High and low orbit ku frequency band satellite terminal surface defect detection method and system

Abstract

The invention provides a high and low orbit ku frequency band satellite terminal surface defect detection method and system, and relates to the technical field of satellite communication, and the method comprises the steps: carrying out the non-uniform deployment of multi-source sensing nodes at a high and low orbit ku frequency band satellite terminal, and constructing a multi-source sensing array; a real-time sensing data array is obtained, and millisecond-level synchronization is carried out through the Beidou time service module; obtaining a plurality of single-dimensional sensing data sets; executing defect feature extraction of multiple parallel threads to obtain a plurality of single-dimensional defect feature sets; carrying out confidence weighted fusion judgment on the defect overlapping region to obtain a fused defect feature set; calling a pre-bound environmental stress parameter sequence; and performing defect time-space long-term and short-term evolution prediction, and outputting critical failure time. The technical problem that the satellite terminal performance and the satellite communication quality are affected due to low defect positioning and judgment precision caused by the fact that a single sensor is adopted to monitor the satellite terminal in surface defect detection in the prior art and the full view of the defect is difficult to comprehensively capture is solved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of satellite communication, in particular to a high-low-orbit ku-band satellite terminal surface defect detection method and system. BACKGROUND

[0002] The occurrence of surface defects of a satellite terminal will seriously affect the performance and service life of the satellite, especially in the high-frequency band (such as the ku-band) communication system of a high-low-orbit satellite. The surface defects not only affect the signal transmission quality, but also can cause system failure or failure. Therefore, real-time monitoring and accurate diagnosis of the surface defects of the satellite terminal are the key to ensuring the long-term stable operation of the satellite system. However, the traditional surface defect detection relies on a single sensor to monitor the surface of the satellite terminal. Such a single sensor can only provide partial and limited defect information, and it is difficult to fully capture the overall situation of the defects, resulting in low positioning and judgment accuracy of the defects. Therefore, the defects at the key positions cannot be identified in a timely and accurate manner, potential failures cannot be found in an early stage, and the performance of the satellite terminal and the quality of satellite communication are affected. SUMMARY

[0003] The application provides a high-low-orbit ku-band satellite terminal surface defect detection method and system, which aims to solve the technical problem that the surface defect detection of the prior art uses a single sensor to monitor the satellite terminal, which is difficult to fully capture the overall situation of the defects, resulting in low positioning and judgment accuracy of the defects, and further affecting the performance of the satellite terminal and the quality of satellite communication.

[0004] According to a first aspect of the application, a high-low-orbit ku-band satellite terminal surface defect detection method is provided, which comprises: according to risk area distribution data, non-uniformly deploying a multi-source sensing node on a high-low-orbit ku-band satellite terminal to construct a multi-source sensing array; a defect detection middle station interacts with the multi-source sensing array at an interval of 0.5 seconds to obtain a real-time sensing data array, wherein the real-time sensing data array is synchronized at a millisecond level via a Beidou timing module; the real-time sensing data array is decomposed according to the sensor type to obtain a plurality of single-dimensional sensing data sets with node position labels; defect feature extraction is performed on the plurality of single-dimensional sensing data sets in multiple parallel threads to obtain a plurality of single-dimensional defect feature sets; after the plurality of single-dimensional defect feature sets are restored according to the node position labels, confidence weighted fusion decision is performed on the defect overlapping area to obtain a fused defect feature set; a pre-bound environmental stress parameter sequence is retrieved according to the device ID of the high-low-orbit ku-band satellite terminal; the fused defect feature set is subjected to defect spatio-temporal long-short-term evolution prediction according to the environmental stress parameter sequence, and a critical failure time is output.

[0005] In a second aspect, the application provides a high-low orbit Ku-band satellite terminal surface defect detection system, which is used for the high-low orbit Ku-band satellite terminal surface defect detection method, and comprises: a sensor node deployment module, configured to perform non-uniform deployment of multi-source sensor nodes on a high-low orbit Ku-band satellite terminal according to risk area distribution data, and construct a multi-source sensing array; a real-time sensing data acquisition module, configured to acquire a real-time sensing data array from the multi-source sensing array at an interval of 0.5 seconds by a defect detection middle station, wherein the real-time sensing data array is synchronized at a millisecond level via a Beidou timing module; a single-dimensional sensing data acquisition module, configured to decompose the real-time sensing data array according to a sensor type, and obtain a plurality of single-dimensional sensing data sets with node position labels; a defect feature extraction module, configured to perform defect feature extraction of the plurality of single-dimensional sensing data sets in multiple parallel threads, and obtain a plurality of single-dimensional defect feature sets; a weighted fusion decision module, configured to perform confidence weighted fusion decision of a defect overlap area after restoring the plurality of single-dimensional defect feature sets according to the node position labels, and obtain a fused defect feature set; an environmental stress parameter retrieval module, configured to retrieve a pre-bound environmental stress parameter sequence according to a device ID of the high-low orbit Ku-band satellite terminal; and an evolution prediction module, configured to perform defect spatiotemporal long-short term evolution prediction on the fused defect feature set according to the environmental stress parameter sequence, and output a critical failure time.

[0006] The one or more technical solutions provided in the application have at least the following beneficial effects: By deploying sensor nodes in satellite terminals according to the risk area distribution data, a multi-source sensing array that can adapt to different regional risk characteristics is constructed, ensuring higher detection density in critical areas and improving the sensitivity and accuracy of detection. The defect detection center interacts with the multi-source sensing array at 0.5 second intervals, obtaining real-time sensing data. The data is synchronized to the millisecond level through the Beidou timing module, ensuring the timeliness and accuracy of the sensor data. The advantage of real-time synchronization is that it can quickly capture changes in the state of the equipment, identify potential surface defects in advance, and enhance real-time monitoring capabilities. Real-time sensing data is decomposed according to sensor type and labeled with node location for each data set, allowing more accurate identification of spatial distribution of data. Through labeling, different sensor types can be effectively distinguished, reducing data analysis errors and improving the efficiency and accuracy of subsequent analysis. Defect feature extraction is performed on multiple single-dimensional sensing data sets through multiple parallel threads, enabling fast processing of large amounts of data. This parallel processing significantly improves the efficiency of feature extraction, reducing data processing time and ensuring real-time and accuracy even in large-scale data collection. Based on node location labels, multiple single-dimensional defect feature sets are restored and defect overlap areas are weighted and fused for decision-making. Through weighted fusion, defect areas can be more accurately determined, effectively reducing false positives and false negatives, and ensuring the accuracy of defect determination. By retrieving pre-bound environmental stress parameter sequences based on device ID and associating them with the fused defect feature set, the depth and accuracy of detection are improved. This method considers the impact of environmental factors on defect development, providing more comprehensive understanding of the defect state of satellite terminals and more accurate data for subsequent failure prediction. Through spatiotemporal long-term evolution prediction of the fused defect feature set based on environmental stress parameter sequences, critical failure time can be accurately predicted. This prediction not only helps the system respond before failure occurs, but also provides scientific maintenance and replacement cycle recommendations for satellite terminals, significantly extending the service life of the equipment, improving satellite communication quality, and enhancing the safety and reliability of the overall system.

