An unmanned aerial vehicle spraying quality monitoring method and system based on distributed control

Abstract

This invention discloses a method and system for monitoring the painting quality of unmanned aerial vehicles (UAVs) based on distributed control, relating to the field of UAV painting technology. Existing technologies suffer from unstable painting quality state estimation and poor multi-UAV coordination. This invention includes: collecting painting execution data, flight status data, and operational environment data to generate local painting quality observation data; performing distributed Kalman filtering based on the observation data to obtain local painting quality prediction state data; constructing local update candidate data and introducing a secure neural network event-triggered mechanism to control communication updates; performing neighborhood data interaction when trigger conditions are met; and performing a consistent fusion update based on the neighborhood update data set to obtain painting quality state estimation data and output online monitoring results. This invention combines event-triggered communication with consistent fusion updates to achieve online consistent monitoring of the painting quality of multiple UAVs, improving the stability of state estimation and the reliability of coordination.

Description

Technical Field

[0001] This invention relates to the field of drone painting technology, and in particular to a method and system for monitoring the quality of drone painting based on distributed control. Background Technology

[0002] Multi-UAV collaborative spraying operations have been applied in scenarios such as wind turbine blade maintenance and surface treatment, anti-corrosion coating of bridge steel structures, and maintenance of storage tank exteriors. These operations typically employ a multi-UAV formation with zoned coverage, achieving continuous coating through trajectory planning, attitude control, and spraying parameter adjustment. They rely on onboard sensors to collect information such as spray gun flow rate, spray pressure, spray width, spray distance, flight speed, attitude angle, and ambient temperature, humidity, wind speed, and direction to infer coating thickness uniformity, missed areas, and the risk of overspray at boundaries. Existing technologies mainly include spraying quality monitoring schemes based on single-UAV closed-loop control and collaborative monitoring schemes based on multi-UAV communication. Single-UAV closed-loop monitoring focuses on fusing and estimating spraying execution data and flight status data locally on the UAV, using a state-space model to recursively calculate the spraying process over time, and using observation updates to correct estimation biases to output the spraying quality status. Multi-UAV collaborative monitoring focuses on sharing local estimation or observation information under neighborhood communication conditions, forming collaborative estimation results among multiple UAVs through a consistency fusion strategy, thereby achieving consistent quality monitoring at the operational level.

[0003] However, existing technologies still have several problems in online consistency monitoring of multi-UAV collaborative painting quality. First, each UAV performs independent local estimation, resulting in the painting quality state quantity being dispersed across different airborne terminals. This lack of a unified representation and fusion mechanism makes it difficult to form a unified, multi-UAV painting quality state estimate covering the entire work surface. Second, the predicted state and observed data come from state recursion models and various types of sensors, respectively, with different physical dimensions, sampling frequencies, and data dimensions. This makes it difficult to uniformly represent them into a format that can be directly used for consistency fusion, increasing the complexity of data alignment and fusion. Furthermore, distributed Kalman filtering relies on neighborhood data sets during the communication update phase. Fluctuations in the communication link and missing neighborhood data can cause incomplete update inputs, leading to instability in the fusion participation scope and making it difficult to maintain the continuity of the consistency fusion update process. In engineering applications, event-triggered communication mechanisms need to simultaneously satisfy constraints on trigger judgment input organization, field integrity, and time index consistency. If there is a lack of tight coupling with the distributed Kalman filter process, the control link between the trigger judgment result and the filter update execution may not be closed, affecting the feasibility of controlled execution of communication updates, and ultimately leading to unstable spraying quality status estimation and poor multi-machine collaboration.

[0004] Therefore, there is an urgent need to provide a drone spraying quality monitoring technology to solve the problems of unstable spraying quality status estimation and poor multi-drone coordination in the existing technology. Summary of the Invention

[0005] The technical problem to be solved and the technical task proposed by this invention is to improve and refine existing technical solutions, and to provide a method and system for monitoring the spraying quality of unmanned aerial vehicles (UAVs) based on distributed control, so as to achieve online consistency monitoring of the spraying quality of multiple UAVs and improve the stability and collaborative reliability of state estimation. To this end, this invention adopts the following technical solution.

