Mechanical arm 3D printing intelligent manufacturing method based on digital twinning

Abstract

The application discloses a mechanical arm 3D printing intelligent manufacturing method based on digital twinning, and relates to the technical field of digital twinning.The mechanical arm 3D printing intelligent manufacturing method based on digital twinning comprises a physical space module of a mechanical arm 3D printing system, data of actions, states and environments of the mechanical arm 3D printing system in an entity space of the mechanical arm 3D printing system is collected, and data transmission between the physical entity and the digital twinning body is established to form a data collection and transmission module.Aiming at the problem that the working state of the mechanical arm 3D printing cannot be predicted, a digital space module is introduced, mechanism description models and data driven models are constructed for the printing device, the working state is considered in real time, possible problems of the mechanical arm 3D printing are predicted, timely alarms are given, defect positions and weak areas of the printing parts are detected, the mechanical arm 3D printing device immediately repairs and reinforces the printing parts, and the digital twinning body model of the printing device is monitored in real time to monitor the state of the printing device in real time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning, in particular to a mechanical arm 3D printing intelligent manufacturing method based on digital twinning. BACKGROUND

[0002] Additive manufacturing technology, i.e. 3D printing technology, can generate digital three-dimensional models of any shape from computer graphics data without the aid of machining or any mold, and quickly manufacture various complex-shaped parts. Compared with traditional subtractive manufacturing, 3D printing technology has the advantages of no mold, short manufacturing cycle, and no limitation on model complexity. The traditional 3D printing printer adopts a 3-axis gantry structure, which has the following shortcomings. The 3D printer with gantry structure adopts a planar slicing layering method, which stacks materials layer by layer. This makes the materials stacked in a single direction, and there are defects in the forming direction. The single direction is low in efficiency, and the thickness of each layer is limited. When approaching the inclined or curved profile, the stepped protrusions will cause the surface of the part to be not smooth enough. The gantry structure limits the flexibility of 3D printing, and the size of the part to be printed is limited by the size of the 3D printer. The gantry structure needs to consume material to support the overhanging part of the printed part, and removing the support wastes material and damages the surface quality of the object. The gantry structure is fixed and is not naturally suitable for more complex requirements such as part repair processing.

[0003] Multi-axis robot is a standardized equipment and has been widely used in industrial applications for many years. Multi-axis robot can carry high load, support high-speed movement, and has great movement range and freedom. Multi-axis robot used for 3D printing can provide very flexible spatial positioning capability, break through the limitation of 3D printing in single direction stacking, support multi-direction forming, and curved surface forming to make the surface of the part more smooth. It breaks through the limitation of the size of the part by the gantry 3D printing structure, eliminates the waste of support material, and supports flexible reprocessing and repair of the part. However, multi-axis robot 3D printing has the following disadvantages. The mechanical arm has its own coordinate system which is different from the fixed structure of the gantry, and the mechanical arm coordinate system needs to be highly unified with the 3D printing coordinate system, otherwise it will seriously affect the precision. The complex task requirements of mechanical arm 3D printing need more complex and accurate inverse kinematics algorithm and trajectory planning algorithm support. Mechanical arm 3D printing lacks high-precision and high-synchronization simulation system to assist complex 3D printing tasks.

[0004] In the prior art, a visual auxiliary positioning device for mechanical arm 3D printing and a positioning method thereof are provided in a patent document with the publication number CN111590899B and the title "Visual auxiliary positioning device for mechanical arm 3D printing and positioning method thereof". The technical solution disclosed in the patent document is as follows: a calibration frame is used to determine the positional relationship between the mechanical arm and the visual positioning device, and the positioning frame is installed at the end of the mechanical arm. A spatial point position acquisition probe acquires the printing boundary. A visual positioning device is arranged on one side for coordinate position acquisition. A computer performs coordinate conversion operation and communicates with the mechanical arm to control the movement of the mechanical arm. The 3D printer trajectory planning in any direction is realized, and the 3D printing trajectory planning and calculation in any cross section and any direction are realized, which saves the complex site modeling and simplifies the preparation work before printing.