[0007] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the following specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 A high-low orbit Ku-band satellite terminal surface defect detection method flowchart is provided for the embodiments of the present application.

[0009] Figure 2A high-low orbit ku frequency band satellite terminal surface defect detection system structure schematic diagram provided by an embodiment of the application.

[0010] The reference signs are explained as follows: a sensor node deployment module 10, a real-time sensing data acquisition module 20, a single-dimensional sensing data acquisition module 30, a defect feature extraction module 40, a weighted fusion decision module 50, an environmental stress parameter retrieval module 60, and an evolution prediction module 70. DETAILED DESCRIPTION

[0011] The embodiment of the application provides a high-low orbit ku frequency band satellite terminal surface defect detection method and system, and solves the technical problem that the surface defect detection in the prior art uses a single sensor to monitor a satellite terminal, it is difficult to comprehensively capture the overall situation of defects, the positioning and judgment accuracy of defects are low, and then the performance of the satellite terminal and the satellite communication quality are affected.

[0012] After introducing the basic principle of the application, various non-limiting embodiments of the application will be specifically introduced in combination with the drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the application and not to limit the application.

[0013] Embodiment one, as shown in the figure, the embodiment of the application provides a high-low orbit ku frequency band satellite terminal surface defect detection method, the method comprises: Figure 1 According to the risk area distribution data, a non-uniform deployment of multi-source sensor nodes is performed on the high-low orbit ku frequency band satellite terminal to construct a multi-source sensing array. According to the risk area distribution data, a non-uniform deployment of multi-source sensor nodes is performed on the high-low orbit ku frequency band satellite terminal to construct a multi-source sensing array.

[0014] The risk area distribution data of the high-low orbit ku frequency band satellite terminal is obtained, the risk area distribution data is obtained based on the surface features of the satellite, historical defect records, environmental conditions and other factors, and can be obtained through historical detection records, simulation or experimental results. The risk area distribution data contains which areas of the satellite surface have a higher frequency of defects, and these high-frequency defect areas are risk areas. According to the risk area distribution data, different types of sensors are selected for deployment, for example, the density of sensors is increased in high-risk areas to more accurately monitor defects. Using this non-uniform deployment method, more sensor nodes are ensured in critical areas (such as high-risk areas) to improve monitoring efficiency and accuracy. The sensor nodes include industrial vision cameras, terahertz imagers, infrared thermal imagers and the like, which are used to capture data related to defects. Through the above deployment, a multi-source sensing array covering the entire high-low orbit ku frequency band satellite terminal is constructed, and the array is composed of multiple sensor nodes, each of which can acquire and transmit data in real time.

[0015] The defect detection middle station interacts with the multi-source sensor array at an interval of 0.5 seconds to obtain a real-time sensor data array, wherein the real-time sensor data array is synchronized at a millisecond level via a Beidou timing module.

[0016] The defect detection middle station periodically obtains data of the sensor nodes by interacting with the multi-source sensor array, and the data includes measurement values of the sensors and corresponding time stamps. The interval of data acquisition is 0.5 seconds, which means that data is obtained from multiple sensor nodes every half second to generate a real-time sensor data array. The obtained real-time sensor data array is synchronized at a millisecond level via a Beidou timing module. Since there may be slight time errors among different sensor nodes, the Beidou timing module can accurately synchronize the data collection times of each node, ensuring that all sensor data in the entire system can be compared and analyzed on the same time axis. This synchronization process ensures that the data obtained by multiple sensors has consistent time labels, avoiding inconsistent data due to time deviation and affecting the accuracy of defect detection.

[0017] The real-time sensor data array is decomposed according to sensor types to obtain multiple single-dimensional sensor data sets with node position labels.

[0018] The obtained real-time sensor data array is a comprehensive data set composed of multiple sensors. The output data of each sensor type has different physical characteristics. According to the sensor types, the entire real-time sensor data array is decomposed into multiple single-dimensional sensor data sets, each corresponding to the data of one sensor type. Each sensor node has a corresponding node position label indicating its specific position on the satellite terminal. When decomposing the data, a corresponding node position label is added to each data item in each single-dimensional sensor data set, ensuring that each piece of data can be traced back to its physical location.

[0019] Defect feature extraction is performed on the multiple single-dimensional sensor data sets in multiple parallel threads to obtain multiple single-dimensional defect feature sets.

[0020] For multiple single-dimensional sensor data sets, multiple parallel threads are started to accelerate data processing, using parallel computing can significantly improve processing efficiency, each thread independently extracts defect features from the corresponding single-dimensional sensor data set, defect features include mutation points, fluctuation amplitude, frequency change, etc., which are used to characterize the occurrence of device surface defects, depending on the characteristics of each sensor type. Each thread runs a pre-trained lightweight defect feature extraction model, which is specifically optimized for sensor type to ensure that the extracted defect features are efficient and accurate, and the extraction process combines spatial and temporal features to ensure that the multi-dimensional dynamic changes of defects can be captured. After each parallel thread performs defect feature extraction, multiple single-dimensional defect feature sets are generated, each single-dimensional defect feature set contains not only defect feature values but also spatial and temporal information for further analysis.

[0021] After restoring the multiple single-dimensional defect feature sets according to the node position label, a confidence weighted fusion decision of the defect overlapping area is made to obtain a fused defect feature set.

[0022] By node position label, each single-dimensional defect feature set is restored to a terminal three-dimensional digital grid. The restoration process is to merge different types of sensor data according to their positions, so that the defect feature set has spatial dimensions. In a multi-source sensor array, different sensors may collect similar defect information in some areas. Determine whether these information overlaps, that is, whether multiple sensors have captured the same defect area. For overlapping areas, weight them according to the recognition confidence and collection confidence of different sensors, where high-confidence sensor data is given greater weight. The weighted and fused data generates a fused defect feature set, which contains all important defect information and has high reliability for subsequent spatiotemporal evolution prediction.

[0023] According to the device ID of the high-low orbit ku frequency band satellite terminal, the pre-bound environmental stress parameter sequence is retrieved.

[0024] The high-low orbit ku frequency band satellite terminal has a unique device ID. By device ID, the stress parameter sequence of the device under different environmental conditions is retrieved. These parameters include stress data such as pressure, temperature, vibration that the satellite surface may experience in actual work. The environmental stress parameter sequence is provided by the environmental monitoring system or historical test data, providing a dynamic change basis for defect spatiotemporal evolution.

[0025] According to the environmental stress parameter sequence, the fused defect feature set is subjected to defect spatiotemporal long-term evolution prediction, and the critical failure time is output.

[0026] In combination with the environmental stress parameter sequence and the fusion defect feature set, a space-time long short-term evolution prediction model is used to model the evolution process of the defect, which includes time series analysis, long short-term memory network, convolutional neural network, etc., for predicting the evolution trend of the defect in the future space-time. This process combines historical data (such as the evolution trajectory of surface defects) and real-time environmental stress data to calculate the possible changes and growth of the defect. The prediction result is the critical failure time of each defect, which is the time point when the defect develops to an unbearable extent and may cause the performance or structure of the satellite terminal to fail. Based on the critical failure time, maintenance instructions are triggered to inform relevant personnel to repair, replace or adjust certain parts of the satellite terminal to avoid failure.

