Batch production type satellite assembly intelligent detection platform and development and application method thereof

Abstract

The invention provides a batch production type satellite assembly intelligent detection platform and a development and application method thereof, and relates to the technical field of satellite assembly, and the batch production type satellite assembly intelligent detection platform comprises a development support module, a hardware construction module, a data processing module, a model training optimization module, a client development and application module and a process data support module. The modules are used for production line application analysis and function parameter definition, hardware construction of an adaptive scene, sample data set acquisition and tagging processing, deep learning-based target detection algorithm training optimization, client software development and model integration, process data storage and support. And intelligent detection and data management of satellite assembly quality are cooperatively realized.

Description

Technical Field

[0001] This invention relates to the field of satellite assembly technology, specifically to a mass production intelligent inspection platform for satellite assembly and its development and application methods. Background Technology

[0002] Cables are crucial components in spacecraft systems for transmitting current and signals. During the final assembly of spacecraft, cable assembly constitutes a significant portion of the process, and its quality directly impacts the spacecraft's operational performance and whether its overall performance indicators meet design and operational requirements. To match the assembly speed of mass-produced satellites, an intelligent assembly inspection platform has been established on the final assembly line to achieve real-time online inspection of the entire satellite. Traditional manual inspection relies on workers' visual judgment. Factors such as the experience level of inspectors, fatigue from prolonged inspections, and differences in management methods can lead to errors such as missing or incorrect cable connectors, negatively impacting the entire production line and significantly reducing efficiency and production effectiveness.

[0003] Artificial intelligence-based target detection technology can effectively avoid the drawbacks of traditional manual inspection, ensuring consistent inspection and assembly speeds, guaranteeing production line continuity, significantly improving final assembly efficiency, and reducing rework rates. Machine vision technology can replace or even surpass the human eye in acquiring image information, and corresponding computer programs can be written to replace the human brain. By extracting features and processing the acquired image information, it can then complete a series of operations such as product identification, inspection, and measurement.

[0004] In recent years, object detection technology has received significant attention in both academia and industry. Early research was mainly based on traditional image processing techniques and feature engineering, relying on manually extracted image features for object recognition and localization. This method has limited adaptability to complex environments and variable conditions, and is highly dependent on domain expert knowledge, making it difficult to generalize.

[0005] In summary, the shortcomings of existing technologies are as follows: 1) Traditional manual inspection relies on experience and is prone to errors due to fatigue, leading to missing or incorrect cable plugs. It is also inefficient and cannot match the assembly speed of mass-produced satellites, affecting production line continuity and production efficiency. 2) Traditional inspection lacks standardized processes and systematic data management, and there is no unified data storage and traceability system, making it difficult to ensure the integrity of inspection data and to trace and analyze the source of assembly quality problems. 3) Manual inspection is limited by human visual perception, making it difficult to cover inspection scenarios such as foreign objects on the satellite and cable bending radii. It cannot accurately complete quantitative inspections and cannot meet the accuracy requirements of mass-produced satellite assembly. 4) The entire satellite is difficult to transport and the workstation is uncertain. Traditional fixed inspection equipment cannot be flexibly adapted, making it difficult to carry out inspections efficiently and restricting the progress of mass-produced satellite assembly. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a batch-production intelligent inspection platform for satellite assembly and its development and application methods.

[0007] According to the present invention, a batch-production intelligent inspection platform for satellite assembly and its development and application method are provided, the solution of which is as follows:

[0008] In the first aspect, a mass production satellite assembly intelligent inspection platform is provided, the inspection platform including: a development support module, a hardware construction module, a data processing module, a model training and optimization module, a process data support module, and a client development and application module;

[0009] The development support module is used to conduct production line application and feasibility analysis for mass production satellite assembly and testing, and to define the core functions and algorithm parameters of the intelligent testing platform.

[0010] The hardware building module is used to build a hardware platform adapted to satellite assembly and testing scenarios. The hardware platform includes at least image acquisition equipment and computing equipment.

[0011] The data processing module is used to collect sample datasets required for satellite assembly and testing, and to label the sample datasets.

[0012] The model training optimization module is used to select a deep learning-based object detection algorithm, train the model using the labeled sample dataset, and optimize the model based on the training effect.

[0013] The process data support module is used to store the process data required for satellite assembly and inspection. The process data includes at least satellite material information, assembly operation instructions and assembly standard parameters, and provides data support for satellite assembly quality inspection.

[0014] The client development and application module is used to develop client software and integrate the optimized target detection model into the client software to realize intelligent detection and data management of satellite assembly quality.

