system B are (X, Y, Z), and convert the coordinates to the global coordinate system A. The coordinate transformation formula is:

[0068]

[0069] Among them: is the rotation matrix from coordinate system B to coordinate system A, is the translation vector from coordinate system B to coordinate system A;

[0070] Suppose the rotation angles between two coordinate systems are α x , α y , α z Then the rotation matrix can be expressed as:

[0071]

[0072] The translation vector can be directly determined by the relative positions of the two coordinate systems;

[0073] Let the length of the upper arm be L1, the length of the lower arm be L2, and the angle between the upper arm and the lower arm be α. According to the cosine theorem, we can obtain:

[0074]

[0075] The rotation angle of the lower arm is:

[0076] θ2 = π - α

[0077] There are two solutions for the rotation angle of the upper arm, which are respectively:

[0078] θ1 = arctan2(Y, X) - arctan2(L2sin(α), L1 + L2cos(α))

[0079] or

[0080] θ′1 = arctan2(Y, X) + arctan2(L2sin(α), L1 + L2cos(α))

[0081] The rotation angle of the wrist is a constant, representing the penetration depth of the needle head, and its value is determined by factors such as the composite material type. The rotation angle of the needle head is also a constant, representing the penetration depth of the needle head, and its value is also determined by various factors.

[0082] The motion vector ξ of the manipulator can be expressed as:

[0083] ξ = (θ1, θ2, θ3, θ4, θ5)

[0084] where θ3, θ4, θ5 are the rotation angles of the other three degrees of freedom respectively.

[0085] For dual-arm collaborative control and path planning, use the Rapidly-Exploring Random Tree (RRT) algorithm to select the initial point and the target point. Randomly select a node from the generated nodes, find the nearest node, generate a new node between them, and check if there is a collision with the obstacle. If not, add the new node to the tree; gradually expand the tree until a collision-free path from the starting point to the ending point is found.

[0086] Ensure a minimum safe distance between the two arms. The condition to avoid collision is:

[0087]

[0088] Among them, (X1, Y1, Z1) and (X2, Y2, Z2) are the positions of the end effectors of the two arms respectively.

[0089] The algorithm module also includes a sensor fusion algorithm to obtain the data transmitted by the monitoring module in real time, and a comprehensive adjustment motion strategy PID control algorithm to accurately control the force of the acupuncture module during acupuncture operations, and correct the acupuncture operations based on the error signal. The PID controller formula is:

[0090]

[0091] Among them, u(t) is the control signal, e(t) is the error signal (the difference between the actual value and the expected value), K p is the proportionality coefficient, K i is the integral coefficient, K d is the differential coefficient; Let the force measured by the force sensor be F m , and the expected force be F e , then the error signal is:

[0092] e(t) = F e -F m

[0093] Construct a process parameter adaptive optimization model based on LSTM + XGBoost: By inputting process parameters, predict quality indicators in real time, and inject new data incrementally into the training set to fine-tune the model and optimize process parameters; The construction of the process parameter adaptive optimization model based on LSTM + XGBoost is as follows:

[0094] Collect the initial data set, obtain process parameters and corresponding quality indicators to make the initial data set

[0095] Use the LSTM model to extract features from time series data, and obtain the hidden layer output, that is, the feature vector, through the last layer of the LSTM model;

[0096] Use the feature vector as the input to train the XGBoost model to effectively process high-dimensional sparse data, and it has good interpretability and high prediction accuracy;

[0097] Construct a process parameter adaptive optimization model according to the stacked model fusion strategy. The input is the output of the LSTM and XGBoost models, and the goal is to learn how to best combine the prediction results of these two models to generate the final process parameter adaptive optimization model;

[0098] Update the model parameters at intervals, collect new data in real time and add it to the training set, and fine-tune the model. When quality problems are detected, that is, process defects or abnormal conditions that do not meet quality standards during the needle punching process, such as delamination defects (delamination rate), uneven fiber distribution, uneven density or excessive porosity, dimensional consistency problems, surface defects or damage. Optimize the production process by adjusting process parameters.

[0099] To ensure that multiple tasks can be executed in parallel and avoid task conflicts, the algorithm module also introduces a multi-threaded scheduling system and a priority task scheduling algorithm, intelligently allocates computing resources, changes the job priority according to task requirements and sends it to the decision-making module to form a priority instruction, controls the multi-process operation of the robot or operates according to task priorities.

