Spaceflight thermal control coating film magnetron sputtering coating manufacturing digital twin system building method

Abstract

The invention discloses a method for constructing a digital twin system for manufacturing a spaceflight thermal control coating film through magnetron sputtering coating. The method comprises the following steps of: firstly, constructing a machine tool three-dimensional model by utilizing Unity 3D and a three-dimensional reconstruction technology; a digital twinborn model is adopted to carry out mapping simulation on equipment operation actions and operation data acquired in real time, and visualization of the manufacturing process is achieved; according to the production requirements of magnetron sputtering coating equipment, an instruction prediction model is established for a specified coating product or process through a digital twin system, and online quality prediction of a real-time production process is realized; by establishing a channel-space attention mechanism, a mapping relation between coating defect strong correlation characteristics and defect categories is established, and magnetron sputtering coating system state early warning based on fusion of multi-source acquisition information is realized.

Description

Technical Field

[0001] The present application relates to the field of digital twin technology, and specifically to a method for building a digital twin system for manufacturing aerospace thermal control coating thin film magnetron sputtering coating. Background Art

[0002] With the advancement of Industry 4.0, the concept of digitalization has entered every enterprise in the industry. All enterprises are constantly trying to realize their own digitalization and how to develop and design more digital products.

[0003] At present, the hottest technology for the digital transformation of enterprises is digital twins. Recently, in the 2023 "Digital Twin City White Paper" released by the China Academy of Information and Communications Technology, the China Internet Association, and the China Communications Standards Association, digital twin academic research in 2023 will mainly focus on five major themes, namely, digital transformation of manufacturing, information-physical system research, digital twin AI intelligent development, digital twin mechanism model application, and virtual and augmented reality technology application. Among them, intelligent manufacturing, as an important component, has received widespread attention in the industry.

[0004] With the continuous increase in spacecraft model missions and the continuous acceleration of model development progress, spacecraft flexible thermal control films are in the transition stage of multi-variety mass production. The traditional production method is to use manual visual inspection + offline instrument detection after production. This method is inefficient and the detection effect is difficult to control. The most important thing is that it cannot be detected and regulated online in real time, which easily leads to defects but still continues production, resulting in production losses, waste of raw materials and energy, etc. This not only causes waste of resources and time, but also fails to guarantee product quality, and it is difficult to meet the requirements of high-quality mass production of products.

[0005] At present, most of the digital twin systems in engineering practice only focus on the state simulation of virtual space to physical space. There is a lack of digital twin systems with two-way data interaction, early warning and control functions between physical space and virtual space. This leads to low utilization rate of big data for manufacturing process monitoring and makes it difficult to achieve real-time and efficient control of complex manufacturing processes. Summary of the invention

[0006] In response to the defects existing in the above-mentioned prior art, the present invention provides a method for building a digital twin system for the manufacture of aerospace thermal control coating thin film magnetron sputtering coating, which focuses on the monitoring and early warning control of the magnetron sputtering coating process based on digital twins.

[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is:

[0008] A method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing, characterized in that it specifically includes the following steps:

[0009] S1: Use Unity3D software and 3D reconstruction technology to build a 3D model of the machine tool; use the digital twin model to map and simulate the equipment operation actions and operation data collected in real time to realize the visualization of the manufacturing process;

[0010] S2: Establish an instruction prediction model for the specified coating product or process through the digital twin system to achieve online quality prediction of the real-time production process. On-site visualization can be directly performed in the virtual environment, and mechanical movement, process simulation and electrical debugging can be performed on the digital twin model of each process. By embedding deep learning algorithms, the production process can be automatically monitored;

[0011] S3: By building a channel-spatial attention mechanism, the defect features extracted by the convolution layer are evaluated after the convolution layer, and the mapping relationship between the strongly correlated features of coating defects and the defect categories is established to reduce the calculation amount of redundant feature information and realize accurate identification of defect categories;

[0012] S4: Conduct technical research on the construction of a status warning mechanism for the magnetron sputtering coating system based on the fusion of multi-source information collection, build a deep learning defect detection model to realize real-time detection of coating defects, and embed it into the digital twin system to achieve production regulation through a classic optimal control algorithm LQR.

[0013] In the above technical solution, further, in S3, a channel-space attention mechanism is built, and the specific steps are:

[0014] S31: Compress the satellite surface coating defect sensitive feature channel through the Squeeze operation, compress the coating defect sensitive feature into a global feature, and input it as the satellite surface coating defect sensitive feature X H×W×M train The output is the compressed global feature F of the satellite surface coating defect 1×1×M sq .