[0005] However, the existing mechanical arm 3D printing does not have a digital twin platform, it is difficult to monitor the 3D printing working state in real time, and it is difficult to predict problems in the multi-axis mechanical arm 3D printing process. It is difficult to ensure the accuracy in the implementation process of the mechanical arm 3D printing due to external disturbance and mechanical arm error. SUMMARY

[0006] Technical problem to be solved

[0007] In view of the deficiencies of the prior art, the present application provides a mechanical arm 3D printing intelligent manufacturing method based on digital twinning, which solves the problem that the existing mechanical arm 3D printing does not have a digital twin platform, it is difficult to monitor the 3D printing working state in real time, and it is difficult to predict problems in the multi-axis mechanical arm 3D printing process. It is difficult to ensure the accuracy in the implementation process of the mechanical arm 3D printing due to external disturbance and mechanical arm error.

[0008] Technical scheme

[0009] To achieve the above purpose, the present application is implemented by the following technical scheme: a mechanical arm 3D printing intelligent manufacturing method based on digital twinning, comprising the following steps:

[0010] S1, constructing a physical space module of the mechanical arm 3D printing system;

[0011] S2, collecting data on the action, state and environment of the mechanical arm 3D printing system in the entity space, and establishing data transmission between the physical entity and the digital twin, to construct a data collection and transmission module;

[0012] S3, constructing a digital space module of the mechanical arm 3D printing system, and completing high-fidelity mapping of the mechanical arm 3D printing device based on the digital twin;

[0013] S4, constructing an algorithm training and verification module to compare and verify the real data collected in S2 and the results predicted by the algorithm model;

[0014] S5, constructing a digital twin display and interaction module.

[0015] Further, the S1 constructing a physical space module specifically includes the following steps:

[0016] S11, configuring a 3D printing extrusion device at the end of the mechanical arm;

[0017] S12, building a camera positioning device, and connecting the mechanical arm coordinate system and the vision coordinate system through calibration equipment;

[0018] S13, building a control system of the mechanical arm and the 3D printing device, and controlling the printing head parameters at the end of the mechanical arm.

[0019] Further, the data collection in S2 is realized by installing various sensors on the mechanical arm 3D printing system. The collected data includes the joint angle, joint speed and acceleration of the mechanical arm, and the speed, position, temperature and humidity of the printing head.

[0020] The data transmission refers to transmitting the collected data to the digital space module, and the transmission methods include Bluetooth communication, wireless network communication, and local area network communication.

[0021] Further, the S3 of constructing a digital space module specifically includes the following steps:

[0022] S31, constructing a three-dimensional visualization model of the mechanical arm 3D printing system;

[0023] S32, constructing a mechanism model of the mechanical arm 3D printing system, which refers to an accurate mathematical model established according to the internal mechanism of the object or production process;

[0024] S33, constructing a data-driven model of the mechanical arm 3D printing system according to an algorithm, which refers to a superior robust model formed by processing and analyzing the collected data, extracting features, and training and fitting on the basis of data.

[0025] Further, the S32 of constructing a mechanism model of the mechanical arm 3D printing system includes establishing a forward and inverse kinematics model of the mechanical arm 3D printing device through a D-H parameter table method.

[0026] Further, the algorithm of constructing a data-driven model of the mechanical arm 3D printing system in S33 includes BP neural network, RBF neural network, convolutional neural network, and recurrent neural network algorithm.

[0027] Further, the S4 of constructing a digital space module specifically includes the following steps:

[0028] S41, collate and analyze the data collected by S2 and transmitted to the state parameter database, and extract data features;

[0029] S42, training motion planning algorithm model and verifying according to numerical twin system;

[0030] S43, training intelligent decision algorithm model composed of deep learning algorithm and verifying;

[0031] S44, training error prediction and intelligent compensation algorithm model and verifying, the error prediction and intelligent compensation algorithm model uses BP back propagation neural network.

[0032] Further, the motion planning algorithm model in S42 adopts RRT trajectory planning algorithm, and the principle of the RRT trajectory planning algorithm is as follows:

[0033] First step, set initial node X init , target node X goal and state sampling space M;

[0034] Second step, randomly sample to obtain sample point X rand ;

[0035] Third step, if X rand has no collision with obstacles, calculate the distance between X rand and all nodes in the generated node set to obtain the nearest node X near , then walk from X near to X rand with step length to generate a new node X new , if the connection line between X new and X near collides with the obstacle, then re-sample randomly;

[0036] Fourth step, repeat the cycle, generate random expansion tree, when the nodes in the tree reach the target area, select a path from X int to X goal as the planning path, and plan a safe printing path for the mechanical arm 3D printing system.