[0027] Further, according to the risk area distribution data, a multi-source sensing array is constructed by non-uniform deployment of multi-source sensing nodes on the LEO and GSO Ku-band satellite terminals, and the method comprises: According to the device ID, the historical surface defect records of the same model device are called online, and the historical surface defect records are differentiated according to a plurality of standard terminal surface defect types to obtain a plurality of high-risk area heat maps, which constitute the risk area distribution data. After matching the sensor types according to the plurality of standard terminal surface defect types, the non-uniform deployment of multi-source sensing nodes is performed by spatially fusing the plurality of high-risk area heat maps to construct the multi-source sensing array.

[0028] According to the device ID, the historical surface defect records related to the model device are called online from the historical database or historical record system. Through online calling, these data can be quickly obtained without manual input or search. The historical surface defect records include defect information encountered by the same model device during past use, such as defect type, location, occurrence time, severity, etc., providing a reference basis for the current defect detection task.

[0029] For each type of defect, a standard terminal surface defect type is defined, such as surface deformation, crack, corrosion, etc. By standardizing different types of defects, uniform classification of various types of defects is ensured. According to a plurality of standard terminal surface defect types, the historical surface defect records are differentiated, and for each type of defect data, it is mapped to the specific area or spatial location of the device to identify which areas have frequently occurred defects in history. Through spatial and temporal analysis of data for each standard terminal surface defect type, high-risk area heat maps for each type of defect are generated, which show the defect occurrence frequency and severity at different locations. The deeper the color, the higher the risk of defect occurrence in that area. By synthesizing the high-risk area heat maps of each defect type, comprehensive risk area distribution data is generated, which comprehensively displays the potential defect areas of the device under different working environments.

[0030] According to each standard terminal surface defect type, the sensor type is matched, for example, for surface deformation type defects, the most suitable sensor is an industrial vision camera, which can capture subtle changes in surface deformation with high precision, especially for detecting surface cracks, depressions and other deformations.

[0031] The multiple high-risk area heat maps are spatially fused, each high-risk area heat map representing a high-risk area of a different defect type. The fusion process combines multiple heat maps according to spatial position and weight, resulting in a comprehensive global risk density distribution map. This global risk density distribution map can show all types of defect areas on the device, highlighting the most critical monitoring areas. Through the fused global risk density distribution map, the deployment position of each sensor node is determined. The deployment of sensor nodes is not uniform, but is arranged differently according to the defect risk size of different areas of the device. This non-uniform deployment method maximizes monitoring efficiency and ensures that high-risk areas are given priority for monitoring. The multi-source sensing array is an array composed of different types of sensors (such as industrial vision cameras, terahertz imagers, infrared thermographic imagers, etc.), which can comprehensively and accurately monitor various defects on the surface and inside the satellite terminal, thereby improving the coverage and accuracy of defect detection.

[0032] Further, after matching the sensor types according to the multiple standard terminal surface defect types, the multiple high-risk area heat maps are spatially fused to deploy the multi-source sensing nodes non-uniformly, and the multi-source sensing array is constructed. The method comprises: Through defect response testing, a mapping relationship between multiple sample defect types and multiple high-response sensor types is established, and a sensor call type library is generated. The multiple standard terminal surface defect types are loaded into the sensor call type library, and multiple adaptive sensors are matched. The multiple high-risk area heat maps are spatially weighted and fused to obtain a global risk density distribution map. According to the density attribute and defect superposition attribute of the global risk density distribution map, the multiple adaptive sensors are combined to deploy the multi-source sensing nodes non-uniformly, and the multi-source sensing array is constructed.

[0033] The defect response test is to evaluate the response of different types of defects to various sensors. The test process includes detecting different defects such as surface deformation, cracks, and thermal stress using different sensors on known sample defect types. Different defects will trigger different response modes of the sensor, such as image changes and thermal radiation changes. According to the test results, a mapping relationship between each defect type and its most suitable sensor type is established. For example, surface deformation defects correspond to industrial vision cameras, which detect surface deformation through image detection; internal structure defects correspond to terahertz imagers, which detect internal defects through electromagnetic wave penetration; and thermal stress defects correspond to infrared thermal imagers, which detect thermal stress through temperature changes. Using the above mapping relationship between defects and sensors, a sensor call type library is created, which is a dynamic data structure containing the correspondence between different defect types and sensor types.

[0034] Various standard terminal surface defect types are loaded into the sensor call type library. Each defect type is matched in the sensor call type library. For each defect, the sensor call type library returns the corresponding adaptive sensor, ensuring that the sensor device that can efficiently detect this type of defect is selected.

[0035] For multiple high-risk area heat maps, the weight of each high-risk area heat map is set according to the severity or frequency of the defect type. Heat maps with larger weights will have a greater impact on the final result. By weighting, different types of high-risk area heat maps are superimposed together. This fusion operation combines the spatial location attribute to ensure that the defect information of different regions is not lost when merged. The result of this fusion process is a comprehensive global risk density distribution map that shows the risk distribution of all types of defects on the satellite terminal, highlighting high-risk areas.

[0036] The density attribute is reflected in the color depth of the global risk density distribution map, indicating which areas need the most monitoring. For example, for areas with higher risk, more sensor nodes should be deployed, while for areas with lower risk, the number of sensor nodes can be appropriately reduced or low-response sensors can be selected. Combining the obtained multiple adaptive sensors, the type and number of sensors required for each region are determined. For example, industrial vision cameras are deployed in high-risk surface deformation areas, terahertz imagers are deployed in areas with high risk of internal structure defects, and infrared thermal imagers are deployed in areas with more thermal stress defects. Each adaptive sensor is deployed according to its characteristics and needs in different areas to ensure accurate monitoring of various defects.

[0037] Non-uniform deployment means that sensors are not evenly distributed on the surface of the equipment or space, but are arranged according to the analysis results of the global risk density distribution map. After deployment, the sensors cover the surface of the equipment and key areas, forming a multi-source sensor array. This array can simultaneously detect defects from multiple angles and multiple sensor types, ensuring the accuracy and integrity of the data.

[0038] Further, performing defect feature extraction of multiple parallel threads on the plurality of single-dimensional sensor data sets to obtain a plurality of single-dimensional defect feature sets, the method comprising: The plurality of standard terminal surface defect types and the plurality of adaptive sensors are used as sample defect model identification retrieval conditions to retrieve a plurality of sample defect sensor data and a plurality of sample defect identification data. The plurality of sample defect sensor data and the plurality of sample defect identification data are used as training data to construct a plurality of lightweight feature extraction models, wherein the plurality of lightweight feature extraction models are deployed to a plurality of heterogeneous computing units. After pre-creating a plurality of independent computing threads for a plurality of single-dimensional sensor data, the plurality of heterogeneous computing units are mapped and bound to the plurality of independent computing threads. The plurality of single-dimensional sensor data sets are distributed to the plurality of independent computing threads to drive the plurality of heterogeneous computing units to perform defect feature extraction of multiple parallel threads and output the plurality of single-dimensional defect feature sets, wherein the single-dimensional defect features have spatial coordinates.