[0015] X H×W×M train = {X1, X2, ..., X M} (Formula 1)

[0016] F i sq =GAP(X i train ) (Formula 2)

[0017]

[0018] Where, XH×W×M train ——Sensitive features of satellite surface coating defects after convolution operation, with M feature channels and a feature map size of H×W;

[0019] F i sq ——Global features of the i-th feature map through global average pooling;

[0020] GAP — global average pooling operation;

[0021] F 1×1×M sq ——Global features corresponding to M channels.

[0022] S32: Through the Excitation operation (FC1 layer + Relu activation + FC2 layer + Sigmoid activation), the fully connected layer FC1 reduces the dimension of the feature map vector along the channel dimension, the Relu activation function realizes nonlinearity, the fully connected layer FC2 restores the channel dimension of the feature vector to the original dimension, and the Sigmoid function converts the processed feature vector into a normalized weight vector S between 0 and 1 1×1×M , that is, to generate weight values ​​of different sensitive features. The use of two fully connected layers FC1 and FC2 has better nonlinear fitting ability, which can better fit the complex associations between channels, so that the important information in the sensitive features of satellite surface coating defects can receive extra attention. The mathematical expression is as follows:

[0023] FC1=W1×F 1×1×M sq (Formula 4)

[0024] FC2=W2×Relu(FC1) (Formula 5)

[0025] S 1×1×M =Sigmoid(FC2) (Formula 6)

[0026] in, M is the number of sensitive features of satellite surface coating defects, and r is the dimensionality reduction coefficient. 1×1×M is the feature vector corresponding to the sensitive features of satellite surface coating defects of M channels, Sigmoid is the activation function, and its mathematical expression is:

[0027]

[0028] S33: Through the Scale operation, the relevant expression is as follows, the normalized channel weight S 1×1×M Satellite surface coating defect sensitive characteristics X H×W×M trainFusion to generate new sensitive features This makes the model more discriminating about the sensitive features of satellite coating defects in different channels.

[0029]

[0030] Furthermore, the S4 specifically includes the following steps:

[0031] S41: inputting the image collected by the industrial camera into the coating defect detection model;

[0032] S42: Determine whether the coating has defects, and continue testing if there are no defects;

[0033] S43: If defects occur in the coating, the defect type is determined, and different control adjustment instructions are issued for different types through the optimal control algorithm LQR. The relevant formula is as follows:

[0034] x=Ax+u (Formula 9)

[0035] In fact, assuming that the above is a single variable equation, the eigenvalue of A is -K, then when x = -Kx, x will approach the steady state at an exponential rate. The above is a substable system, and the steady state point is 0, because x = e -kt , let, u = -K1x will be very effective, then the state equation becomes x = -(K + K1)x, so the steady-state response speed of the system can be adjusted. When u = b, the system steady-state point migrates from 0 to x = -Kx + b = 0 → x = b / K. Therefore, u = -Kx + b, the steady-state point can be changed while adjusting the response rate. b can be a PID value, b = K p e+K i ∫e+K d e can also be other functions of e designed by you to constrain the system to a specified value.

[0036] Furthermore, in S1, the three-dimensional model of the machine tool equipment is constructed by using Unity3D software. All models are derived from actual on-site equipment, and are consistent with the on-site equipment in appearance, so as to achieve a comprehensive analysis of the equipment. In addition, for the vacuuming process, different color particle effects are used to distinguish gases with different flow rates. During the magnetron sputtering process, the charged particles are represented by setting the tailing effect of the particle effects. The charged particles are emitted from the emission port, and the speed and direction of the particle animation movement are set to characterize the influence of the magnetic field and electric field on the charged particles in the manufacturing space. The target material is accelerated to collide with the target material through the action of the magnetic field and electric field. The material on the target is splashed out by the particle bombardment and attached to the flexible film to complete the coating.