[0037] Further, the error prediction and intelligent compensation algorithm model in S44 takes the end coordinates, posture and printing device extrusion state of the mechanical arm 3D printing device as input, and the joint angle, speed control parameter of the mechanical arm as output, and fits, predicts and eliminates errors through BP back propagation neural network.

[0038] Further, the BP neural network loss function is:

[0039]

[0040] wherein t is a true value, and y is a model predicted value;

[0041] The BP neural network gradient is:

[0042]

[0043] wherein omega is a weight value of a neural network layer, x is an input value, f(omega T x) is a predicted value.

[0044] omega = omega + eta (t-y)f'(omega T x)

[0045] wherein omega is a weight value of a neural network layer, and eta is a learning rate of weight updating.

[0046] Advantages

[0047] The present application has the following advantages:

[0048] (1) The mechanical arm 3D printing intelligent manufacturing method based on digital twinning can predict the working state of the mechanical arm 3D printing, introduce a digital space module, construct a mechanism description model and a data driven model for the printing device, consider the working state in real time, predict possible problems of the mechanical arm 3D printing, timely alarm, detect the defect position and weak area of the printing part, and the mechanical arm 3D printing device can repair and reinforce it in real time, and the digital twinning model of the printing device can monitor the state of the printing device in real time.

[0049] (2) The mechanical arm 3D printing intelligent manufacturing method based on digital twinning introduces an error prediction and intelligent compensation algorithm, extracts features from the collected multi-dimensional data, trains a superior robust deep learning algorithm, optimizes the printing precision of the mechanical arm 3D printing, provides a virtual simulation model, can effectively reduce the cost and prevent the device loss and danger caused by decision-making mistakes.

[0050] Of course, any product implementing the present application does not necessarily need to achieve all the advantages described above at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 FIG. 1 is a schematic diagram of the mechanical arm 3D printing digital twinning system workflow of the present application;

[0052] Figure 2 FIG. 2 is a schematic diagram of the digital space module of the present application;

[0053] Figure 3 FIG. 3 is a schematic diagram of the overall data transmission and interaction mode of the digital twinning system of the present application. DETAILED DESCRIPTION

[0054] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0055] In the description of the present application, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inner", "periphery" and the like indicate the orientation or positional relationship, which are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the indicated component or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0056] Please refer to Figures 1-3 The embodiments of the present application provide a technical solution: a mechanical arm 3D printing intelligent manufacturing method based on digital twinning, as shown in Figure 1 The method comprises the following steps:

[0057] S1, constructing a physical space module of a mechanical arm 3D printing system;

[0058] S2, collecting data on the action, state and environment of the mechanical arm 3D printing system in the entity space, and establishing data transmission between the physical entity and the digital twin to construct a data collection and transmission module;

[0059] S3, constructing a digital space module of the mechanical arm 3D printing system, and completing high-fidelity mapping of the mechanical arm 3D printing device based on the digital twin;

[0060] S4, constructing an algorithm training and verification module to compare and verify the real data collected in S2 and the results predicted by the algorithm model;

[0061] S5, constructing a digital twin display and interaction module.

[0062] Specifically, S1 constructing the physical space module specifically comprises the following steps:

[0063] S11, configuring a 3D printing extrusion device at the end of the mechanical arm;

[0064] S12, building a camera positioning device to connect the mechanical arm coordinate system and the vision coordinate system through a calibration device;

[0065] S13, building a control system of the mechanical arm and the 3D printing device to control the parameters of the printing head at the end of the mechanical arm.

[0066] In this embodiment, a 3D printing base device is added to the end of the mechanical arm, and the 3D printing device is connected with the control module, and the camera positioning device assists the mechanical arm 3D printing device to accurately position the printing target. The mechanical arm and the printing target coordinate system need to be converted, the hand-eye calibration of the mechanical arm is performed, the conversion relationship between the end coordinate system and the camera coordinate system is solved, the control system is responsible for the motion control of the mechanical arm 3D printing, and the trajectory, speed and other parameters of the printing head at the end of the mechanical arm are controlled; the control system will use digital twins to combine control and feedback to realize the control of the mechanical arm 3D printing.