[0039] According to the plurality of standard terminal surface defect types and the plurality of adaptive sensors that have been determined as retrieval conditions, a plurality of sample defect sensor data related to the conditions are retrieved. These data are collected by actual sensors and obtained through multiple sample detections on equipment or in experiments. At the same time, a plurality of sample defect identification data related to the conditions are retrieved, i.e. the labels of each sample defect data, indicating that the defect corresponding to the sensor data is a crack, corrosion or other type.

[0040] The obtained plurality of sample defect sensor data and plurality of sample defect identification data are integrated into training data to construct a plurality of lightweight feature extraction models. During the training process, the model will learn how to extract useful features from the sensor data and optimize these features to improve the recognition accuracy of defects. Through backpropagation and optimization algorithms, the model continuously adjusts the weights, and finally can effectively identify different types of defects from the sensor data. Lightweight means that the model consumes low computing resources, but still extracts enough feature information. The design of these models usually includes convolutional neural networks, sparse representation models, etc., which can perform efficient feature extraction with less computing resources.

[0041] The heterogeneous computing unit refers to the allocation of computing tasks in a computing system by using different types of hardware units such as CPU, GPU, TPU, etc. Different hardware units have different computing advantages, for example, GPU is suitable for processing large-scale parallel computing tasks, while CPU is more suitable for processing serial tasks. Deploying multiple lightweight feature extraction models to multiple heterogeneous computing units can improve computing efficiency and shorten training time.

[0042] According to different types of sensing data, a corresponding independent computing thread is created for each single-dimensional sensing data, and each thread is responsible for processing a specific data set to parallelize multiple tasks. The heterogeneous computing unit is mapped and bound to the corresponding independent computing thread, which ensures efficient use of computing resources and accelerates task execution in a parallel computing environment.

[0043] The multiple single-dimensional sensing data sets are distributed to the corresponding independent computing threads, and the data distribution is based on the type of data. In each independent computing thread, defect feature extraction is performed in parallel, and each thread independently processes a single-dimensional sensing data set to extract relevant defect features through a pre-trained model. The extracted defect features have spatial coordinates, which means that each defect feature is related to a specific location. Multiple parallel threads mean that multiple threads work simultaneously without blocking each other, which greatly improves processing efficiency. During execution, computing threads do not interfere with each other, and the output of each thread is completed independently and stored. After each thread completes its task, a single-dimensional defect feature set is generated, which contains detailed information about the defect and its spatial coordinates.

[0044] Further, after restoring the multiple single-dimensional defect feature sets according to the node position label, a confidence weighted fusion decision is made for the defect overlap area to obtain a fused defect feature set. The method comprises: According to the recognition accuracy of the multiple heterogeneous computing units, multiple recognition confidences are associated. According to the sensing performance degradation parameters of the multiple adaptive sensors, multiple collection confidences are associated. The multiple single-dimensional defect feature sets are projected onto a terminal three-dimensional digital grid according to the node position label to obtain a global multi-dimensional defect feature distribution. According to the multiple recognition confidences and the multiple collection confidences, a confidence weighted fusion decision is made for the defect overlap area in the global multi-dimensional defect feature distribution to obtain a fused defect feature set.

[0045] Each heterogeneous computing unit generates a corresponding recognition confidence according to its recognition accuracy when performing defect feature extraction. Recognition accuracy refers to the correctness and accuracy of the heterogeneous computing unit when processing data. For example, the recognition accuracy of a GPU when processing image data is affected by factors such as image quality, noise interference, and the like, which in turn affect the accuracy of its output. Recognition confidence is closely related to recognition accuracy. The higher the recognition accuracy, the stronger the confidence. The recognition results output by different heterogeneous computing units are evaluated, their recognition accuracy is calculated, and the recognition accuracy is converted into a corresponding recognition confidence value, indicating the trustworthiness of the corresponding heterogeneous computing unit and the reliability of the output results.

[0046] The adaptation of the sensor's sensing performance degradation parameter refers to the degree of gradual decline in sensor performance under conditions such as long-term operation and changes in environmental factors. For example, the sensitivity of a sensor may decrease and its accuracy may deteriorate due to wear and tear, aging, or environmental influences such as excessive temperature or humidity. The sensing performance degradation parameter can be calculated by monitoring the service life of the sensor, environmental conditions, and historical data. For example, if the temperature sensing ability of a sensor has decreased over time, the degree of degradation can be calculated by comparing historical data with current data. The collection confidence reflects the reliability of the current data collected by the sensor. The calculation of the collection confidence is based on the degree of degradation of the sensor. For example, if the performance of the sensor has degraded severely, the collection confidence may be low, and vice versa. This value reflects the accuracy and reliability of the data collected by the sensor.

[0047] The node position label is used to mark the specific position of each sensor in space. Based on these node position labels, the one-dimensional defect feature set obtained from the sensor is projected into a three-dimensional digital grid. This means that each defect feature is mapped to a corresponding position in the three-dimensional digital grid according to its spatial coordinates. Each point on the three-dimensional digital grid represents the defect feature at that position, forming a comprehensive global multi-dimensional defect feature distribution. This can visually display the defect situation on the entire terminal surface and provide a basis for subsequent fusion and evaluation.

[0048] In the global multi-dimensional defect feature distribution, there may be overlapping regions between multiple defect features. These overlapping regions may be the same defect detected by multiple sensors or computing units, or similar defects detected by multiple sensors. For each overlapping region, the recognition confidence and the collection confidence are used for weighted fusion. The recognition confidence determines the credibility of the defect features in the region, and the collection confidence reflects the reliability of the data from the sensor itself. By weighted fusion of the defect features in the overlapping region, a fused defect feature set is finally generated. The fused defect feature set contains the optimal estimate of each region defect. The weighted confidence is used to determine which defect features are most credible and which may be false positives or low-reliability detection results.

[0049] Further, according to the sequence of environmental stress parameters, the defect spatiotemporal long and short term evolution prediction is performed on the fusion defect feature set, and a critical failure time is output, and the method comprises: According to the Beidou timestamp, the sequence of environmental stress parameters and the fusion defect feature set are aligned and mapped to the terminal three-dimensional digital grid to obtain a spatiotemporal coupling defect field; a double-scale prediction engine is loaded to perform defect evolution coupling iteration on the spatiotemporal coupling defect field, and a multi-stage defect damage accumulation feature is output; a plurality of defect critical failure thresholds of the plurality of sample defect types are preset; and the plurality of defect critical failure thresholds are used to traverse and compare the multi-stage defect damage accumulation feature to locate the critical failure time.