[0015] Preferably, the hardware platform further includes a display for showing the test results, and the image acquisition device and the computing device are both communicatively connected to the display;

[0016] The hardware platform is also equipped with lighting equipment. When in use, the position of the image acquisition equipment and the lighting equipment relative to the workbench is adjusted to ensure that the acquired assembly images meet the clarity requirements, making it suitable for satellite assembly and inspection scenarios where the entire satellite is difficult to transport and the work position is uncertain.

[0017] Preferably, the deep learning-based target detection algorithm selected by the model training optimization module is the YOLOv5 algorithm;

[0018] The YOLOv5 algorithm divides the input satellite assembly image into m×n cells. Each cell outputs B bounding boxes and N predicted object category probability information. Each bounding box contains one confidence score and four coordinate position information. The confidence score is used to characterize the matching degree between the predicted box and the ground truth box and the probability of the presence of a target object within the box. By adjusting the parameters of the algorithm's inference speed and detection accuracy, the optimized model's detection rate is matched with the satellite assembly speed.

[0019] Preferably, the client software developed by the client development and application module includes: a data management submodule;

[0020] The data management submodule adopts a three-level folder structure of satellite-subsystem-unit, supports the creation of custom satellite model folders and the entry of basic satellite information, and supports batch uploading of unit assembly images to the corresponding folders; uploaded images are automatically stored in the system database, and support quick retrieval by satellite name and unit name, realizing the classified storage and traceability of detection data.

[0021] Preferably, the client software further includes a quality inspection submodule;

[0022] The quality inspection submodule supports the creation of inspection tasks. The inspection tasks require the input of satellite name and satellite type information, and can be queried by satellite name, type, and number.

[0023] The quality inspection submodule also standardizes the image acquisition of individual sub-components. For cable sub-components, images of plugs, interfaces, uninserted states, and connected states need to be acquired. After the image acquisition is completed, the target detection model is triggered to perform assembly quality inspection and generate the single-unit pass rate and the system judgment conclusion of pass / fail.

[0024] Preferably, the quality inspection submodule also supports a manual review function, which is only available to administrators.

[0025] Based on the detection results of the target detection model, the administrator performs the approval or re-detection operation. If there are abnormalities in the detection results, the system will report the error and indicate the reason for the abnormality, guiding the correct assembly and re-detection. After approval, the system automatically generates a Word format detection report containing basic satellite information, detection overview statistics, and abnormal details. The detection report supports export and archiving.

[0026] Preferably, the process data support module can also be extended to interface with external CAD systems for importing satellite assembly models;

[0027] The satellite assembly quality inspection includes at least one or more of the following: incorrect or missing connector installation, incorrect or missing fastener installation, detection of extraneous objects on the satellite, detection of clamp pasting position, detection of cable bending radius, identification of unit and connector number and corresponding relationship identification, covering the quality control needs of the entire satellite assembly scenario.

[0028] Secondly, a method for developing and applying a batch-production intelligent inspection platform for satellite assembly is provided, the system comprising:

[0029] Step S1: Conduct production line application and feasibility analysis for mass production satellite assembly and testing, define the core functions and algorithm parameters of the intelligent testing platform, and complete the development support work;

[0030] Step S2: Build the hardware platform, adjust the relative positions of the devices to ensure image acquisition clarity, and complete the hardware construction;

[0031] Step S3: Collect the sample dataset required for satellite assembly and inspection, and label the sample dataset with target locations and categories to complete the data processing;

[0032] Step S4: Select the YOLOv5 algorithm, use the labeled sample dataset for model training, and optimize the algorithm's inference speed and detection accuracy by adjusting parameters to make the model's detection beat match the satellite assembly speed, thus completing the model training optimization.

[0033] Step S5: Develop client software, integrate the optimized YOLOv5 model into the software, and realize the creation of detection tasks, image uploading, AI detection, manual review and report export through the client software, thus completing the client development and application.

[0034] Preferably, the specific process of satellite assembly quality inspection via client software in step 5 includes:

[0035] Step S51: The user logs into the client software, selects the detection task corresponding to the satellite model, and enters the standalone image acquisition and import interface;

[0036] Step S52: Upload the single-unit assembly image to be inspected. The client software backend filters whether the image meets the acquisition specifications. If it does not meet the specifications, return to step S51 and repeat the image acquisition and upload operation.

[0037] Step S53: Call the YOLOv5 model to detect images that meet the requirements. If there are any abnormalities in the detection results, the system will report an error and indicate the reason for the error. After correction and assembly, return to step S52 to re-upload.