[0067] Specifically, the data collection in S2 is realized by installing various sensors on the mechanical arm 3D printing system, and the collected data includes the joint angle, joint speed and acceleration of the mechanical arm, and the speed, position, temperature and humidity of the printing head;

[0068] Data transmission refers to transmitting the collected data to the digital space module, and the transmission methods include Bluetooth communication, wireless network communication, and local area network communication.

[0069] In this embodiment, various sensors are installed on the mechanical arm 3D printing device to collect signals such as joint angle, joint speed, acceleration of the mechanical arm, and signals such as speed, position, temperature, and humidity of the printing head. The sensors are distributed on the mechanical arm body of the mechanical arm printing device, the 3D printing head device, and the like, and the data is transmitted to the data acquisition card through the lead-out line, and then connected with the control system. According to the collected printing head state parameters, the position and speed of the end of the mechanical arm are controlled to ensure the synchronization of the 3D printing process and the movement of the mechanical arm;

[0070] The data acquisition card receives sensor data and camera data, and transmits the data to the control system. The analog signals in the data acquisition card are converted into digital signals and transmitted to the control system. The collected data is transmitted to the digital space module by using methods such as Bluetooth communication, wireless network communication, and local area network communication.

[0071] Specifically, the construction of the digital space module in S3 specifically includes the following steps:

[0072] S31, constructing a three-dimensional visualization model of the mechanical arm 3D printing system;

[0073] S32, constructing a mechanism model of the mechanical arm 3D printing system, which is a precise mathematical model established according to the internal mechanism of the object or production process;

[0074] S33, constructing a data-driven model of the mechanical arm 3D printing system according to an algorithm, which is a superior robust model formed by processing and analyzing the collected data, extracting features, and training and fitting on the basis of data;

[0075] The mechanism model of the 3D printing system of the mechanical arm is constructed in S32, including establishing the forward and inverse kinematics model of the 3D printing device of the mechanical arm by the D-H parameter table method;

[0076] The algorithm for constructing the data-driven model of the 3D printing system of the mechanical arm in S33 includes the BP neural network, the RBF neural network, the convolutional neural network, and the recurrent neural network algorithm.

[0077] In the embodiment, the digital twin is constructed, and the digital twin completes the highly real mapping of the 3D printing device of the mechanical arm.

[0078] Specifically, S4 constructing the digital space module specifically includes the following steps:

[0079] S41, the data collected in S2 and transmitted to the state parameter database is sorted and analyzed, and the data features are extracted;

[0080] S42, the motion planning algorithm model is trained and verified according to the numerical twin system;

[0081] S43, the intelligent decision algorithm model composed of the deep learning algorithm is trained and verified;

[0082] S44, the error prediction and intelligent compensation algorithm model is trained and verified, and the error prediction and intelligent compensation algorithm model uses the BP back propagation neural network;

[0083] The motion planning algorithm model in S42 adopts the RRT trajectory planning algorithm, and the principle of the RRT trajectory planning algorithm is as follows:

[0084] First step, set the initial node X init , the target node X goal and the state sampling space M;

[0085] Second step, randomly sample to obtain a sampling point X rand ;

[0086] Third step, if X rand does not collide with the obstacle, the distance between X rand and all nodes in the generated node set is calculated to obtain the nearest node X near , and then X near moves to X rand with a step length of step to generate a new node X new , if the connection line between X new and X near collides with the obstacle, the random sampling is re-performed;

[0087] Fourth step, repeatedly cycle to generate a random expansion tree, and when the nodes in the tree reach the target area, a path from X int to Xgoal the path as a planning path, to plan a safe printing path for the robotic arm 3D printing system;

[0088] The error prediction and intelligent compensation algorithm model in S44 takes the end coordinates of the robotic arm 3D printing device, the posture, and the extrusion state of the printing device as input, and outputs the joint angles and speed control parameters of the robotic arm. The BP back propagation neural network is fitted to predict and eliminate errors;

[0089] The BP neural network loss function is:

[0090]

[0091] where t is the true value and y is the model prediction value;

[0092] The BP neural network gradient is:

[0093]

[0094] where ω is the weight value of the neural network layer, x is the input value, and f(ω T x) is the prediction value;

[0095] ω = ω + η(t-y)f'(ω T x)

[0096] where ω is the weight value of the neural network layer, and η is the learning rate of weight update.