[0050] The Beidou timestamp is used to ensure the consistency of all data in time. The Beidou system provides a high-precision time synchronization function, which can accurately synchronize the time marks of each data source and ensure the time sequence consistency of multi-source data. According to the Beidou timestamp, the sequence of environmental stress parameters and the fusion defect feature set are aligned in time sequence, and the timestamp of each data point is used as the reference for alignment to ensure the synchronization of environmental stress and defect feature data. After time alignment, these data points are mapped to a three-dimensional digital grid, which represents the physical space on the surface of the satellite terminal. Each grid point in the three-dimensional digital grid can store corresponding defect data and stress data. After alignment and mapping, a spatiotemporal coupling defect field is obtained. The spatiotemporal coupling defect field couples spatial position, time information, environmental stress and defect features to form a dynamically changing defect space, providing a comprehensive data basis for subsequent defect evolution prediction.

[0051] The double-scale prediction engine refers to a prediction model that combines short-period and long-period evolution mechanisms. In this step, the double-scale prediction engine simulates the evolution process of defects and analyzes the changes of defects in different time scales. Defect evolution coupling iteration is a process of repeatedly iterating and calculating the spatiotemporal coupling defect field by the double-scale prediction engine. Through short-period evolution, the evolution process of defects in a short period of time is simulated, such as the expansion of surface cracks caused by temperature changes, which usually takes hours or days as the period. Through long-period evolution, the development process of defects in a long time scale is simulated, such as material fatigue caused by long-term stress load, which usually takes months or years as the period. After iteration is completed, a multi-stage defect damage accumulation feature is output. These features represent the damage degree of defects at a specific time point and in different stages, such as the length, depth of cracks, and accumulation of thermal stress. Multi-stage represents the evolution process of defects in different time periods, reflecting the changes and damage accumulation of defects in multiple time periods.

[0052] The sample defect type refers to different types of surface or internal defects, such as cracks, corrosion, thermal expansion, etc., each of which has its own failure mode and severity. For each sample defect type, a corresponding defect critical failure threshold is preset, which indicates that the material or equipment will fail when the defect reaches a certain level at different stages. For example, the defect critical failure threshold of a certain surface crack indicates that the equipment will not work properly when the crack expands to a certain length. The defect critical failure threshold can be set through experimental data, historical records or simulation analysis.

[0053] According to the defect critical failure threshold of each sample defect type, the multi-stage defect damage accumulation characteristics output previously are compared to determine under what conditions the defect will reach the failure threshold. During the comparison process, when the multi-stage defect damage accumulation characteristics reach or exceed the defect critical failure threshold of a certain defect type, the critical failure time of the defect is determined. The critical failure time reflects the time node at which the equipment fails at certain key positions under specific environmental stress.

[0054] Further, a double-scale prediction engine is loaded to perform defect evolution coupling iteration of the spatio-temporal coupling defect field, and output multi-stage defect damage accumulation characteristics. The method comprises: A short-period evolution step and a long-period evolution step are predefined, wherein the long-period evolution step is P times the short-period evolution step. P concatenated environmental stress slices are segmented from the environmental stress parameter sequence. The P concatenated environmental stress slices are used to drive the time series convolution network in the double-scale prediction engine to perform defect morphology iterative evolution, and output a defect morphology evolution vector. The defect morphology evolution vector is used to update the spatio-temporal coupling defect field to obtain an updated coupling defect field. After extracting cumulative stress data from the updated coupling defect field, a material performance degradation vector is calculated by the LSTM model in the double-scale prediction engine. After dynamically rewriting the boundary conditions of the time series convolution network using the material performance degradation vector, short-period evolution and long-period update are repeatedly performed until the environmental stress parameter sequence is exhausted, and the multi-stage defect damage accumulation characteristics are output.

[0055] The short-period evolution step refers to a smaller time step in the defect evolution process, usually representing the evolution process in a short time, such as hours or days, and reflects the small changes in defects caused by the instantaneous changes in environmental stress factors such as temperature and pressure; the long-period evolution step refers to a larger time step in the defect evolution process, usually representing the evolution process in a long time, such as months or years, and reflects the long-term stress and fatigue effects, such as material fatigue, thermal expansion, etc. The long-period evolution step is set to be P times the short-period evolution step, where P is a number greater than 1, i.e. the long-period evolution step is P times larger than the short-period evolution step, for example, if the short period is 1 hour, the long period is P hours. This setting allows the long-period evolution to cover multiple time steps of the short-period evolution to adapt to the characteristics of long-term changes.

[0056] The environmental stress parameter sequence is usually composed of time series data, which records the value of environmental stress at each moment of change. These data are usually arranged in chronological order, representing the change of environmental stress over time. The environmental stress sequence is divided into P linked environmental stress slices according to certain time intervals, and each slice represents the change of environmental stress in a certain time period. These slices are linked, i.e. there is a certain overlap between the end of each slice and the beginning of the next slice. In this way, each slice is tightly connected to avoid information loss.

[0057] The time series convolution network is a deep learning model for processing time series data, which can capture long-term dependencies in time series. Here, it is used to simulate the morphology changes of defects in short-period and long-period evolution processes. The P linked environmental stress slices are used as the input of the time series convolution network, and the network iteratively evolves the morphology of the defects based on the data of these slices. For short-period evolution, the time series convolution network predicts the small morphological changes of defects according to the short-period evolution step; for long-period evolution, through the P times relationship, the time series convolution network calculates the defect morphology evolution in a longer time scale, covering the long-term accumulation effect of environmental stress. Under the driving of the time series convolution network, the defect morphology will continuously evolve, and finally output the updated state of the evolved defect morphology. The evolution result is represented by the defect morphology evolution vector, which describes the change direction and amplitude of the defect at a certain time point.

[0058] The defect morphology evolution vector is applied to the spatiotemporal coupled defect field to dynamically update the state of the defect field. Specifically, the defect morphology evolution vector provides the morphological changes of the defect at a certain time, such as crack growth, and reflects these changes in the spatiotemporal coupled defect field. The updated coupled defect field contains the defect distribution state at the current time and provides input for subsequent defect damage prediction.

[0059] Cumulative stress data is extracted from the updated coupling defect field, which represents the total stress generated at each time point under the interaction of environmental stress and material. The stress may be caused by long-term mechanical load, thermal stress, or material fatigue accumulation, etc. LSTM model is a recurrent neural network for processing time series data, which is particularly good at capturing long-term dependencies in data. LSTM model is used to calculate material performance degradation vector, which predicts the change of material performance by analyzing cumulative stress data. LSTM model is trained and inferred on cumulative stress data to generate corresponding material performance degradation vector, which represents the degree of performance degradation of the material at each time point. As the defect evolves and stress accumulates, the strength, toughness and other properties of the material gradually decrease. This material performance degradation vector can accurately reflect the state change of the material, especially under long-term stress.

[0060] Boundary conditions refer to the initial state or limiting conditions that the model relies on during calculation. In defect evolution prediction, the time series convolution network dynamically adjusts its boundary conditions according to the evolution of defect morphology and the degradation state of material performance to ensure that the model can accurately reflect the true state of the material at each time point. Through dynamic rewriting of boundary conditions and short-term and long-term updating cycles, the defect evolution model is iteratively updated until the environmental stress sequence is exhausted. Each update makes a new prediction of the morphology and damage accumulation of the defect. In this process, defect damage accumulates gradually, and eventually a complete defect evolution process is formed. Through multiple iterations, multi-stage defect damage accumulation features are output to describe the damage accumulation degree of the defect at different stages, such as initial micro-cracks, significant damage at the middle stage, and critical failure stage.