[0100] The present invention also discloses a needle punching method for a carbon fiber preform double-arm humanoid needle punching robot, and the method steps are as follows:

[0101] S1: Start the robot and complete the system self-check, use the monitoring module to scan and identify the position and state of the carbon fiber preform and send it to the algorithm module;

[0102] S2: The algorithm module constructs and runs a double-arm cooperation control and collision avoidance algorithm: Based on the forward kinematics algorithm, calculate the position and posture of the hand according to the axis angles of the robot arms of the movable 7-DOF double arms, use the inverse kinematics algorithm to determine the angles that each joint needs to reach to reach the predetermined position, and apply the RRT algorithm to plan a collision-free path for the two arms to form a working strategy;

[0103] S3: The algorithm module sends the working strategy to the decision-making module to be converted into a control instruction, moves the fixed arm to the predetermined position by controlling the axis servo motor, fixes the carbon fiber preform at the execution end, and the operating arm starts the needle punching operation through the needle punching module according to the preset process parameters of the needle punching module. The needle punching module feeds back the needle punching data to the algorithm module in real time through the torque sensor;

[0104] S4: The algorithm module inputs the preset process parameters to form a data set, trains a process parameter adaptive optimization model based on LSTM+XGBoost to optimize the process parameters and feeds back the working strategy to the decision-making module. The decision-making module is converted into a control instruction and sent to the needle punching module to adjust the process parameters of the needle punching module in real time for operation;

[0105] S5: After the needle punching operation is completed, the operator checks or replaces the preform, and the fixed arm releases the carbon fiber preform to prepare for the next round of operation. The comparison between the present invention and the performance of artificial parts is shown in Table 1:

[0106] Table 1

[0107]

[0108] The delamination defect rate decreased from 15% to 2.3%; the fiber utilization rate increased from 58% to 92%; the CPK value of dimensional consistency increased from 1.2 to 1.8.

[0109] The above description of the embodiments enables those skilled in the art to implement or use the present invention. Various modifications to the embodiments will be obvious to those skilled in the art. The general principles of the present invention can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention should not be limited to the embodiments shown herein, but should cover the widest scope that conforms to the principles and novel features disclosed in the present invention.

Claims

1. A carbon fiber preform double-arm humanoid needle punching robot system, characterized in that, The system includes a mechanical structure module, a decision-making module, and an algorithm module; The mechanical structure module includes a monitoring module, a robot with a movable 7-DOF dual-arm, and a needle punching module, and the robot is controlled by a shaft servo motor; The monitoring module is used to monitor and obtain in real time the state of the carbon fiber preform at the execution end of the fixed arm of the robot and the process parameters of the needle punching module, and send them to the algorithm module; The needle punching module is arranged at the execution end of the working arm of the robot and is used to perform processing operations on the carbon fiber preform when the fixed arm reaches the target position; The decision-making module is connected to the shaft servo motor and the needle punching module and is used to receive the working strategy given by the algorithm module and convert it into a control instruction, which is transmitted to the shaft servo motor and the needle punching module; The algorithm module constructs and outputs a working strategy: Construct a dual-arm cooperative control and collision avoidance algorithm: Based on the forward and inverse kinematic algorithms, obtain the position of the needle punching module and the angles of the robotic arm axes, and plan a collision-free working path for the dual-arm through the RRT algorithm; Construct a process parameter adaptive optimization model based on LSTM+XGBoost: By inputting process parameters, predict the quality index in real time, and inject new data increments into the training set to achieve model fine-tuning and optimize process parameters; The input end of the algorithm module is connected to the monitoring module, and the output end is connected to the decision-making module.

2. The carbon fiber preform double-arm humanoid needle punching robot system according to claim 1, wherein The monitoring module includes an omnidirectional computer vision camera installed on the top of the robot with a movable 7-DOF dual-arm and a torque sensor configured on the needle punching module; The omnidirectional computer vision camera is equipped with a high-resolution camera and a variety of sensors, including an infrared sensor, a temperature and humidity sensor, and a force sensor. The camera scans and identifies the working environment to provide real-time image feedback. Combining with the built-in computer vision algorithm, it identifies the state parameters of the carbon fiber preform at the execution end of the fixed arm; The torque sensor is used to collect and feedback the process parameters of the needle punching module when the needle punching module is working.

3. The carbon fiber preform double-arm humanoid needle punching robot system according to claim 2, wherein The specific operation process of the dual-arm cooperative control and collision avoidance algorithm is as follows: Taking the center of the robot base as the origin OA, the vertical downward direction as the positive x-axis direction, and the horizontal right direction as the positive y-axis direction, and taking the needle punching module as the origin OB, define the direction of the local coordinate system. Assume that the coordinates of the needle punching target point P in the coordinate system B are (X, Y, Z), and convert the coordinates to the global coordinate system A. The coordinate transformation formula is: Wherein: is the rotation matrix from coordinate system B to coordinate system A, is the translation vector from coordinate system B to coordinate system A; Let the length of the upper arm be L1, the length of the lower arm be L2, and the angle between the upper arm and the lower arm be α. According to the cosine theorem, obtain the rotation angles θ1 of the upper arm and θ2 of the lower arm. The rotation angle of the wrist is a constant, which represents the penetration depth of the needle tip, and its value is determined by factors such as the type of composite material. The rotation angle of the needle tip is also a constant, which represents the penetration depth of the needle tip; The motion vector ξ of the manipulator can be expressed as: ξ = (θ1, θ2, θ3, θ4, θ5) Among them, θ3, θ4, θ5 are the rotation angles of the other three degrees of freedom respectively; Two-arm collaborative control and path planning. Use the Rapidly-exploring Random Tree (RRT) algorithm to select the initial point and the target point. Randomly select a node from the generated nodes, find the nearest node, generate a new node between them, and check if there is a collision with the obstacle. If not, add the new node to the tree. Gradually expand the tree until a collision-free path from the starting point to the ending point is found. Ensure a minimum safety distance between the two arms. The collision avoidance condition is: where (X1, Y1, Z1) and (X2, Y2, Z2) are the positions of the end effectors of the two arms respectively.