[0097] In this embodiment, the real data collected by the sensor is compared with the result predicted by the algorithm model, and the evaluation method is used to evaluate the effect of the model. The simulation results of the mechanism model are compared with the prediction results, and the joint verification method of real data and reliable simulation data is used. The algorithm of step S3 for constructing the data-driven model needs to be trained and verified. These algorithms include BP neural network, RBF neural network, convolutional neural network, recurrent neural network, etc. Python and pytorch deep learning framework are used for programming. These algorithms are called by scripts written in C# in unity. The data required for algorithm training and verification comes from various running parameters stored in the database and various data simulated by the simulation model.

[0098] A 3D printing extrusion device is configured at the end of the robotic arm, and the 3D printing device is connected with the control module. The camera positioning device assists the robotic arm 3D printing device in accurately positioning the printing target. The robotic arm and the printing target coordinate system need to be converted, and the hand-eye calibration of the robotic arm is performed to solve the conversion relationship between the end coordinate system and the camera coordinate system;

[0099] The robot base and the calibration board are fixed, and the robot end coordinate system and the camera coordinate system will move with the robot. Assuming that a point in the calibration board coordinate system is P0, and in the base coordinate system is P1, the conversion relationship between the end coordinate system and the camera coordinate system can be solved according to the relationship between several coordinate systems. Assuming that P0 can be converted to the camera coordinate system through the conversion matrix T1, and then the coordinates can be converted to the end coordinate system according to the conversion matrix T2 of the camera and the end, and then the coordinates can be converted to the base coordinate system according to the robot conversion matrix T3, that is, the equation:

[0100] T3T2T1T0=P1

[0101] Move the robot several times, P0 and P1 will not change, and perform the second measurement:

[0102] T 3' T2T 1' P0=P1

[0103] Solving the simultaneous equations can obtain:

[0104] T 3' T2T 1' P0=T3T2T1;

[0105] Since T 3' , T 1' , T3, T1 are known, the conversion matrix T2 of the robot end coordinate system and the camera coordinate system can be solved. With the conversion matrix T2, the 3D printing device at the end of the robot can be assisted by vision.

[0106] After building the control system of the robot and the 3D printing device, the robot 3D printing motion control is responsible for controlling the trajectory, speed and other parameters of the printing head at the end of the robot. The control system will use digital twins to combine control and feedback to achieve control of the robot 3D printing.

[0107] Data collection is performed on the action, state, environment, etc. in the physical space of the robot 3D printing device, and data transmission between the physical entity and the digital twin is established.

[0108] A variety of sensors are installed in the robot 3D printing device to collect signals such as robot joint angle, joint speed, acceleration, and signals such as printing head speed, position, temperature, humidity. The sensors are distributed on the robot main body, 3D printing head device and the like of the robot printing device, and the data collected by the sensors are transmitted to the data acquisition card through wires, and then connected to the control system. According to the collected printing head state parameters, the robot end pose and speed are controlled to ensure the synchronization of the 3D printing process and the robot motion;

[0109] The data acquisition card receives sensor data and camera data, and transmits the data to the control system. The analog signals of the data acquisition card are converted into digital signals and transmitted to the control system. The collected data is transmitted to the digital space module using methods including Bluetooth communication, wireless network communication, and local area network communication.

[0110] A digital twin is constructed, and the digital twin completes a highly realistic mapping of the robotic arm 3D printing device. The robotic arm 3D printing digital space module is as shown in Figure 2

[0111] First, a three-dimensional visualization model is constructed for the robotic arm 3D printing system. According to the robotic arm and the printing device, a Solidworks assembly model is built. The model file is converted to.max format using 3Dmax, and the assembly model is imported into unity. Scripts are written in C# language in unity to construct the functions of virtual-real combination of the robotic arm digital twin, real-time data communication, etc.

[0112] Second, a mechanism model is constructed for the robotic arm 3D printing system, including establishing the forward and inverse kinematics model of the robotic arm 3D printing device through the D-H parameter table method. Admas is used to perform dynamics simulation on the robotic arm 3D printing device.

[0113] Ansys finite element simulation is used to perform statics simulation on the robotic arm 3D printing device. The forward and inverse kinematics model is constructed because the path parameters of the printing path cannot be directly used for robotic arm control during printing path planning. The pose of the end printing head needs to be calculated, and the inverse kinematics of the robotic arm is solved to complete the motion planning of the robotic arm.