[0038] Step S54: After the test is passed, the administrator will perform a manual review. If the review is successful, a Word format test report will be generated. If the review fails, return to step S53 for retesting.

[0039] Preferably, in step S5, the client software adopts a three-level data management architecture of satellite-subsystem-standby, supporting batch image uploading and name-based retrieval; through a dual-stage judgment mechanism of AI automatic detection and administrator manual review, combined with the exported standardized inspection report, satellite assembly inspection is achieved.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. This invention solves the problem of low efficiency in traditional manual inspection, making the inspection and batch production satellite assembly speed consistent, ensuring the continuity of the production line, improving acceptance efficiency by 80%, and reducing rework rate (rework rate reduced by 98%). It also reduces the adverse effects of misoperation such as missing or incorrect assembly on production, thereby improving production efficiency.

[0042] 2. This invention simplifies the traditional inspection mode that requires 5 experienced workers to one person with no assembly experience, greatly reducing labor costs and dependence on personnel experience.

[0043] 3. This invention ensures the quality and accuracy of testing. Relying on AI target detection technology (YOLOv5 algorithm), the detection accuracy rate reaches 95%. It can cover scenarios that are difficult to detect accurately by humans, such as incorrect or missing connectors, extraneous objects on the satellite, and cable bending radius. It realizes multi-dimensional and quantitative assembly quality inspection and meets the strict requirements of mass-produced satellites for assembly accuracy.

[0044] 4. This invention improves management and traceability capabilities. Through a three-level data management architecture of "satellite-subsystem-standalone machine" and historical record traceability function, it realizes classified storage, rapid retrieval and long-term archiving of test data. Combined with standardized test processes and automatically generated standard reports, it provides data support for quality traceability, problem analysis and subsequent production optimization, and improves the level of assembly quality control.

[0045] 5. Enhanced scenario adaptability: The use of portable cameras and computers and displays with portable power supplies makes it suitable for scenarios where satellite transportation is difficult and workstations are uncertain. It does not rely on fixed testing equipment and can flexibly carry out on-site testing, avoiding equipment compatibility issues that restrict assembly progress.

[0046] Other beneficial effects of the present invention will be explained in detail through the introduction of specific technical features and technical solutions in specific embodiments. Those skilled in the art should be able to understand the beneficial technical effects brought about by these technical features and technical solutions through the introduction of these technical features and technical solutions. Attached Figure Description

[0047] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0048] Figure 1 This is a schematic diagram illustrating the overall development process of the testing platform;

[0049] Figure 2 This is a schematic diagram illustrating the working principle of the testing platform;

[0050] Figure 3 This is a flowchart illustrating the process of satellite assembly quality inspection using client software.

[0051] Figure 4 This is a diagram showing the device connections;

[0052] Figure 5 This is a schematic diagram of the process database;

[0053] Figure 6 This is a schematic diagram illustrating software design and application.

[0054] Figure 7 This is a schematic diagram illustrating software design and application.

[0055] Figure 8 This is a layout diagram of the hardware equipment. Detailed Implementation

[0056] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0057] This invention provides a batch-production intelligent inspection platform for satellite assembly, referring to... Figure 1 As shown, the testing platform specifically includes: a development support module, a hardware construction module, a data processing module, a model training and optimization module, a process data support module, and a client development and application module.

[0058] Among them, the development support module is used to carry out production line application and feasibility analysis for mass production satellite assembly and testing, identify the pain points of the existing manual testing mode (such as low efficiency, poor consistency, and difficulty in data traceability), and define the core functions and algorithm parameters of the intelligent testing platform to provide guidance for subsequent development.

[0059] The hardware building module is used to build a hardware platform adapted to satellite assembly and testing scenarios, including image acquisition equipment and computing equipment, providing hardware support for image acquisition and data processing.

[0060] Furthermore, the image acquisition device in this hardware module is a portable high-definition camera, and the computing device is a computer with a portable power supply. The hardware platform also includes a display screen for showing the inspection results. The portable high-definition camera, the computer with a portable power supply, and the display screen are communicatively connected. In use, the position of the portable high-definition camera and the lighting equipment relative to the worktable is adjusted to ensure that the acquired assembly images meet the clarity requirements, thus adapting to satellite assembly and inspection scenarios where satellite transport is difficult and the workstation position is uncertain.

[0061] The data processing module is used to collect sample datasets required for satellite assembly and inspection, and to label the sample datasets to provide a data foundation for model training.

[0062] The model training and optimization module selects a deep learning-based object detection algorithm, uses a labeled sample dataset for model training, and optimizes the model based on the training results. The model is used to perform satellite assembly quality inspection.

