A transformer-based identification method for anti-interference satellite flexible attachment coupling torque identifier

Abstract

The present invention proposes a transformer-based identification method for an anti-interference satellite-based flexible attachment coupling torque identifier, comprising the following steps: Step S1: using a depth camera to obtain a depth map of a flexible sailboard, and using image recognition technology to obtain the vibration displacement of the flexible attachment identification point; Step S2: judging whether all identification points are identified in step S1, if all are identified, then entering step S3, if not all are identified, then using a displacement predictor to complete the displacement of the sailboard identification point; Step S3: inputting the vibration displacement of the four identification points on the flexible sailboard that are identified or predicted into the coupling torque identifier to identify the vibration coupling torque generated by the current flexible sailboard. The present invention can intelligently and real-timely obtain the coupling interference torque of the flexible attachment sailboard through computer vision processing technology and the powerful nonlinear fitting ability of neural networks, and provide assistance for the attitude control system of spacecraft. It can be applied to various industrial scenarios to help enterprises work and produce better and safer.

Description

Technical Field

[0001] The invention belongs to the technical field of spacecraft intelligent control, and in particular relates to an identification method of a transformer-based anti-interference on-board flexible attachment coupling torque identifier. Background Art

[0002] With the continuous development of science and technology, most of the spacecraft currently in orbit carry large flexible accessories, such as solar panels or antennas. Large flexible accessories will vibrate during the attitude maneuvers of the spacecraft, and the coupling torque generated will have a huge impact on the spacecraft control system and even cause the spacecraft mission to fail.

[0003] However, there is no suitable solution to this problem in the current spacecraft attitude control algorithm. Summary of the invention

[0004] The problem solved by the present invention is: to overcome the deficiencies of the prior art and to provide an identification method for an anti-interference onboard flexible attachment coupling torque identifier based on a transformer, which can identify the coupling interference torque generated by the flutter of a large flexible attachment carried by a spacecraft.

[0005] The object of the present invention is achieved through the following technical solution: a transformer-based identification method for an anti-interference on-board flexible attachment coupling torque identifier, the method comprising the following steps:

[0006] Step S1: using a depth camera to obtain a depth map of the flexible sailboard, and using image recognition technology to obtain the vibration displacement of the flexible attachment identification point;

[0007] Step S2: Determine whether all the identification points are recognized in step S1. If all are recognized, proceed to step S3. If not all are recognized, use the displacement predictor to complete the displacement of the windsurfing board identification points.

[0008] Step S3: inputting the identified or predicted vibration displacements of the four marking points on the flexible sailboard into a coupling torque identifier to identify the vibration coupling torque generated by the current flexible sailboard.

[0009] In the above-mentioned transformer-based identification method of the anti-interference on-board flexible attachment coupling torque identifier, in step S1, the sailboard is first identified, and black dots are marked on the surface of the sailboard as identification points. Then, a depth camera is used to obtain a depth map of the front sailboard. The dot detection method is used to detect the image coordinates of the first four identification points of the central axis position in the depth image, and then the vibration displacement of the four identification points selected at each moment is obtained. Since the sailboard is vibrating, the vibration displacement is changing at each moment.

[0010] In the above-mentioned transformer-based identification method of the anti-interference on-board flexible attachment coupling torque identifier, in step S2, since the positions of all the identification points may not be identified due to external interference during the operation of the identifier, resulting in the inability of the torque identifier to work, a displacement predictor based on the transformer network is designed. The vibration displacement of the four identification points in the first four time series and the incomplete identification point displacement information identified in the current time series are used as input, and the displacement predictor outputs the displacement information of the missing points. The transformer-based displacement predictor works as follows: First, the vibration displacement data of the four identification points in the first four time series are organized into a 1×4×4 displacement array matrix Ms, and the displacement data of the identification points of the incomplete current time series are organized into a 1×4×1 auxiliary array matrix Ma. The displacement array matrix Ms and the auxiliary array matrix Ma are input into the data processing network. The data processing network has two fully connected layers L1 and L2. The number of input neurons of L1 is 4 and the number of output neurons is 512; the number of input neurons of L2 is 1 and the number of output neurons is 512. The displacement array matrix Ms and the auxiliary array matrix Ma are respectively input into the data processing network. The group matrix Ma is input into L1 and L2, and the displacement array matrix Ms and the auxiliary array matrix Ma are upgraded to 1×4×512 encoder input matrix and 1×4×512 decoder input matrix. The encoder input matrix and the decoder input matrix are input into the encoder and decoder respectively. The encoder is composed of an 8-head self-attention mechanism network and an 8-head attention mechanism network. After 3 layers of decoding and encoding, the output of the decoder is sent to the data integrator. The data integrator consists of a fully connected layer with 512 input neurons and 4 output neurons respectively. The output completes the displacement data of all identification points of the current time series.