[0037] Beneficial effects of the present invention:

[0038] The present invention combines digital twin technology with the production control of flexible thin film coating of spacecraft, focusing on the real-time detection and early warning control of defects in the actual production coating process. For the real-time detection process, the present invention adopts the channel-space attention mechanism to evaluate the value of defect features, establishes the mapping relationship between the strongly correlated features of coating defects and the defect categories, reduces the calculation amount of redundant feature information, and realizes the accurate identification of defect categories. The satellite surface coating defect sensitive feature channel is compressed by the Squeeze operation, and the coating defect sensitive feature is compressed into a global feature. The input is the satellite surface coating defect sensitive feature, and the output is the compressed satellite surface coating defect global feature. The two fully connected layers FC1 and FC2 have better nonlinear fitting ability, which can better fit the complex associations between channels, so that the important information in the satellite surface coating defect sensitive feature can get extra attention. Through the Scale operation, the normalized channel weights are merged with the satellite surface coating defect sensitive features of the original channel to generate new sensitive features, so that the model has more discriminative ability for satellite coating defect sensitive features of different channels. For the early warning and control process, when defects occur in the coating, the system determines the defect type, and then chooses to stop the machine for inspection or control and continue detection through the digital twin model according to different types of defects. The monitoring and prediction of the magnetron sputtering coating process based on digital twin can improve the efficiency and stability of the process, reduce production costs, improve the quality of thin film production, and provide strong support for process optimization and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is implemented as a flow chart of the overall solution. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0041] This embodiment discloses a method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing, which is mainly used for spacecraft coating processing equipment. In this solution, it is necessary to collect the appearance and status of the main products in the internal operation process of the coating equipment to build a three-dimensional model, and it is necessary to choose to stop the inspection or control and continue the inspection through the digital twin model according to the different types of defects detected in real time by defect recognition.

[0042] The overall solution implementation flow chart is as follows: Figure 1 As shown, refer to Figure 1 , a method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing, comprising:

[0043] Using computer-aided design software or 3D modeling tools, we began to build a 3D model of aerospace thermal control coating thin film magnetron sputtering coating equipment based on the collected information. The key parts of the model include chamber base, beams, gears, targets, coating spindles, rollers, vacuum systems, condensing devices, mobile devices, power distribution control equipment, equipment appearance, etc., a total of 87 parts, and then imported the model into Unity3D software.

[0044] A digital twin system interface for aerospace thermal control coating thin film magnetron sputtering coating manufacturing. The first thing created is the system homepage. The left side is the system navigation bar. By clicking different buttons created in the navigation bar, you can enter the corresponding module of the system. The current status of the coating equipment is displayed below the navigation bar. The right side of the system homepage shows the recognition accuracy of different coating defects in the production process identified and counted by the system, the proportion of each defect, the specific time and distribution location of the defect. The overall situation of each defect in the production process can be observed through the homepage.

[0045] Click the Twin button in the navigation bar to enter the Twin module. This module has 6 monitoring windows for the production process, where Camera 1 and Camera 2 are real-time production images of the coating equipment captured by the camera. Camera 1 captures the image of the thin film product after coating, and Camera 2 captures the main view of the coating equipment, which are used to monitor the overall production environment and conditions. The remaining four windows correspond to the digital twin model of the coating equipment, where the upper window is the main view of the coating equipment, and the lower three windows are partial views of the interior of the production coating equipment, which are used to monitor the internal conditions of the coating equipment during the production process and whether each production link is operating normally.

[0046] After clicking the twin button in the navigation bar, three secondary menu buttons will appear below the twin button, corresponding to the production links of the thin film, including feeding, vacuum and magnetron sputtering stages.

[0047] Click the feed button to enter the feed link. The left side of the feed interface is divided into two parts: processing status and material information. The right side is the loading view, pretreatment view and main view of the coating equipment. There are 4 states in production pretreatment, namely idle, loading, closed processing chamber and completed. These states correspond to the pretreatment state of the coating equipment in actual production and correspond to the state of the virtual coating equipment in the right view. Material information is divided into film-related information, target material-related information and coating equipment-related information. These information correspond to the actual production information. To the right of the coating equipment-related information are the start machine tool and auxiliary device button and the start loading button. At the beginning, the coating equipment is in an idle state. When the start loading button is clicked, the progress bar below the processing status begins to load, and the coating equipment enters the stage of loading and closing the processing chamber in turn. The loading effect can be seen from the front window of loading. After the loading process is completed, the processing pretreatment enters the stage of closing the processing chamber. At this time, the coating equipment can be clearly seen from the main view moving to the processing chamber on the right. After the above two stages are completed, the progress bar reaches 100%, and the processing status is completed.