[0006] Firstly, a method for monitoring the quality of drone painting based on distributed control is provided, which includes the following steps: S1: Real-time acquisition of spraying execution data, flight status data, and operational environment data; time synchronization and validity processing of the acquired data; and generation of local spraying quality observation data. S2: Perform a local prediction process using distributed Kalman filtering based on local spraying quality observation data to obtain local spraying quality prediction status data; S3: Based on the local spraying quality prediction status data and the local spraying quality observation data, construct local update candidate data for filtering and updating; S4: Embed the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter, using local update candidate data and historical published data as input to generate an event triggering determination result for whether to trigger a communication update; S5: Based on the event trigger determination result, the update process of the distributed Kalman filter is executed in a controlled manner. When the preset event trigger conditions are met, local update candidate data is published to the neighboring UAVs and neighboring update data is received to form a neighboring update data set. S6: Based on the neighborhood update dataset, perform a distributed Kalman filter to perform a consistent fusion update on the local spraying quality prediction status data to obtain the fused spraying quality status estimation data; S7: Output the estimated spraying quality status data as the online consistency monitoring result of multi-UAV collaborative spraying operation.

[0007] This technical solution simultaneously collects spraying execution data, flight status data, and operational environment data, and generates local spraying quality observation data through time synchronization and validity processing. This achieves a unified representation of heterogeneous data sources, providing a standardized input foundation for subsequent distributed filtering. The NN-ETM safe neural network event triggering mechanism is embedded into the communication update process of the distributed Kalman filter. Local update candidate data and historical published data are used as direct inputs for trigger determination, forming a closed-loop control between event trigger determination and filter update execution, improving the controllability and predictability of communication updates. Neighborhood data interaction is determined based on the event trigger determination result. Local update candidate data is published and neighborhood update data is received only when the trigger condition is met, effectively reducing unnecessary communication and lowering communication bandwidth usage and energy consumption in distributed collaborative scenarios. Based on the neighbor update data set, a consistent fusion update of the local spraying quality prediction state data is performed using distributed Kalman filtering. This allows each UAV to not only rely on its own sensor data but also integrate effective information from neighboring UAVs, suppressing state estimation biases caused by sensor noise, spraying parameter fluctuations, or local environmental disturbances, and enhancing the continuity, stability, and anti-disturbance capability of the estimation results. Through the distributed Kalman filter consensus fusion update in step S6 and the direct output of monitoring results in step S7, in multi-UAV collaborative operation scenarios, each UAV can collaboratively correct its local predicted state based on the neighborhood update data set, ultimately forming a unified spraying quality state estimate and outputting it online, thereby achieving consistent construction and continuous monitoring of spraying quality state among multiple UAVs. Step S4 introduces an NN-ETM secure neural network event triggering mechanism, and step S5 executes communication updates in a controlled manner based on the trigger judgment result, ensuring that communication interaction only occurs when necessary, avoiding communication congestion and latency jitter caused by fixed-period broadcasting, reducing the impact of link fluctuations on the integrity of the neighborhood data set, and thus improving the communication reliability and overall collaborative reliability of the distributed collaborative system in complex operating environments. Each UAV obtains its own scattered local spraying quality prediction status data through local prediction; it acquires updated data from neighboring UAVs through event-triggered communication, forming a neighborhood updated data set; then, it uses distributed Kalman filtering for consistent fusion updates to weightedly fuse the local predicted status with the neighborhood information, making the state estimates of multiple UAVs tend to be consistent, thereby fusing the quality state quantities scattered at each UAV into unified, collaborative spraying quality state estimation data; finally, the output is a shared online consistent monitoring result for multiple UAVs, solving the problem of scattered state quantities, forming a consistent quality state estimate for multiple UAVs, supporting global optimization scheduling of collaborative operations, and avoiding spraying conflicts or missed spraying caused by inconsistent estimates.

[0008] As a preferred technical means, step S1 specifically includes: S11: Collect spraying execution data, flight status data, and work environment data according to a unified sampling rhythm, and write a collection time identifier for each collection record to obtain raw data with collection time identifier; S12: Based on the acquisition time identifier, perform time synchronization, validity verification, missing data filling, and normalization on the raw data to obtain standardized observation feature data; S13: The standardized observation feature data is spliced ​​and encapsulated according to the field order of spraying execution data, flight status data, and operation environment data to generate local spraying quality observation data.

[0009] This technical solution collects multi-source data using a unified sampling rhythm and adds sampling time markers to ensure the temporal alignment of the data from the source, avoiding subsequent processing errors caused by asynchronous sampling from different sensors. Through a full-process data preprocessing process including time synchronization, validity verification, missing data filling, and normalization, outlier data is removed, data gaps are filled, and data units are standardized, significantly improving the quality and reliability of the original observation data. The data is then spliced ​​and encapsulated according to a fixed field order to generate local spraying quality observation data, achieving structured integration of multi-source data and providing standardized and consistent input data for the subsequent local prediction process of distributed Kalman filtering.