4. The carbon fiber preform double-arm humanoid needle punching robot system according to claim 3, characterized in that, The construction of the process parameter adaptive optimization model based on LSTM + XGBoost is as follows: Collect the initial data set, obtain the process parameters and the corresponding quality indicators to make the initial data set. Use the LSTM model to extract features from the time series data, and obtain the hidden layer output, that is, the feature vector, through the last layer of the LSTM model. Take the feature vector as the input, and train the XGBoost model to effectively process high-dimensional sparse data, and it has good interpretability and high prediction accuracy. Construct the process parameter adaptive optimization model according to the stacking method model fusion strategy. The input is the output of the LSTM and XGBoost models, and the goal is to learn how to best combine the prediction results of these two models to generate the final process parameter adaptive optimization model. Update the model parameters at intervals, collect the new data obtained by the real-time acquisition monitoring module and add it to the training set, and fine-tune the model. When a quality problem is detected, optimize the production process by adjusting the process parameters.

5. The carbon fiber preform double-arm humanoid needle punching robot system according to claim 1, characterized in that, The algorithm module also includes a sensor fusion algorithm to obtain the data transmitted by the monitoring module in real time, and comprehensively adjust the motion strategy PID control algorithm, which is used to accurately control the force of the needle punching operation of the needle punching module, and correct the needle punching operation based on the error signal. The PID controller formula is: where, u(t) is the control signal, e(t) is the error signal (the difference between the actual value and the expected value), K p is the proportionality coefficient, K i is the integral coefficient, K d is the differential coefficient; Let the force measured by the force sensor be F m , and the desired force be F e , then the error signal is: e(t) = F e -F m 。 6. The carbon fiber preform double-arm humanoid needle punching robot system according to claim 1, characterized in that To ensure that multiple tasks can be executed in parallel and avoid task conflicts, the algorithm module also introduces a multi-thread scheduling system and a priority task scheduling algorithm, intelligently allocate computing resources, change the job priority according to the task requirements and send it to the decision-making module to form a priority instruction, and control the multi-process operation of the robot or operate according to the task priority.

7. The carbon fiber preform double-arm humanoid needle punching robot system according to claim 1, characterized in that, The needle punching module is a replaceable part, and the replacement modules are a winding module and a cutting module. After the system controls the needle punching module to complete the needle punching operation, replace the winding module and the cutting module according to the operation requirements to perform the winding and cutting work of the carbon fiber preform.

8. The needling method of the carbon fiber preform double-arm humanoid needling robot according to any one of claims 1-7, characterized in that, The method includes the following steps: S1: Start the robot and complete the system self-check. Use the monitoring module to scan and identify the position and state of the carbon fiber preform and send it to the algorithm module. S2: The algorithm module constructs and runs the two-arm collaborative control and collision avoidance algorithm: Based on the forward kinematics algorithm, calculate the position and posture of the hand according to the axis angles of the robot arm of the movable 7-DOF two arms, use the inverse kinematics algorithm to determine the angles that each joint needs to reach to reach the predetermined position, and apply the RRT algorithm to plan a collision-free path for the two arms to form a working strategy. S3: The algorithm module sends the working strategy to the decision-making module to be converted into control instructions. The fixed arm is moved to a predetermined position by controlling the servo motor of the shaft part, and the carbon fiber preform is fixed at the execution end. The operating arm starts the needling operation through the needling module according to the preset process parameters of the needling module. The needling module feeds back the needling data to the algorithm module in real time through the torque sensor; S4: The algorithm module inputs the preset process parameters to form a data set, trains the process parameter adaptive optimization model constructed based on LSTM+XGBoost to optimize the process parameters and feeds back the working strategy to the decision-making module. The decision-making module is converted into control instructions and sent to the needling module to adjust the process parameters of the needling module in real time for operation; S5: After the needling operation is completed, the operator checks or replaces the preform, and the fixed arm releases the carbon fiber preform to prepare for the next round of operation.

Images