[0061] Further, according to the density attribute and defect superposition attribute of the global risk density distribution map, combined with the plurality of adaptive sensors, a non-uniform deployment of multi-source sensing nodes is performed to construct the multi-source sensing array, and the method comprises: a plurality of multi-scale density intervals and multi-scale sensor densities are predefined; the global risk density distribution map is divided into a plurality of risk area distributions according to the multi-scale density intervals; the multi-scale risk area distribution is positioned by the multi-scale sensor densities to obtain a multi-level deployment point distribution; N risk weight levels of N standard terminal surface defect types are extracted at a first deployment point; after the N standard terminal surface defect types are arranged in descending order according to the N risk weight levels, a main detection sensor and N-1 auxiliary verification sensors are called from the plurality of adaptive sensors according to the sorting result; after the main detection sensor is installed at the first deployment point, the N-1 auxiliary verification sensors are superimposed and installed; sensor deployment is performed by analogy traversal of the multi-level deployment point distribution to construct the multi-source sensing array.

[0062] Multi-scale density intervals refer to dividing risk areas into multiple different density intervals according to their risk levels and distribution characteristics, for example, dividing the density of risk areas into several ranges according to risk levels (high, medium, low), and setting corresponding threshold values for each interval. In this way, the distribution of risk areas can be refined, and a foundation is provided for subsequent deployment. Multi-scale sensor density refers to using different sensor densities in different risk areas, for example, higher density sensors are needed in high-risk areas to provide more accurate data collection, while fewer sensors are deployed in low-risk areas.

[0063] According to the predefined multi-scale density intervals, the global risk density distribution map is divided into multiple levels of areas, each level representing a different risk intensity, and the distribution range can cover large areas or smaller local areas. During the division process, each area in the map is mapped to the corresponding density interval according to its risk intensity value, for example, areas with higher risk are divided into high-risk areas, and areas with lower risk are divided into low-risk areas. This zoning method allows risk management to take different monitoring and management measures according to different risk levels.

[0064] Through multi-scale sensor density, the deployment location of the sensor is determined according to the density requirements of different risk areas. More sensor nodes are deployed in high-risk areas for high-frequency data collection and monitoring, and the number of sensors can be appropriately reduced in low-risk areas. These deployment locations cover areas of different risk levels in the global risk density distribution map.

[0065] The first deployment point is any one of the multi-level deployment point distribution, serving as the current analysis object. N standard terminal surface defect types are extracted at the first deployment point, where N is a positive integer. These standard terminal surface defect types include but are not limited to surface deformation defects, internal structure defects, and thermal stress defects. Each standard terminal surface defect type is associated with a different sensor type, such as an industrial vision camera, a terahertz imager, and an infrared thermal imager. For each standard terminal surface defect type, the risk level of these standard terminal surface defect types is evaluated through historical data, simulation experiments, or real-time monitoring results. For example, some defects can have a greater impact on the safety or performance of the terminal, so their risk level will be higher. According to the evaluation results, risk weight levels are extracted to represent the risk degree of each defect type relative to other defects.

[0066] According to the extracted N risk weight grades, the N standard terminal surface defect types are sorted in descending order of risk weight grades, and the defect types with higher risk weight grades are usually the most concerned and need to be detected first. According to the sorting result, the appropriate main detection sensor and N-1 auxiliary verification sensors are selected from the multiple adaptive sensors. The main detection sensor is responsible for core defect detection, and its performance and accuracy requirements are higher. The auxiliary verification sensor is used for auxiliary detection, provides supplementary data or verifies the detection result, and helps to improve the accuracy and reliability of detection.

[0067] At the first deployment point, the selected main detection sensor is installed first, which is responsible for the main defect detection task to ensure the accuracy of data acquisition. On the basis of the main detection sensor, N-1 auxiliary verification sensors are installed. The function of the auxiliary verification sensor is to supplement or verify the data of the main detection sensor to reduce the error of the detection result of a single sensor. According to the deployment process of the first deployment point, the same sensor deployment operation is performed on each deployment point to finally build a multi-source sensor array. This array covers all areas that need to be monitored and can accurately detect different types of defects. At the same time, the reliability of the monitoring result is improved through data fusion of multiple sensors.

[0068] Further, if the critical failure time is less than 1 / W of the total length of the environmental stress parameter sequence, a maintenance instruction is generated and sent to the terminal control unit.

[0069] W is a scale factor, which is set based on the safety margin, fault diagnosis accuracy, and other factors of the system. If the critical failure time is less than 1 / W of the total length of the environmental stress sequence, it means that the device may rapidly reach a failure state due to accumulated damage, environmental stress, and other factors in the next period of time, and immediate maintenance of the device is needed to prevent failure. In this case, a maintenance instruction is generated and sent to the terminal control unit. The terminal control unit executes specific operations according to the received maintenance instruction, including pausing the system, adjusting the operating parameters, performing emergency repair, or starting the standby system, etc.

[0070] In the second embodiment, based on the same inventive concept as the high-low orbit Ku-band satellite terminal surface defect detection method in the foregoing embodiments, as shown in FIG. 2, the present application embodiment provides a high-low orbit Ku-band satellite terminal surface defect detection system, which comprises: Figure 2 ​The sensing node deployment module 10 is configured to perform non-uniform deployment of multi-source sensing nodes on the high and low orbit Ku-band satellite terminals according to risk area distribution data, and construct a multi-source sensing array; the real-time sensing data acquisition module 20 is configured to interact with the multi-source sensing array at an interval of 0.5 seconds to acquire a real-time sensing data array in the defect detection middle station, wherein the real-time sensing data array is synchronized at a millisecond level via a Beidou timing module; the single-dimensional sensing data acquisition module 30 is configured to decompose the real-time sensing data array according to the sensor type, and obtain a plurality of single-dimensional sensing data sets with node position labels; the defect feature extraction module 40 is configured to perform defect feature extraction of the plurality of single-dimensional sensing data sets in multiple parallel threads, and obtain a plurality of single-dimensional defect feature sets; the weighted fusion decision module 50 is configured to perform confidence weighted fusion decision of the defect overlapping area after restoring the plurality of single-dimensional defect feature sets according to the node position labels, and obtain a fused defect feature set; the environmental stress parameter retrieval module 60 is configured to retrieve a pre-bound environmental stress parameter sequence according to the equipment ID of the high and low orbit Ku-band satellite terminal; and the evolution prediction module 70 is configured to perform defect spatio-temporal long and short term evolution prediction on the fused defect feature set according to the environmental stress parameter sequence, and output a critical failure time.

[0071] Further, the sensing node deployment module 10 is configured to perform the following operation steps: According to the equipment ID, the historical surface defect records of the same type of equipment are called online; according to a plurality of standard terminal surface defect types, the historical surface defect records are differentiated to obtain a plurality of high-risk area heat maps, which constitute the risk area distribution data; after matching the sensor types according to the plurality of standard terminal surface defect types, the non-uniform deployment of multi-source sensing nodes is performed by spatially fusing the plurality of high-risk area heat maps, and the multi-source sensing array is constructed.