[0063] Furthermore, referring to Figure 2 As shown, the model training and optimization module selects the YOLOv5 algorithm as the deep learning-based object detection algorithm. This algorithm divides the input satellite assembly image into m×n cells, and each cell outputs B bounding boxes and N predicted object category probability information. Each bounding box contains one confidence score and four coordinate position information. The confidence score is used to characterize the matching degree between the predicted box and the ground truth box and the probability of the presence of a target object within the box. By adjusting the parameters of the algorithm's inference speed and detection accuracy, the optimized model's detection cycle is matched with the satellite assembly speed, avoiding the detection process from slowing down the assembly progress.

[0064] The process data support module stores the process data required for satellite assembly and inspection. This data includes at least satellite material information, assembly operation instructions, and standard assembly parameters, providing data support for satellite assembly quality inspection. The module can also be extended to interface with external CAD systems to import satellite assembly models, enriching the sources of process data. Satellite assembly quality inspection includes at least one of the following: connector misinstallation / omission detection, fastener misinstallation / omission detection, onboard foreign object detection, clamp adhesion location detection, cable bending radius detection, and identification of unit and connector serial numbers and their corresponding relationships. This comprehensively covers the quality control needs of the entire satellite assembly process, avoiding the incomplete coverage of traditional manual inspection scenarios.

[0065] The client development and application module is used to develop client software and integrate the optimized target detection model into the client software to realize intelligent detection and data management of satellite assembly quality.

[0066] Furthermore, the client software developed by this client development and application module contains multiple functional sub-modules, specifically implemented as follows:

[0067] (1) Data Management Submodule: It adopts a three-level folder structure of "satellite-subsystem-unit", supports users to create custom satellite model folders and enter basic satellite information; it supports batch uploading of unit assembly images to the corresponding folders, and the uploaded images are automatically stored in the system database. It can be quickly retrieved by satellite name and unit name, realizing the classification, storage and traceability of test data.

[0068] (2) Quality Inspection Submodule: Supports the creation of inspection tasks. Inspection tasks require the input of satellite name and satellite type information, and support the query of tasks by satellite name, type and number; standardizes the image acquisition of single-unit sub-components. Cable sub-components need to acquire images of plugs, interfaces, uninserted state and connected state. After the image acquisition is completed, the target inspection model is triggered to perform assembly quality inspection and generate the single-unit pass rate and the system judgment conclusion of "pass / fail".

[0069] (3) Manual review and reporting submodule: The quality inspection submodule also supports manual review function, which is only open to administrators. Administrators can perform "approval" or "re-inspection" operations based on the inspection results of the target inspection model. If there are abnormalities in the inspection results, the system will report the error and prompt the reason for the abnormality, and guide the re-inspection after correction and assembly. After approval, the system will automatically generate a Word format inspection report containing basic satellite information, inspection overview statistics and abnormal details. The inspection report can be exported and archived.

[0070] This invention also provides a method for developing and applying a batch-production intelligent inspection platform for satellite assembly, including the following:

[0071] Development Support Phase (Step S1): Conduct production line application and feasibility analysis for mass production satellite assembly and testing, define the core functions and algorithm parameters of the intelligent testing platform, and complete the development support work;

[0072] Hardware construction phase (step S2): Build the hardware platform, adjust the relative positions of the devices to ensure image acquisition clarity, and complete the hardware construction;

[0073] Data processing stage (step S3): Collect the sample dataset required for satellite assembly and inspection, and label the sample dataset with target location and category annotations to complete the data processing;

[0074] Model training optimization stage (step S4): Select the YOLOv5 algorithm, use the labeled sample dataset for model training, and optimize the algorithm's inference speed and detection accuracy by adjusting parameters to make the model's detection beat match the satellite assembly speed, thus completing the model training optimization.

[0075] Client Development and Application Phase (Step S5): Develop client software, integrate the optimized YOLOv5 model into the software, and realize the creation of detection tasks, image uploading, AI detection, manual review and report export through the client software, thus completing the client development and application.

[0076] Among them, reference Figure 3 As shown, the specific process of satellite assembly quality inspection via client software in step S5 includes:

[0077] Step S51: The user logs into the client software, selects the detection task corresponding to the satellite model, and enters the standalone image acquisition and import interface;

[0078] Step S52: Upload the single-unit assembly image to be inspected. The client software backend filters whether the image meets the acquisition specifications. If it does not meet the specifications, return to step S51 and repeat the image acquisition and upload operation.