[0048] Click the vacuum processing button in the navigation bar to enter the vacuuming phase. On the right side of the interface are three monitoring windows of the virtual coating equipment. A table is created in the process parameter section to show the current, voltage, pressure and vacuum parameters set for the coating equipment. The magnetron sputtering coating parameters are three curves of vacuum, pressure and magnetic field strength. Click the start vacuuming button, the progress bar in the processing state starts to load, and the coating state enters the equipment preparation stage. At this time, the equipment is fed with cooling water and cold temperature medium to adjust the temperature. This process can be seen from the right view. When the temperature reaches the appropriate range, enter the vacuuming phase. When the vacuum degree of the vacuum chamber of the coating equipment reaches the standard, the progress bar is loaded and the vacuuming phase ends.

[0049] Click the magnetron sputtering button in the navigation bar to enter the magnetron sputtering link, as shown in the figure above. Similar to the vacuum link interface, the left side of the magnetron sputtering production interface is divided into three parts: processing status, process parameters and vacuum chamber parameters, and the right side is the coating equipment monitoring window. The processing status is divided into three stages: idle, coating pretreatment, and magnetron sputtering coating. The process parameters show the current, voltage, coating speed, vacuum degree and coating length parameters set by the coating equipment. The parameter part is the same as the vacuum link. Click the start magnetron sputtering button, the processing status switches from idle to coating pretreatment, and the progress bar starts to load in the processing status. After the coating pretreatment is completed, the processing status enters the magnetron sputtering coating stage. At this time, it can be seen from the monitoring window that the charged particles are emitted upward from the emission port, and are accelerated to collide with the target material through the action of the magnetic field and electric field. The material on the target is splashed out by the particle bombardment and attached to the substrate to complete the coating. During the coating process, the system's built-in defect detection algorithm will begin to detect defects in the coated film. If defects are detected, the system will issue an alarm prompt to instruct the staff to adjust the production process. After all flexible thermal control films are coated, the progress bar reaches 100% and the loading is completed.

[0050] Click the real-time detection button in the navigation bar, and two modules, industrial camera and real-time monitoring, will appear. Click the industrial camera button, and the system will jump to the system interface of the industrial camera. Through this page, the parameters and operating status of the production process can be monitored, and the parameters can be reversely controlled and adjusted.

[0051] Click the real-time monitoring button in the navigation bar to enter the defect recognition and detection module. The defect detection interface is mainly divided into 4 parts. The first part is the classification statistics of this defect recognition. This part displays the defects generated in the production process in the form of a pie chart, and clearly sees the proportion of each defect type in the production process. The second part is the classification statistics of this defect recognition. This part shows a waterfall chart. The horizontal axis is the time of this detection, and the vertical axis is the number of occurrences of the corresponding defect type in this detection. From this chart, we can know the corresponding relationship between the time and the number of occurrences of all defect types in this production process. The third part is the corresponding third part-related information on this defect recognition. During the operation of the system, the defect pictures taken by the industrial camera will be displayed in this part, and the defect details after identification by the defect recognition algorithm will also be displayed at the same time. Behind it, there is also the type of the defect and the coordinate information of the defect, which is convenient for the staff to find the location of the defect in the film during the production process. When clicking on the defect picture in the third part, the fourth part will show the detailed information of the corresponding defect, including the detailed enlarged picture of the defect, the type of defect, the time of generation, the location, the name of the defect picture and the storage path.

[0052] Click the Start Detection button in the lower right corner of the interface to start the inspection of this production. The four areas in the interface will update data as the inspection results appear, allowing real-time observation of defect information during the production process.

[0053] Click the Historical Database button to enter the Historical Data module. This page displays a coating defect information table. From this module, you can see the causes of various defect types obtained from database statistics, the number of defects generated, and the recognition rate of various defects by the thermal control coating thin film magnetron sputtering coating manufacturing digital twin system. After completing a coating production, this system will store the product defect information in the specified path and add it to the defect information database.

[0054] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing, characterized in that: The specific steps include: S1: Use Unity3D software and 3D reconstruction technology to build a 3D model of the machine tool; use the digital twin model to map and simulate the equipment operation actions and operation data collected in real time to realize the visualization of the manufacturing process; S2: Establish an instruction prediction model for the specified coating product or process through the digital twin system to realize the online quality prediction of the real-time production process. It can directly visualize the site in the virtual environment, perform mechanical movement, process simulation and electrical debugging on the digital twin model of each process, and automatically monitor the production process by embedding deep learning algorithms; S3: By building a channel-spatial attention mechanism, the defect features extracted by the convolution layer are evaluated after the convolution layer, and the mapping relationship between the strongly correlated features of coating defects and the defect categories is established to reduce the calculation amount of redundant feature information and realize accurate identification of defect categories; S4: Conduct technical research on the construction of a status warning mechanism for the magnetron sputtering coating system based on the fusion of multi-source information collection, build a deep learning defect detection model to realize real-time detection of coating defects, and embed it into the digital twin system to achieve production regulation through a classic optimal control algorithm LQR.