[0072] Further, the sensing node deployment module 10 is configured to perform the following operation steps: Through defect response testing, a mapping relationship between a plurality of sample defect types and a plurality of high-response sensor types is established, and a sensor calling type library is generated; the plurality of standard terminal surface defect types are loaded into the sensor calling type library, and a plurality of adaptive sensors are matched; the spatial weighted fusion of the plurality of high-risk area heat maps is performed to obtain a global risk density distribution map; according to the density attribute and defect superposition attribute of the global risk density distribution map, in combination with the plurality of adaptive sensors, the non-uniform deployment of multi-source sensing nodes is performed, and the multi-source sensing array is constructed.

[0073] Further, the defect feature extraction module 40 is configured to perform the following operation steps: The plurality of standard terminal surface defect types and the plurality of adaptive sensors are identified as sample defect model identification retrieval conditions, and a plurality of sample defect sensor data and a plurality of sample defect identification data are retrieved; the plurality of sample defect sensor data and the plurality of sample defect identification data are used as training data to construct a plurality of lightweight feature extraction models, wherein the plurality of lightweight feature extraction models are deployed to a plurality of heterogeneous computing units; a plurality of independent computing threads are pre-created for a plurality of single-dimensional sensor data, and then the plurality of heterogeneous computing units are mapped and bound to the plurality of independent computing threads; the plurality of single-dimensional sensor data sets are distributed to the plurality of independent computing threads, the plurality of heterogeneous computing units are driven, defect feature extraction of multiple parallel threads is performed, and the plurality of single-dimensional defect feature sets are output, wherein the single-dimensional defect features have spatial coordinates.

[0074] Further, the weighted fusion decision module 50 is configured to perform the following operation steps: The plurality of recognition confidence levels are associated according to the recognition accuracy of the plurality of heterogeneous computing units; the plurality of acquisition confidence levels are associated according to the sensing performance degradation parameters of the plurality of adaptive sensors; the plurality of single-dimensional defect feature sets are projected to a terminal three-dimensional digital grid according to the node position label to obtain a global multi-dimensional defect feature distribution; and the confidence weighted fusion decision of the defect overlapping area in the global multi-dimensional defect feature distribution is performed according to the plurality of recognition confidence levels and the plurality of acquisition confidence levels to obtain a fusion defect feature set.

[0075] Further, the evolution prediction module 70 is configured to perform the following operation steps: The environment stress parameter sequence and the fusion defect feature set are aligned and mapped to the terminal three-dimensional digital grid according to the Beidou timestamp to obtain a spatiotemporal coupling defect field; a double-scale prediction engine is loaded to perform defect evolution coupling iteration of the spatiotemporal coupling defect field to output multi-stage defect damage cumulative features; a plurality of defect critical failure thresholds of the plurality of sample defect types are preset; the plurality of defect critical failure thresholds are used to traverse and compare the multi-stage defect damage cumulative features to locate the critical failure time.

[0076] Further, the evolution prediction module 70 is configured to perform the following operation steps: predefined short period evolution step and long period evolution step, wherein the long period evolution step is P times of the short period evolution step; P concatenated environmental stress slices are segmented from the environmental stress parameter sequence; the P concatenated environmental stress slices are used to drive a time convolution network in the double-scale prediction engine to perform defect morphology iterative evolution, and output a defect morphology evolution vector; the defect morphology evolution vector is used to update the spatiotemporal coupling defect field to obtain an updated coupling defect field; after extracting cumulative stress data from the updated coupling defect field, a material performance degradation vector is calculated by an LSTM model in the double-scale prediction engine; after dynamically rewriting boundary conditions of the time convolution network by using the material performance degradation vector, short period evolution and long period update are repeatedly performed until the environmental stress parameter sequence is exhausted, and the multi-stage defect damage cumulative feature is output.

[0077] Further, the sensor node deployment module 10 is configured to perform the following operation steps: predefined multi-scale density interval and multi-scale sensor density; the global risk density distribution map is divided into a multi-level risk area distribution according to the multi-scale density interval; the multi-level risk area distribution is positioned by the multi-scale sensor density to obtain a multi-level deployment point distribution; N risk weight levels of N standard terminal surface defect types are extracted at a first deployment point; after arranging the N standard terminal surface defect types in descending order according to the N risk weight levels, a main detection sensor and N-1 auxiliary verification sensors are called from the plurality of adaptive sensors according to the sorting result; after installing the main detection sensor at the first deployment point, the N-1 auxiliary verification sensors are installed in a superimposed manner; the sensor deployment is performed by analogy traversing the multi-level deployment point distribution to construct the multi-source sensor array.

[0078] Further, if the critical failure time is less than 1 / W of the total length of the environmental stress parameter sequence, a maintenance instruction is generated and sent to the terminal control unit.

[0079] Through the foregoing detailed description of the high and low orbit Ku frequency band satellite terminal surface defect detection method, those skilled in the art can clearly understand the high and low orbit Ku frequency band satellite terminal surface defect detection system in the present embodiment. Since it corresponds to the method disclosed in the embodiment, it is described relatively simply, and the relevant part is referred to the method part description.

[0080] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Therefore, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting surface defects of a high-low orbit Ku-band satellite terminal, characterized in that, The method comprises: According to the risk area distribution data, the non-uniform deployment of multi-source sensing nodes is carried out on the high-low orbit ku frequency band satellite terminal to construct a multi-source sensing array; The defect detection middle station interacts with the multi-source sensing array at an interval of 0.5 seconds to obtain a real-time sensing data array, wherein the real-time sensing data array is synchronized at a millisecond level via a Beidou timing module; According to the sensor type decomposition, the real-time sensing data array is obtained, and a plurality of single-dimensional sensing data sets with node position labels are obtained; Defect feature extraction is performed on the plurality of single-dimensional sensing data sets in multiple parallel threads to obtain a plurality of single-dimensional defect feature sets; After restoring the plurality of single-dimensional defect feature sets according to the node position labels, confidence weighted fusion decision is made on the defect overlapping area to obtain a fused defect feature set; According to the device ID of the high-low orbit ku frequency band satellite terminal, a pre-bound environmental stress parameter sequence is called; According to the environmental stress parameter sequence, the fused defect feature set is subjected to defect space-time long-term evolution prediction, and a critical failure time is output.

2. The method of claim 1, wherein the method further comprises: According to the risk area distribution data, the non-uniform deployment of multi-source sensing nodes is carried out on the high-low orbit ku frequency band satellite terminal to construct a multi-source sensing array, the method comprising: According to the device ID, the historical surface defect records of the same type of device are called online; According to a plurality of standard terminal surface defect types, the historical surface defect records are differentiated to obtain a plurality of high-risk area heat maps, which constitute the risk area distribution data; After matching the sensor types according to the plurality of standard terminal surface defect types, the non-uniform deployment of multi-source sensing nodes is carried out by spatially fusing the plurality of high-risk area heat maps to construct the multi-source sensing array.