[0079] Step S53: Call the YOLOv5 model to detect images that meet the requirements. If there are any abnormalities in the detection results, the system will report an error and indicate the reason for the error. After correction and assembly, return to step S52 to re-upload.

[0080] Step S54: After the test is passed, the administrator will perform a manual review. If the review is successful, a Word format test report will be generated. If the review fails, return to step S53 for retesting.

[0081] Meanwhile, in step S5, the client software adopts a three-level data management architecture of "satellite-subsystem-standby machine", which supports batch image uploading and name retrieval. Through a two-stage judgment mechanism of "AI automatic detection + administrator manual review" and combined with exportable standardized inspection reports, the efficiency and accuracy of satellite assembly inspection are achieved, reducing labor costs and rework rate.

[0082] The present invention will now be described in more detail.

[0083] The present invention provides a batch-production intelligent inspection platform for satellite assembly, (1) the working principle of which is as follows: Figure 2 As shown, the core of this intelligent inspection platform lies in how to accurately and efficiently automatically identify the assembly result images captured and uploaded during the production process, and clearly feed the judgment results back to the production end. Achieving this key point requires employing a one-stage object detection algorithm based on neural network deep learning. This algorithm divides the acquired input image into m×n cells, with each cell outputting B bounding boxes and N predicted object category probabilities. Each bounding box consists of one confidence score and four coordinate position information points. The confidence score indicates how close the predicted box output by the network is to the ground truth box, and the probability that it contains an object.

[0084] For specific work procedures, please refer to... Figure 3 As shown, the platform's basic functions include:

[0085] a) Detection of incorrect or missing connectors and fasteners;

[0086] b) Identification of stand-alone unit and connector numbers and their corresponding relationships;

[0087] c) Detection of unwanted objects on the satellite;

[0088] d) Inspection of clamp adhesion position;

[0089] e) Cable bending radius detection;

[0090] Client-side software development and interface design should meet the following functions:

[0091] a) System Homepage Function Navigation

[0092] Interface layout: The top of the homepage displays a welcome message and user information, the middle displays key data indicators (number of newly inspected surfaces today, number of abnormal single machines this week, etc.), and the bottom presents satellite structure drawings and a list of single machine assembly progress.

[0093] Navigation bar settings: The left navigation bar contains entry points for functional modules such as "Data Management", "Quality Inspection", and "History", allowing users to quickly switch between operating scenarios according to their inspection needs.

[0094] b) Data Management

[0095] Hierarchical Management: Click "Add Satellite" to create a custom folder for the corresponding satellite model, and enter basic information such as satellite name and type; if it has already been added, click on a satellite folder to jump to the single-user operation interface.

[0096] Logical architecture: A three-level folder structure of "satellite-subsystem-standby" is adopted to realize the classification, storage and management of detection data, which facilitates subsequent retrieval and access.

[0097] Upload Guidelines: Click on the single-user folder to enter the image upload interface, which supports batch uploading of single-user images (no quantity limit).

[0098] Data storage: All uploaded images are automatically stored in the system database, and can be quickly searched and viewed by satellite or single-camera name.

[0099] c) Quality Inspection

[0100] Task Creation: On the quality inspection interface, click "Create Task," fill in the necessary information such as satellite name and type, and optionally enter non-essential conditions such as satellite number. After task creation, a list of inspection tasks will be automatically generated, displaying the task status (pending inspection, in progress, completed, etc.).

[0101] Query function: You can quickly search for corresponding tasks by satellite name, type, number and other conditions, which makes it easy to track the detection progress in real time.

[0102] Data Acquisition Specifications: Upon entering the standalone image acquisition interface, the left-hand list displays each standalone machine and its sub-components. For cable components, four images must be uploaded: plug, interface, unplugged, and plugged-in. Click "Upload Image" to select the standard image from the uploaded images and import it. All sub-component images for each standalone machine must be imported. Only when the acquisition status displays "Completed" can the testing phase begin.

[0103] Progress monitoring: The left side of the single-machine list displays the image import completion status intuitively by checking the status boxes, ensuring the integrity of the test data.

[0104] Inspection Process: Clicking the "Inspect" button triggers the AI ​​inspection algorithm, and the system automatically identifies defects and analyzes assembly deviations in the single-machine image. After the system completes the AI ​​inspection, the user can click "Submit for Review" for further manual review.

[0105] Data output: The system automatically calculates the single-machine pass rate and generates the system judgment conclusion (pass / fail), providing data support for subsequent review.