[0011] In the above transformer-based identification method of the anti-interference satellite flexible attachment coupling torque identifier, in step S3, after obtaining the complete identification point vibration displacement data in step S2, the complete identification point vibration displacement data is input into the coupling torque identifier, and the coupling torque identifier outputs the coupling vibration torque of the flexible sailboard in the current time series. The coupling torque identifier consists of a network with two fully connected layers. The number of input neurons and the number of output neurons of the first layer of the neural network are 512 and 256 respectively, and the number of input neurons and the number of output neurons of the second layer of the neural network are 256 and 1 respectively.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] Through computer vision processing technology and the powerful nonlinear fitting ability of neural networks, it is possible to intelligently and real-time obtain the coupling interference torque of the flexible accessory sailboard, providing assistance for the attitude control system of the spacecraft.

[0014] It can be applied in various industrial scenarios to help companies work and produce better and safer. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the overall framework diagram of the network.

[0016] Figure 2 This is a schematic diagram of the ground experiment.

[0017] Figure 3 It is a schematic diagram of dot detection.

[0018] Figure 4 It is the comparison between the identification result of the coupling torque identifier and the actual coupling torque. DETAILED DESCRIPTION

[0019] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to be able to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0020] Step S1: Use a depth camera to obtain a depth map of the flexible sailboard, and use image recognition technology to obtain the vibration displacement of the flexible attachment identification point. First, mark the sailboard, and mark black dots on the surface of the sailboard as identification points. In this ground experiment, a total of 15 identification points were marked on the sailboard, which were evenly distributed at the center axis and upper edge of the sailboard, covering the entire sailboard. Considering that the data information of the upper edge marking points overlaps, and in order to reduce the amount of network calculation, only the vibration displacement of the first four marking points at the center axis is selected as input. Then use the depth camera to obtain the depth map of the front sailboard, and use the dot detection method to detect the image coordinates of the first four identification points at the center axis in the depth image, and then obtain the vibration displacement of the four identification points selected at each moment.

[0021] Step S2: Determine whether the displacement of all the marking points is identified in step S1. If so, proceed to step S3. If not all are identified, use the displacement predictor to complete the displacement of the windsurfing marking points. Specifically, a displacement predictor based on a transformer network is designed. The vibration displacement of the four marking points in the first four time series and the incomplete displacement information of the marking points identified in the current time series are used as inputs. The displacement predictor outputs the displacement information of the missing points. The transformer-based displacement predictor works as follows: first, the vibration displacement data of the four identification points in the first four time series are organized into a 1×4×4 displacement array matrix Ms, and the displacement data of the identification points of the incomplete current time series are organized into a 1×4×1 auxiliary array matrix Ma. The displacement array matrix Ms and the auxiliary array matrix Ma are input into the data processing network. The data processing network has two fully connected layers, which respectively upgrade the displacement array matrix Ms and the auxiliary array matrix Ma to a 1×4×512 encoder input matrix and a 1×4×512 decoder input matrix. The encoder input matrix and the decoder input matrix are respectively input into the encoder and decoder. The encoder and decoder are respectively composed of an 8-head self-attention mechanism network and an attention mechanism network. After 3 layers of decoding and encoding, the output of the decoder is sent to the data integrator. The data integrator consists of a fully connected layer and outputs the completed displacement data of all identification points of the current time series.

[0022] Step S3: Input the vibration displacement of the four identified or predicted points on the flexible sailboard into the coupling torque identifier to identify the vibration coupling torque generated by the current flexible sailboard. Specifically, after obtaining the complete vibration displacement data of the identification points in step S2, the complete vibration displacement data of the identification points is input into the coupling torque identifier, and the coupling torque identifier outputs the coupling vibration torque of the current time series flexible sailboard. The coupling torque identifier is composed of a network of three fully connected layers. The output is the coupling torque generated by the current flexible sailboard.

[0023] Example 1:

[0024] The flexible sailboard whose flutter moment needs to be measured is marked with dots, the sailboard is photographed using a depth camera, and the position of the dot marks in the photo is identified using image recognition technology, such as Figure 2 In this example, considering that the information of the two rows of marking points on the upper edge overlaps, and the network calculation amount will increase as the number of marking points increases, only the vibration displacement of the first four marking points at the center axis position is selected as the network input, as shown in Figure 3 shown.

[0025] In the experiment, a force of 10 Newtons was first applied to the upper end of the sailboard to make it vibrate, and then the sailboard was photographed with a camera, and all the dots on the flexible sailboard were identified. After obtaining the positions of all the dots in the photo, the first four marking points of the central axis were selected, and the vibration displacement of the four dot marks at each moment was obtained using the depth photo. When the vibration displacement of all the marked points cannot be obtained in the photo due to interference from the external environment, the displacement predictor was used to predict the vibration displacement of the missing marked points. After obtaining the vibration displacement of all the required marked points, the trained moment identifier was used to identify the vibration coupling torque at that moment, and then the vibration coupling torque of the flexible sailboard at each moment was obtained. In this example, at the 5th moment, the 55th moment, the 105th moment, and the 155th moment, the coordinates of the four marking points on the sailboard in the Z direction of the camera coordinate system are, from bottom to top: [19.533, 20.100, 20.700, 22.000] mm, [19.000, 19.700, 20.200, 21.900] mm, [19.800, 20.500, 21.000, 22.200] mm, [20.400, 20.800, 21.300, 22.200] mm;