3. The method of claim 2, wherein the method further comprises: After matching the sensor types according to the plurality of standard terminal surface defect types, the non-uniform deployment of multi-source sensing nodes is carried out by spatially fusing the plurality of high-risk area heat maps to construct the multi-source sensing array, the method comprising: Through defect response testing, a mapping relationship between a plurality of sample defect types and a plurality of high-response sensor types is established, and a sensor calling type library is generated; The plurality of standard terminal surface defect types are loaded into the sensor calling type library to match a plurality of adaptive sensors; The plurality of high-risk area heat maps are spatially weighted and fused to obtain a global risk density distribution map; According to the density attribute and defect superposition attribute of the global risk density distribution map, combined with the plurality of adaptive sensors, the non-uniform deployment of multi-source sensing nodes is carried out to construct the multi-source sensing array.

4. The method of claim 3, wherein the method further comprises: Defect feature extraction is performed on the plurality of single-dimensional sensing data sets in multiple parallel threads to obtain a plurality of single-dimensional defect feature sets, the method comprising: The plurality of standard terminal surface defect types and the plurality of adaptive sensors are used as sample defect model identification retrieval conditions to retrieve a plurality of sample defect sensing data and a plurality of sample defect identification data; The plurality of sample defect sensing data and the plurality of sample defect identification data are used as training data to construct a plurality of lightweight feature extraction models, wherein the plurality of lightweight feature extraction models are deployed to a plurality of heterogeneous computing units; After pre-creating multiple independent computing threads for multiple single-dimensional sensing data, the multiple heterogeneous computing units are mapped and bound to the multiple independent computing threads; The multiple single-dimensional sensing data sets are distributed to the multiple independent computing threads, driving the multiple heterogeneous computing units to perform multi-parallel thread defect feature extraction, and outputting the multiple single-dimensional defect feature sets, wherein the single-dimensional defect features have spatial coordinates.

5. The method of claim 4, wherein the method further comprises: After restoring the multiple single-dimensional defect feature sets according to the node position label, confidence weighted fusion decision is made for the defect overlapping area to obtain a fused defect feature set, and the method comprises: According to the recognition accuracy of the multiple heterogeneous computing units, multiple recognition confidences are associated; According to the sensing performance degradation parameters of the multiple adaptive sensors, multiple acquisition confidences are associated; According to the node position label, the multiple single-dimensional defect feature sets are projected to a terminal three-dimensional digital grid to obtain a global multi-dimensional defect feature distribution; According to the multiple recognition confidences and the multiple acquisition confidences, confidence weighted fusion decision is made for the defect overlapping area in the global multi-dimensional defect feature distribution to obtain a fused defect feature set.

6. The method of claim 5, wherein the method further comprises: According to the sequence of environmental stress parameters, defect spatio-temporal long-term and short-term evolution prediction is performed on the fused defect feature set to output a critical failure time, and the method comprises: According to the Beidou timestamp, the sequence of environmental stress parameters and the fused defect feature set are aligned and mapped to the terminal three-dimensional digital grid to obtain a spatio-temporal coupled defect field; A double-scale prediction engine is loaded to perform defect evolution coupled iteration of the spatio-temporal coupled defect field to output multi-stage defect damage accumulation characteristics; Multiple defect critical failure thresholds of the multiple sample defect types are preset; The multiple defect critical failure thresholds are used to traverse and compare the multi-stage defect damage accumulation characteristics to locate the critical failure time.

7. A method for detecting surface defects in high and low orbit Ku-band satellite terminals as described in claim 6, characterized in that, A double-scale prediction engine is loaded to perform defect evolution coupled iteration of the spatio-temporal coupled defect field to output multi-stage defect damage accumulation characteristics, and the method comprises: The long-period evolution step is P times the short-period evolution step; P concatenated environmental stress slices are segmented from the sequence of environmental stress parameters; The P concatenated environmental stress slices are used to drive the time series convolution network in the double-scale prediction engine to perform defect morphology iterative evolution to output a defect morphology evolution vector; The defect morphology evolution vector is used to update the spatio-temporal coupled defect field to obtain an updated coupled defect field; After extracting cumulative stress data from the updated coupled defect field, a material performance degradation vector is calculated through an LSTM model in the double-scale prediction engine; After dynamically rewriting the boundary conditions of the time series convolution network using the material performance degradation vector, short-period evolution and long-period update are repeatedly performed until the sequence of environmental stress parameters is exhausted, and the multi-stage defect damage accumulation characteristics are output.

8. The method for detecting surface defects in high and low orbit Ku-band satellite terminals as described in claim 3, characterized in that, According to the density attribute and defect superposition attribute of the global risk density distribution map, combined with the multiple adaptive sensors, non-uniform deployment of the multi-source sensing nodes is performed to construct the multi-source sensing array, and the method comprises: Predefined multi-scale density interval and multi-scale sensor density; dividing the global risk density distribution map into a multi-level risk area distribution according to the multi-scale density interval; sensing node positioning of the multi-level risk area distribution with the multi-scale sensor density to obtain a multi-level deployment point distribution; extracting N risk weight levels of N standard terminal surface defect types at a first deployment point; after arranging the N standard terminal surface defect types in descending order according to the N risk weight levels, calling a main detection sensor and N-1 auxiliary verification sensors from the plurality of adaptive sensors according to the sorting result; after installing the main detection sensor at the first deployment point, superimposedly installing the N-1 auxiliary verification sensors, and iteratively deploying sensors at the multi-level deployment point distribution to construct the multi-source sensing array.

9. A method for detecting surface defects in high and low orbit Ku-band satellite terminals as described in claim 7, characterized in that, If the critical failure time is less than 1 / W of the total length of the environmental stress parameter sequence, a maintenance instruction is generated and sent to the terminal control unit.

10. A high-low orbit ku-band satellite terminal surface defect detection system, characterized by, A high-low orbit ku frequency band satellite terminal surface defect detection method according to any one of claims 1-9, the system comprising: a sensing node deployment module for non-uniform deployment of multi-source sensing nodes at high-low orbit ku frequency band satellite terminals according to risk area distribution data to construct a multi-source sensing array; a real-time sensing data acquisition module for the defect detection middle station to interact with the multi-source sensing array to acquire a real-time sensing data array at an interval of 0.5 seconds, wherein the real-time sensing data array is synchronized at a millisecond level via a Beidou timing module; a single-dimensional sensing data acquisition module for decomposing the real-time sensing data array according to sensor types to obtain a plurality of single-dimensional sensing data sets with node position labels; a defect feature extraction module for performing defect feature extraction of the plurality of single-dimensional sensing data sets in multiple parallel threads to obtain a plurality of single-dimensional defect feature sets; a weighted fusion decision module for performing confidence weighted fusion decision of defect overlapping areas after restoring the plurality of single-dimensional defect feature sets according to node position labels to obtain a fused defect feature set; an environmental stress parameter calling module for calling a pre-bound environmental stress parameter sequence according to the equipment ID of the high-low orbit ku frequency band satellite terminal; an evolution prediction module for performing defect spatiotemporal long-short term evolution prediction of the fused defect feature set according to the environmental stress parameter sequence to output a critical failure time.

Images