[0106] Review Process: System administrators can access the manual review interface (invisible to regular users) to check, verify, and correct errors based on the AI ​​detection results. Reviewers can select "Approved" or "Re-check," and the review results are recorded in the system in real time.

[0107] Report generation: After approval, the system automatically generates a Word format detection report (including basic satellite information, detection overview statistics, anomaly details, etc.) based on the AI ​​detection results and manual review results. It supports export and archiving.

[0108] d) Historical Records and Data Tracing

[0109] Filtering function: The history interface supports querying corresponding detection records by conditions such as task name, number, satellite, and detection result.

[0110] Status Display: The center of the interface displays all detection records, including task status, detection results, and executable operations (view details, export report, delete, etc.). Users can view the review status on this page; if it is "completed," a report can be exported to view detailed information.

[0111] Report Export: Click "Export Report" to generate the corresponding satellite's inspection report. The report is complete and standardized, and can be used as a basis for quality traceability and decision-making.

[0112] Data archiving: All test reports and historical records are automatically archived to the system, supporting long-term query and traceability, which facilitates subsequent data analysis and system optimization.

[0113] e) System Settings

[0114] The system settings are divided into three sub-modules: model management, user management, and personal center.

[0115] Model management and user management can effectively manage algorithm models and user information, display relevant information, and implement functions such as adding, deleting, modifying, and querying.

[0116] The user center displays the user's basic information and communication information, as well as account type, account status, and login time.

[0117] Device connection:

[0118] Reference Figure 4 As shown, due to the relative difficulty of transporting the entire satellite, the computer and display screen need to be equipped with a portable power supply, and the camera also needs to be portable. For personnel performing the inspection operation, they only need to hold the camera and take pictures or videos of the area to be inspected on the satellite, and the system can automatically generate the inspection results and display them on the screen.

[0119] The hardware devices are portable, high-definition, high-performance, and multi-angle, providing a guarantee for the processing and operation of software algorithms.

[0120] Based on comprehensive process inputs, the software application can construct a complete operational process from task creation to quality inspection, data management, and historical records.

[0121] (1) Process Database

[0122] Reference Figure 5 As shown, the process database serves as the information input and construction basis for the machine vision module, enabling data sharing and mutual feedback between the various functional modules of the two, which is the key aspect of this invention.

[0123] The process database module is primarily responsible for storing and managing process data, including material information, guiding resources, 3D models, operating instructions, and monitoring data. It can also be extended to external CAD systems to import assembly models. The accuracy and completeness of this process information directly impact the testing precision and efficiency of the software product.

[0124] In the process of building the process database, issues such as the updating and iteration of spacecraft products may be encountered. Therefore, ensuring that the process data of aerospace products and the calculation code of the software application are updated is a challenge in this stage.

[0125] (2) Software Design and Application

[0126] Reference Figure 6 and Figure 7 As shown, the software is the core of the intelligent inspection platform, and it needs to include multiple modules such as data management, online inspection, and historical record query to ensure that the inspection platform can realize the functions of each part. During software development, based on complete process inputs, a complete operational flow can be constructed from task creation to quality inspection, data management, and historical record query.

[0127] The software homepage displays key data metrics (number of newly inspected surfaces today, number of abnormal single units this week, etc.), and the bottom shows satellite structure drawings and a list of single unit assembly progress.

[0128] The software needs to clearly define image upload specifications to ensure the integrity of the data to be tested. For example, a single-camera image must include images taken from six different angles and a torque graph.

[0129] The software must have a real-time query function. Users should be able to quickly search for corresponding tasks using criteria such as satellite name, type, and number, facilitating real-time tracking of testing progress.

[0130] After the operator uploads the captured image, the software backend needs to screen and judge whether the uploaded image meets the feasibility of detection, and needs to feed back the judgment criteria and results to the operator. Therefore, ensuring the timeliness of backend information processing is a challenge in this process.

[0131] After the AI ​​inspection is completed, the system automatically calculates the single-machine pass rate, generates system judgment conclusions and analysis of assembly anomalies (if any), providing data support for subsequent manual review.

[0132] (3) Equipment selection

[0133] Hardware and software are like the skeleton and muscles of an intelligent detection platform. The quality and sophistication of the equipment will affect the speed and accuracy of the detection.

[0134] In this project, because the entire satellite being inspected currently has the characteristic of uncertain positioning, the inspection platform needs to be established with the help of portable cameras and corresponding computing equipment, so as not to affect the transportation of the entire satellite.

[0135] Hardware equipment layout such as Figure 8 As shown, the inspection of the assembly area is achieved by scanning, capturing, and uploading images and videos to the client software.