[0114] The D-H parameter table of a certain robotic arm is as shown in the following table:

[0115] i α a d θ 1 0 0 0 [theta1] 2 -90° 0 0 [theta2] 3 0 [a2] [d3] [theta]3 4 -90° [a3] [d4] [theta]4 5 90° 0 0 [theta]5 6 -90° 0 0 [theta]6

[0116] Wherein, parameter a is the length value of the connecting rod, α is the torsion angle of the connecting rod, d is the offset distance of the connecting rod, θ is the joint angle, and the robotic arm coordinate is the transformation matrix converted from the joint i-1 coordinate system to the joint i coordinate system

[0117]

[0118] Wherein, s represents the sin function, c represents the cos function, and the values of α, a, d, and θ come from the DH parameter table.

[0119] The inverse kinematics is solved using the following kinematics equation, is a transformation matrix from the base coordinate system of a six-axis robotic arm to the end coordinate system:

[0120] ​

[0121] Finally, a data-driven model is constructed for the mechanical arm 3D printing system, the collected data is processed and analyzed, the features are extracted, and a superior robust model is trained and fitted based on the data, including a running state parameter database, a motion optimization algorithm, an intelligent decision-making algorithm, etc.

[0122] The intelligent decision-making algorithm uses the processed multi-modal data stored in the database in the digital space, combines the characteristics of the constructed mechanism description model, trains the state prediction algorithm for the mechanical arm 3D printing, and if the printing system is about to fail, it will issue a warning; according to the image recognition algorithm, the image data collected by the camera is detected, and whether the target workpiece printing has defects is detected, and the defect position is located, and the defect position is reinforced; according to the mechanical simulation, the place where the printed workpiece has excessive stress is detected and the position is located, and the position with excessive stress is printed and reinforced.

[0123] Combined with the mechanical arm stress state obtained by simulating the mechanism description model, and the angle acceleration and 3D printing head printing speed data collected by the current sensors, an error prediction and compensation model for optimizing 3D printing precision is trained through a deep learning neural network structure, which predicts the error in the actual printing process and intelligently compensates for it, improves the printing precision of the mechanical arm 3D printing system, and solves the error stacking problem caused by the multi-axis mechanical arm.

[0124] A trajectory planning algorithm model is constructed for the mechanical arm 3D printing. RRT trajectory planning algorithm is adopted, the mechanical arm 3D printing task environment is complex, and the end printing device needs to be planned for trajectory planning when printing complex parts. A collision-free route is planned. RRT algorithm is a general trajectory planning algorithm, which randomly samples the state space and detects the collision of the sampling points, avoiding the huge calculation amount caused by accurate modeling, and can well handle obstacle and differential constraint problems. The principle of RRT algorithm is as follows:

[0125] First step, set the initial node X init , the target node X goal and the state sampling space M;

[0126] Second step, randomly sample to get sampling point X rand ;

[0127] Third step, if X rand has no collision with obstacles, calculate the distance between X rand and all nodes in the generated node set to get the nearest node X near , then from X near to X rand with step length, generate a new node X new , if Xnew with X near collide with obstacles, then re-sampling;

[0128] Fourth step, repeat the cycle, generate random expansion tree, when the tree nodes reach the target area, select a path from X int to X goal as the planning path.

[0129] Using RRT algorithm for 3D printing system planning safe printing path for mechanical arm.

[0130] Using real data collected by sensors and algorithm model prediction results for comparison, using evaluation method to evaluate the effect of the model. Combined with the simulation results of the mechanism model and the prediction results, the joint verification method of real data and reliable simulation data is used.

[0131] The algorithm for building data-driven models needs to be trained and verified. These algorithms include BP neural network, RBF neural network, convolutional neural network, recurrent neural network, etc. Python and pytorch deep learning framework are used to write these algorithms, which are called by scripts written in C# in unity. The data required for algorithm training and verification comes from various running parameters stored in the database and various data simulated by the simulation model;

[0132] Sort and analyze the data collected and transmitted to the state parameter database in step S2, and extract the data features;

[0133] For the verification of the motion planning algorithm, the digital twin system has the inherent advantage of being able to test whether the motion trajectory collides with obstacles in the simulation space. Depending on high-precision simulation and camera data collection, the digital twin system will issue a warning for dangerous paths;

[0134] The intelligent decision-making algorithm is mainly composed of deep learning algorithms. The deep learning training model needs a training set to train the model, and the validation set is used to test the model error during neural network training. When the model error in the validation set reaches the minimum, the model training is considered complete. The test set is used to evaluate the final result of the model. For the image recognition algorithm of defect position detection, real images collected by the camera are used as test set data to verify the effectiveness of the training model.