[0010] As a preferred technical means, step S2 specifically includes: S21: Based on local spraying quality observation data, read the spraying execution data, flight status data, and operation environment data corresponding to the current moment, and construct the prediction calculation input data for distributed Kalman filtering; S22: Based on the input data for prediction calculation, generate local initial prediction data for distributed Kalman filtering to obtain local initial prediction data; S23: Based on the local predicted initial value data, perform distributed Kalman filtering state-time recursive processing, write the motion change information represented by the flight state data into the state recursive process, write the spraying process information represented by the spraying execution data into the state recursive process, and obtain the local recursive predicted state data. S24: Perform distributed Kalman filtering prediction consistency verification on the local recursive prediction state data based on the work environment data, identify the state components that are inconsistent with the work environment data, and perform replacement and update processing on the inconsistent components to obtain the verified prediction state data. S25: Perform observation consistency pre-check processing of distributed Kalman filtering on the verified predicted state data, calculate the difference information between the verified predicted state data and the local spraying quality observation data, and write the difference information into the predicted state data to obtain the predicted state data carrying the difference information. S26: Perform numerical stabilization processing of the predicted state data carrying difference information using distributed Kalman filtering to complete outlier removal and range constraint processing, and obtain stable local spraying quality predicted state data. S27: Output stable local spraying quality prediction status data.

[0011] This technical solution ensures the accuracy and relevance of the prediction input by constructing targeted input data for distributed Kalman filtering. During the state-time recursion process, it simultaneously integrates motion change information from the flight state with information from the spraying execution process, enabling the prediction model to accurately reflect the coupling relationship between the UAV's motion and the spraying process, thus improving the physical rationality of the state prediction. It introduces operational environment data for prediction consistency verification, correcting the impact of environmental factors on spraying quality and eliminating prediction biases caused by environmental disturbances. Through observation consistency pre-detection, it identifies discrepancies between prediction and observation in advance, providing a basis for subsequent candidate data construction and event trigger determination. Numerical stabilization processing is performed to remove abnormal prediction values ​​and constrain state components within a physically feasible range, preventing filter divergence and ensuring the stability and reliability of the local spraying quality prediction state data.

[0012] As a preferred technical means, step S3 specifically includes: S31: Based on local spraying quality prediction status data and local spraying quality observation data, and performing time index consistency verification on the two types of data, a data input pair with consistent time is obtained; S32: Based on time-consistent data input pairs, perform field mapping processing on the state components in the local spraying quality prediction state data and the observation components in the local spraying quality observation data to establish a one-to-one correspondence between the state components and the observation components, and obtain the field-aligned prediction state data and the field-aligned observation data. S33: Based on the field-aligned predicted state data and the field-aligned observed data, calculate the difference information between the predicted state component and the corresponding observed component item by item to form a set of difference information; S34: Perform consistency filtering on the set of difference information, identify the difference components that do not meet the constraints according to the preset difference constraints, and remove the difference components that do not meet the constraints to obtain the filtered set of difference information. S35: The field-aligned predicted state data, field-aligned observation data, and filtered difference information set are structured and encapsulated, and local update candidate data is generated in a fixed order of state data, observation data, and difference information. S36: Output local update candidate data and use the local update candidate data as input data for the event trigger determination in step S4.

[0013] This technical solution ensures strict alignment between predicted state data and observed data in the time dimension through time index consistency verification, avoiding calculation errors caused by temporal misalignment. It establishes a one-to-one correspondence between state components and observed components through field mapping, resolving the issues of dimensional mismatch and physical inconsistency between the state space and observation space. It calculates and filters differences item by item, eliminating invalid difference components and retaining valid information that accurately reflects prediction deviations, thus improving the accuracy of subsequent event trigger determination. Finally, it performs structured encapsulation in a fixed order to generate local update candidate data, achieving standardized data organization and ensuring that neighboring UAVs can accurately parse the received data, providing a unified and effective input carrier for subsequent event trigger determination and consistency fusion.

[0014] As a preferred technical means, step S4 specifically includes: S41: Introduce the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter, and set the NN-ETM secure neural network event triggering mechanism as the determination method for communication update execution; S42: Use locally updated candidate data as input data for the event triggering determination method; S43: Perform time index consistency check and field integrity check on the locally updated candidate data to obtain candidate judgment data; S44: Based on the candidate decision data, according to the preset input rules in the NN-ETM secure neural network event triggering mechanism, perform decision input construction processing on the candidate decision data to generate event trigger decision input data; S45: Based on the NN-ETM secure neural network event triggering mechanism, perform event triggering judgment processing on the event triggering judgment input data to obtain the communication update triggering judgment result; S46: Output the communication update trigger determination result.