[0136] Note: Before performing assembly inspection in a specific area, adjust the relative positions of the camera and lighting equipment to the workbench to meet the inspection conditions such as image acquisition clarity.

[0137] In this embodiment of the invention, regarding the algorithm: deep learning-based object detection algorithms can be divided into two categories, namely two-stage object detection algorithms and one-stage object detection algorithms.

[0138] While two-stage object detection algorithms significantly outperform traditional object detection algorithms, the selective region search (ROI) algorithm leads to a large amount of redundant computation within R-CNN, and the R-CNN algorithm requires scaling the ROI to a fixed size. Popular two-stage object detection algorithms in recent years include Libra R-CNN, CenterNet, Dynamic R-CNN, QueryDet, and Sparse R-CNN.

[0139] One-stage object detection algorithms eliminate the need for the region search process of two-stage algorithms, directly outputting the location and category information of objects in an image using convolutional neural networks. Compared to two-stage object detection algorithms, one-stage algorithms, while slightly less accurate, offer a significant speed advantage. The YOLO algorithm uses regression to directly obtain the location and category of objects, greatly reducing the computational load on the network, but it also suffers from poor detection performance when detecting small or densely packed objects.

[0140] In terms of software development: First, the client software can be developed and iterated as an independent product. This allows the software to integrate the detection, data processing, and system management of different batches of satellites.

[0141] Accordingly, the software can be selectively imported into the production line MES system for compatible use and merged with the satellite production management system, which enables collaborative sharing of satellite assembly process and result data.

[0142] Based on the above solutions, the current inspection of assembly results relies on the visual judgment of workers. The experience level of the inspectors, fatigue caused by long inspection periods, and differences in management methods can all lead to errors such as missing parts or incorrect assembly at critical locations, greatly increasing the rework rate, reducing production efficiency, and thus having an adverse impact on the entire satellite production line.

[0143] The establishment of an intelligent inspection platform for satellite assembly has the following four advantages and benefits compared to the traditional manual inspection mode:

[0144] (1) Labor costs: can be reduced from 5 experienced workers to 1 person with no assembly experience;

[0145] (2) Acceptance efficiency improved by 80%;

[0146] (3) Accuracy rate 95%;

[0147] (4) The rework rate was reduced by 98%;

[0148] (5) The platform adopts the YOLOv5 algorithm model. Compared with other existing algorithm models, this model is based on the convolutional neural network algorithm. By optimizing the algorithm or using a specially designed model structure, the accuracy of target detection and the inference speed have been greatly improved.

[0149] This invention provides a batch-production satellite assembly intelligent inspection platform and its development and application method. By integrating hardware, algorithms and software through modular design, it realizes intelligent and standardized batch-production satellite assembly inspection, solves a series of pain points of traditional manual inspection, and can be widely used in the assembly quality control of batch-production spacecraft.

[0150] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0151] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A batch-production intelligent testing platform for satellite assembly, characterized in that, include: The system includes a development support module, a hardware construction module, a data processing module, a model training and optimization module, a process data support module, and a client development and application module. The development support module is used to conduct production line application and feasibility analysis for mass production satellite assembly and testing, and to define the core functions and algorithm parameters of the intelligent testing platform. The hardware building module is used to build a hardware platform adapted to satellite assembly and testing scenarios. The hardware platform includes at least image acquisition equipment and computing equipment. The data processing module is used to collect sample datasets required for satellite assembly and testing, and to label the sample datasets. The model training optimization module is used to select a deep learning-based object detection algorithm, train the model using the labeled sample dataset, and optimize the model based on the training effect. The process data support module is used to store the process data required for satellite assembly and inspection. The process data includes at least satellite material information, assembly operation instructions and assembly standard parameters, and provides data support for satellite assembly quality inspection. The client development and application module is used to develop client software and integrate the optimized target detection model into the client software to realize intelligent detection and data management of satellite assembly quality.

2. The intelligent testing platform for batch-production satellite assembly according to claim 1, characterized in that, The hardware platform also includes a display for showing the test results, and the image acquisition device and the computing device are both communicatively connected to the display; The hardware platform is also equipped with lighting equipment. When in use, the position of the image acquisition device and the lighting equipment relative to the workbench is adjusted to ensure that the acquired assembly images meet the clarity requirements.