[0135] The classification effect of image recognition task verified by confusion matrix is shown in the following table:

[0136] positive examples negative examples positive examples TP FN negative examples FP TN

[0137] Calculate the accuracy, precision, recall and other indicators of the model to evaluate the pros and cons of the algorithm training.

[0138] The error prediction and intelligent compensation model uses a BP back propagation neural network, trains the model with real data collected by sensors, and compensates for errors, thereby achieving the purpose of optimizing the accuracy of the 3D printing of the robot arm. The model input is the end coordinates, posture and printing device extrusion state of the 3D printing device of the robot arm, and the model output is the control parameters such as the angle and speed of each joint of the robot arm. Through neural network fitting, the effect of predicting and eliminating errors is achieved.

[0139] The BP neural network loss function, where t is the true value and y is the model prediction value:

[0140]

[0141] The BP neural network gradient, where ω is the weight value of the neural network layer, x is the input value, and f(ω T x) is the predicted value:

[0142]

[0143] The BP neural network back propagation weight update, where ω is the weight value of the neural network layer and η is the learning rate of the weight update:

[0144] ω = ω + η(t - y)f'(ω T x) x

[0145] The overall data transmission and interaction mode of the robot arm 3D printing digital twin system is as shown in Figure 3 The client written in Unity is used to display and interactively control the digital twin system. Based on the mechanism model and data-driven model, the digital twin model is constructed to realize the faithful mapping of the digital twin to the real physical entity, and the working state of the robot arm 3D printing device can be monitored online in real time. The optimization algorithm of the data-driven model can be used to analyze and optimize the robot arm 3D printing process in real time, and intelligent prediction can be achieved. The digital twin system is used for data bidirectional communication and reliable control of the robot arm 3D printing device.

[0146] The digital twin system display module has complete display functions. The robot arm 3D printing device model is displayed on the unity front-end interface, and various data of the device are displayed. Some of these data come from real-time collection by sensors, and some come from prediction and decision-making by the digital space module. These data are displayed in the form of pie charts, graphs and stress maps.

[0147] The digital twin system sets multiple control modes, sets a model import interface in the unity front end, and imports a to-be-printed component model in the interface. The system automatically processes the model. Interfaces for controlling the printing speed and printing specifications of the mechanical arm are set in the unity front end. With the assistance of visual algorithms, the selection of defective parts and other operations can be completed. The front end has multiple display modes, and the Unity game engine is used to write the mechanical arm 3D printing digital twin system client. Web3D technology can also be used to realize real-time display of the digital twin system on a webpage.

[0148] In summary, for the mechanical arm 3D printing scene, a digital twin model for real-time monitoring of the printing device is constructed to monitor the state of the printing device in real time, and a digital space module is introduced. Mechanism description models and data-driven models are constructed for the printing device, the working state is considered in real time, possible problems in the mechanical arm 3D printing work are predicted, timely alarms are given, defective parts and weak areas of the printed parts are detected, the mechanical arm 3D printing device performs immediate repair and reinforcement, error prediction and intelligent compensation algorithms are introduced to solve the precision problem encountered in the mechanical arm 3D printing process, superior robust deep learning algorithms are trained through dimensionality reduction feature extraction on the collected multi-dimensional data, and the printing precision of the mechanical arm 3D printing is optimized.

[0149] It should be noted that, in this document, the terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0150] The preferred embodiments of the application disclosed above are only used to help explain the application. The preferred embodiments do not describe all the details and do not limit the application to the specific embodiments described. Obviously, many modifications and changes can be made according to the content of the present application. The embodiments are selected and described in detail in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application. The application is limited only by the claims and their full scope and equivalents.