2. According to the method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing according to claim 1, in said S3, a channel-space attention mechanism is built, and the specific steps are: S31: Compress the satellite surface coating defect sensitive feature channel through the Squeeze operation, compress the coating defect sensitive feature into a global feature, and input it as the satellite surface coating defect sensitive feature X H×W×M train The output is the compressed global feature F of the satellite surface coating defect 1×1×M sq ; X H×W×M train = {X1, X2, ..., X M } (Formula 1) Where, X H×W×M train ——Sensitive features of satellite surface coating defects after convolution operation, with M feature channels and a feature map size of H×W; F i sq ——Global features of the i-th feature map through global average pooling; GAP — global average pooling operation; F 1×1×M sq ——Global features corresponding to M channels. S32: Through the Excitation operation (FC1 layer + Relu activation + FC2 layer + Sigmoid activation), the fully connected layer FC1 reduces the dimension of the feature map vector along the channel dimension, the Relu activation function realizes nonlinearity, the fully connected layer FC2 restores the channel dimension of the feature vector to the original dimension, and the Sigmoid function converts the processed feature vector into a normalized weight vector S between 0 and 1 1×1×M , that is, to generate weight values ​​of different sensitive features, and use two fully connected layers FC1 and FC2 to fit the complex associations between channels, so that the important information in the sensitive features of satellite surface coating defects can receive extra attention. Its mathematical expression is as follows: FC1=W1×F 1×1×M sq (Formula 4) FC2 = W2 × Relu (FC1) (Formula 5) S 1×1×M =Sigmoid(FC2)(Formula 6) in, M is the number of sensitive features of satellite surface coating defects, r is the dimensionality reduction coefficient, S 1×1×M is the feature vector corresponding to the sensitive features of satellite surface coating defects of M channels, Sigmoid is the activation function, and its mathematical expression is: S33: Through the Scale operation, the relevant expression is as follows, the normalized channel weight S 1×1×M Satellite surface coating defect sensitive characteristics X H×W×M train Fusion to generate new sensitive features This makes the model more discriminating about the sensitive features of satellite coating defects in different channels.

3. According to the method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing according to claim 1, in said S4, production control is realized by a classic optimal control algorithm LQR, and the specific steps are: S41: inputting the image collected by the industrial camera into the coating defect detection model; S42: Determine whether the coating has defects, and continue testing if there are no defects; S43: If defects occur in the coating, the defect type is determined and different control adjustment instructions are issued for different types through a classic optimal control algorithm LQR. The relevant formula is as follows: x=Ax+u (Formula 9) Assuming that the above is a single variable equation, the eigenvalue of A is -K, then when x = -Kx, x approaches the steady state at an exponential rate. The above is a substable system, and the steady state point is 0, because x = e -kt , let u=-K1x, then the state equation becomes x=-(K+K1)x, which can adjust the steady-state response speed of the system. When u=b, the system steady-state point migrates from 0 to x=-Kx+b=0→x=b / K, so u=-Kx+b. While adjusting the response rate, the steady-state point is changed. b can be the PID value, b=K p e+K i ∫e+K d e can also be other functions of e designed by yourself, so as to constrain the system to a specified value.

4. According to the method for building a digital twin system for aerospace thermal control coating thin film magnetron sputtering coating manufacturing according to claim 1, in S1, building a three-dimensional model of the machine tool is specifically completed by using Unity3D software, and all models are derived from actual on-site equipment, and are consistent with the on-site equipment in appearance, so as to achieve a comprehensive analysis of the equipment; during the vacuuming process, different color particle effects are used to distinguish gases with different flow rates. During the magnetron sputtering process, the tailing effect of the particle effects is set to represent the charged particles. The charged particles are emitted from the emission port, and the speed and direction of the particle animation movement are set to characterize the charged particles being affected by the magnetic field and electric field in the manufacturing space, and being accelerated to collide with the target material through the action of the magnetic field and electric field. The material on the target is splashed out after the particle bombardment and attached to the flexible film to complete the coating.

Images