3. The intelligent testing platform for batch-production satellite assembly according to claim 1, characterized in that, The model training optimization module selected the YOLOv5 algorithm as the deep learning-based object detection algorithm. The YOLOv5 algorithm divides the input satellite assembly image into m×n cells. Each cell outputs B bounding boxes and N predicted object category probability information. Each bounding box contains one confidence score and four coordinate position information. The confidence score is used to characterize the matching degree between the predicted box and the ground truth box and the probability of the presence of a target object within the box. By adjusting the parameters of the algorithm's inference speed and detection accuracy, the optimized model's detection rate is matched with the satellite assembly speed.

4. The intelligent testing platform for batch-production satellite assembly according to claim 1, characterized in that, The client software developed by the client development and application module includes: a data management sub-module; The data management submodule adopts a three-level folder structure of satellite-subsystem-unit, supports the creation of custom satellite model folders and the entry of basic satellite information, and supports batch uploading of unit assembly images to the corresponding folders; uploaded images are automatically stored in the system database, and support quick retrieval by satellite name and unit name, realizing the classified storage and traceability of detection data.

5. The intelligent testing platform for batch-production satellite assembly according to claim 4, characterized in that, The client software also includes a quality inspection submodule; The quality inspection submodule supports the creation of inspection tasks. The inspection tasks require the input of satellite name and satellite type information, and can be queried by satellite name, type, and number. The quality inspection submodule also standardizes the image acquisition of individual sub-components. For cable sub-components, images of plugs, interfaces, uninserted states, and connected states need to be acquired. After the image acquisition is completed, the target detection model is triggered to perform assembly quality inspection and generate the single-unit pass rate and the system judgment conclusion of pass / fail.

6. The intelligent testing platform for batch-production satellite assembly according to claim 5, characterized in that, The quality inspection submodule also supports a manual review function, which is only available to administrators. The administrator performs approval or re-detection operations based on the detection results of the target detection model. If there are abnormalities in the detection results, the system will report an error and indicate the reason for the abnormality, guiding the correct assembly and re-detection. Once approved, the system automatically generates a Word-formatted inspection report containing basic satellite information, inspection overview statistics, and anomaly details. The inspection report can be exported and archived.

7. The intelligent testing platform for batch-production satellite assembly according to claim 1, characterized in that, The process data support module can also be extended to interface with external CAD systems for importing satellite assembly models; The satellite assembly quality inspection includes at least one or more of the following: incorrect or missing connector installation, incorrect or missing fastener installation, detection of extraneous objects on the satellite, detection of clamp pasting position, detection of cable bending radius, identification of unit and connector number and corresponding relationship identification, covering the quality control needs of the entire satellite assembly scenario.

8. A method for developing and applying a batch-production intelligent inspection platform for satellite assembly, characterized in that, include: Step S1: Conduct production line application and feasibility analysis for mass production satellite assembly and testing, define the core functions and algorithm parameters of the intelligent testing platform, and complete the development support work; Step S2: Build the hardware platform, adjust the relative positions of the devices to ensure image acquisition clarity, and complete the hardware construction; Step S3: Collect the sample dataset required for satellite assembly and inspection, and label the sample dataset with target locations and categories to complete the data processing; Step S4: Select the YOLOv5 algorithm, use the labeled sample dataset for model training, and optimize the algorithm's inference speed and detection accuracy by adjusting parameters to make the model's detection beat match the satellite assembly speed, thus completing the model training optimization. Step S5: Develop client software, integrate the optimized YOLOv5 model into the software, and realize the creation of detection tasks, image uploading, AI detection, manual review and report export through the client software, thus completing the client development and application.

9. The development and application method of the intelligent inspection platform for batch-production satellite assembly according to claim 8, characterized in that, The specific process of satellite assembly quality inspection via client software in step 5 includes: Step S51: The user logs into the client software, selects the detection task corresponding to the satellite model, and enters the standalone image acquisition and import interface; Step S52: Upload the single-unit assembly image to be inspected. The client software backend filters whether the image meets the acquisition specifications. If it does not meet the specifications, return to step S51 and repeat the image acquisition and upload operation. Step S53: Call the YOLOv5 model to detect images that meet the requirements. If there are any abnormalities in the detection results, the system will report an error and indicate the reason for the error. After correction and assembly, return to step S52 to re-upload. Step S54: After the test is passed, the administrator will perform a manual review. If the review is successful, a Word format test report will be generated. If the review fails, return to step S53 for retesting.

10. The development and application method of the intelligent inspection platform for batch-production satellite assembly according to claim 8, characterized in that, In step S5, the client software adopts a three-level data management architecture of satellite-subsystem-standby machine, supports batch image uploading and name retrieval; through a two-stage judgment mechanism of AI automatic detection and administrator manual review, combined with the exported standardized inspection report, satellite assembly inspection is realized.

Images