Claims

1. A digital-twin-based intelligent manufacturing method for a mechanical arm 3D printing, characterized in that: It comprises the following steps: S1, constructing a physical space module of the mechanical arm 3D printing system; S2, collecting data on the action, state and environment of the mechanical arm 3D printing system in the physical space, and establishing data transmission between the physical entity and the digital twin to form a data collection and transmission module; S3, constructing a digital space module of the mechanical arm 3D printing system, and completing high-precision mapping of the mechanical arm 3D printing device based on the digital twin; S4, constructing an algorithm training and verification module to compare and verify the real data collected in S2 and the results predicted by the algorithm model; The S4 constructing the digital space module specifically comprises the following steps: S41, analyzing and extracting data features from the data collected in S2 and transmitted to the state parameter database; S42, training a motion planning algorithm model and verifying it according to the numerical twin system; S43, training an intelligent decision algorithm model composed of a deep learning algorithm and verifying it; S44, training an error prediction and intelligent compensation algorithm model and verifying it, wherein the error prediction and intelligent compensation algorithm model uses a BP back propagation neural network; The error prediction and intelligent compensation algorithm model in S44 takes the end coordinates, posture and extrusion state of the mechanical arm 3D printing device as input, and the joint angles, speed control parameters of the mechanical arm as output, and fits, predicts and eliminates errors through the BP back propagation neural network; The loss function of the BP neural network is: ； wherein is the true value, is the model predicted value; The gradient of the BP neural network is: ； wherein, are weight values of a neural network layer, are input values, i.e. predicted values; ； wherein, are weight values of a neural network layer, is a learning rate for weight updates; S5, constructing a digital twin display and interaction module.

2. The digital-twin-based intelligent manufacturing method of a mechanical arm 3D printing according to claim 1, characterized in that: The S1 constructing the physical space module specifically comprises the following steps: S11, configuring a 3D printing extrusion device at the end of the mechanical arm; S12, building a camera positioning device to connect the mechanical arm coordinate system and the visual coordinate system through calibration equipment; S13, building a control system for the mechanical arm and the 3D printing device to control the parameters of the printing head at the end of the mechanical arm.

3. The digital-twin-based robot 3D printing intelligent manufacturing method according to claim 1, characterized in that: The data collection in S2 is realized by installing various sensors on the mechanical arm 3D printing system, and the collected data includes the joint angles, joint speeds and accelerations of the mechanical arm, and the speed, position, temperature and humidity of the printing head; The data transmission refers to transmitting the collected data to the digital space module, and the transmission methods include Bluetooth communication, wireless network communication and local area network communication.

4. The digital-twin-based intelligent manufacturing method of a mechanical arm 3D printing according to claim 1, characterized in that: The S3 constructing the digital space module specifically comprises the following steps: S31, constructing a three-dimensional visualization model of the mechanical arm 3D printing system; S32, constructing a mechanism model of the mechanical arm 3D printing system, which is a precise mathematical model established according to the internal mechanism of the object or production process; S33, constructing a data-driven model of the mechanical arm 3D printing system based on algorithms, which is a superior robust model formed by processing and analyzing the collected data to extract features based on the data.

5. The digital-twin-based intelligent manufacturing method of a mechanical arm 3D printing according to claim 4, characterized in that: The S32 constructing the mechanism model of the mechanical arm 3D printing system includes establishing the forward and inverse kinematics model of the mechanical arm 3D printing device through the D-H parameter table method.

6. The digital-twin-based intelligent manufacturing method of a mechanical arm 3D printing according to claim 4, characterized in that: The algorithm for constructing the data-driven model of the mechanical arm 3D printing system in S33 includes a BP neural network, an RBF neural network, a convolutional neural network, and a recurrent neural network algorithm.

7. The digital-twin-based intelligent manufacturing method of a mechanical arm 3D printing according to claim 1, characterized in that: The motion planning algorithm model in S42 adopts an RRT trajectory planning algorithm, and the principle of the RRT trajectory planning algorithm is as follows: First step, set initial node X init , target node X goal and state sampling space M; Second step, random sampling to get sampling points X rand ; Third step, if X rand No collision with obstacles, calculate X rand The distance from all nodes in the generated node set, get the nearest node X near From X near Walk to X rand with step, generate a new node X new If X new The line of X near collides with the obstacle, then resample randomly; Fourth step, repeat the cycle, generate random expansion tree, when the node in the tree reaches the target area, select a path from X int to X goal as the planning path, and plan a safe printing path for the mechanical arm 3D printing system.

Images