[0026] At the 5th moment, the 55th moment, the 105th moment, and the 155th moment, the actual coupling torque of the windsurfing board is: [-0.0272] Nm, [-0.0416] Nm, [-0.0148] Nm, [-0.0013] Nm;

[0027] At the 5th moment, the 55th moment, the 105th moment, and the 155th moment, the identification results of the coupling torque identifier are: [-0.0241] Nm, [-0.0450] Nm, [-0.0151] Nm, [-0.0080] Nm;

[0028] The comparison between the coupling torque identification results of the trained coupling torque identifier and the real coupling torque is shown in Figure 4 The dotted line in the figure represents the actual vibration coupling torque of the sailboard at 200 sampling moments in this experiment, and the solid line with asterisks represents the coupling torque identified by the coupling torque identifier based on the information collected by the camera at 200 sampling moments. The two are basically consistent.

Claims

1. A transformer-based identification method for an anti-interference satellite onboard flexible attachment coupling torque identifier, characterized in that: The following steps are involved: Step S1: using a depth camera to obtain a depth map of the flexible attachment, and using image recognition technology to obtain a vibration displacement of a marking point of the flexible attachment; Step S2: Determine whether all the identification points are recognized in step S1. If all are recognized, proceed to step S3. If not all are recognized, use the transformer-based displacement predictor to complete the vibration displacement of the flexible attachment identification points. Step S3: inputting the identified or predicted vibration displacements of the four marking points on the flexible attachment into a coupling torque identifier to identify the vibration coupling torque generated by the current flexible attachment; In step S1, the flexible attachment is first identified by marking black dots on the surface of the flexible attachment as identification points, and then a depth camera is used to obtain a depth map of the front flexible attachment. The dot detection method is used to detect the image coordinates of the first four identification points of the central axis position in the depth map, and then the vibration displacement of the four identification points selected at each moment is obtained. Since the flexible attachment is vibrating, the vibration displacement is changing at each moment, wherein the flexible attachment is a flexible sailboard.

2. According to claim 1, a transformer-based identification method for an anti-interference satellite onboard flexible attachment coupling torque identifier is characterized by: In step S2, since the coupling torque identifier cannot identify the positions of all the identification points due to external interference, causing the coupling torque identifier to fail to work, a transformer-based displacement predictor is designed. The vibration displacements of the four identification points in the first four time series and the incomplete identification point displacement information identified in the current time series are used as inputs, and the displacement predictor outputs the displacement information of the missing points.

3. The identification method of the transformer-based anti-interference satellite onboard flexible attachment coupling torque identifier according to claim 2 is characterized in that: The transformer-based displacement predictor works as follows: First, the vibration displacement data of the four identification points in the first four time series are organized into a 1×4×4 displacement array matrix Ms, and the displacement data of the identification points of the incomplete current time series are organized into a 1×4×1 auxiliary array matrix Ma. The displacement array matrix Ms and the auxiliary array matrix Ma are input into the data processing network. The data processing network has two fully connected layers L1 and L2. The number of input neurons of L1 is 4 and the number of output neurons is 512; the number of input neurons of L2 is 1 and the number of output neurons is 512. The displacement array matrix Ms and the auxiliary array matrix Ma are respectively input into the data processing network. The auxiliary array matrix Ma is input into L1 and L2, and the displacement array matrix Ms and the auxiliary array matrix Ma are upgraded to 1×4×512 encoder input matrix and 1×4×512 decoder input matrix, and the encoder input matrix and the decoder input matrix are input into the encoder and decoder respectively. The encoder and decoder are composed of an 8-head self-attention mechanism network and an attention mechanism network respectively. After 3 layers of decoding and encoding, the output of the decoder is sent to the data integrator. The data integrator consists of a fully connected layer with 512 input neurons and 4 output neurons respectively. The output completes the displacement data of all identification points of the current time series.

4. The identification method of the transformer-based anti-interference satellite onboard flexible attachment coupling torque identifier according to claim 1 is characterized in that: In step S3, after the complete vibration displacement data of the identification point in step S2 is obtained, the complete vibration displacement data of the identification point is input into the coupling torque identifier, and the coupling torque identifier outputs the coupling vibration torque of the flexible attachment in the current time series; The coupled torque identifier consists of a network with two fully connected layers.

5. The identification method of the transformer-based anti-interference satellite onboard flexible attachment coupling torque identifier according to claim 4 is characterized in that: In the network with two fully connected layers, the number of input neurons and the number of output neurons of the first layer of the neural network are 512 and 256 respectively, and the number of input neurons and the number of output neurons of the second layer of the neural network are 256 and 1 respectively.

Images