[0015] This technical solution embeds a neural network event triggering mechanism into the communication update stage of Kalman filtering to ensure the closure of the control link. By performing time index consistency checks and field integrity checks on locally updated candidate data, invalid input data is eliminated to avoid false triggering or missed triggering caused by erroneous data. Standardized judgment input data is constructed according to the input rules preset by NN-ETM to ensure the standardization and consistency of neural network input, thereby improving the accuracy and stability of event triggering judgment. By performing event triggering judgment through the NN-ETM safe neural network, it can adaptively learn the characteristics of changes in spraying quality status, which has higher judgment accuracy and robustness compared with traditional threshold triggering methods.

[0016] As a preferred technical means, step S5 specifically includes: S51: Based on the communication update trigger determination result and local update candidate data, form communication update control input data; S52: Based on the communication update control input data, perform communication update condition judgment processing; if the communication update trigger judgment result indicates that communication update is triggered, generate communication update execution data; if the communication update trigger judgment result indicates that communication update is not triggered, generate the neighborhood update data set initialization result. S53: After the communication update execution data is generated, send local update candidate data to neighboring UAVs and generate a sending completion record to obtain the published local update candidate data; S54: Based on the published local update candidate data, receive neighborhood update data from the neighboring drones and perform reception registration processing on the neighborhood update data to obtain a neighborhood update data set; S55: Perform time index consistency check and field integrity check on the neighborhood update data set, delete data records that do not meet the time index consistency requirement and delete data records that do not meet the field integrity requirement, and obtain the verified neighborhood update data set; S56: Output the set of neighborhood update data after verification.

[0017] This technical solution achieves precise and controlled execution of communication updates through communication update condition judgment processing, initiating data interaction only when trigger conditions are met to avoid invalid communication; it achieves traceability of the communication process and ensures the integrity of data interaction by generating a send completion record and receiving registration processing; it performs time index consistency verification and field integrity verification on the received neighborhood update data, eliminating invalid data that is timed out, missing, or formatted incorrectly, ensuring the timeliness and validity of the neighborhood update data set, avoiding interference from invalid data in the subsequent consistency fusion update process, and improving the reliability of the fusion result.

[0018] As a preferred technical means, step S6 specifically includes: S61: Based on the local spraying quality prediction status data and the verified neighborhood update data set, a consistent fusion update input data is formed; S62: Perform time index alignment and field alignment processing on the consistent fusion update input data to obtain the aligned local prediction state data and the aligned neighborhood update data set; S63: Perform fusion participation verification processing based on the aligned neighborhood update data set, identify data records that do not meet the fusion participation conditions and perform removal processing to obtain a valid neighborhood update data set; S64: Based on the effective neighborhood update data set, generate fusion update constraint data according to the distributed Kalman filter consensus fusion update rule; S65: Based on the fusion update constraint data, the effective neighborhood update data set is written into the aligned local prediction state data fusion update process, and distributed Kalman filter consistency fusion update processing is performed to obtain the fused spraying quality state estimation data. S66: Output the estimated spray quality status data after fusion.

[0019] This technical solution ensures complete matching between local predicted state data and neighborhood update data in terms of time dimension and physical meaning through time index alignment and field alignment, laying the foundation for consistent fusion. By participating in the verification process during fusion, valid neighborhood data that meets the requirements is further filtered out, eliminating interference from low-quality or abnormal data and improving the robustness of the fusion process. Fusion update constraint data is generated based on the valid neighborhood data, enabling the fusion process to dynamically adjust the fusion weights according to the quality of the neighborhood data, ensuring the rationality and accuracy of the fusion results. Distributed Kalman filtering consistent fusion update is performed to globally align the local estimation results of multiple drones, effectively eliminating estimation biases between different drones and obtaining globally consistent spraying quality state estimation data.

[0020] As a preferred technical means, step S7 specifically includes: S71: Based on the fused spraying quality status estimation data, generate input data to form a consistent monitoring result; S72: Generate input data for consistency monitoring results, perform time index consistency verification and field integrity verification, and generate candidate data from the results; S73: Generate candidate data based on the results, perform output field mapping processing on the state components in the generated candidate data, establish a one-to-one correspondence between the state components and the output fields of the online consistency monitoring results, and obtain the field-aligned result data; S74: Perform structured encapsulation processing on the field-aligned result data, and generate online consistency monitoring result output data according to the fixed order of time index and output fields; S75: Perform output consistency verification on the online consistency monitoring result output data, delete data records that do not meet the time index consistency requirements and data records that do not meet the field integrity requirements, and obtain the verified online consistency monitoring result output data.

[0021] This technical solution ensures the validity of the input data for generating the results by performing time index consistency checks and field integrity checks on the fused state estimation data; through output field mapping processing, the abstract state components are converted into monitoring result fields that can be directly understood in engineering, realizing the connection between technical data and application requirements; structured encapsulation is performed in a fixed order to generate standardized online consistent monitoring result output data, which is convenient for subsequent system parsing, storage and display; through output consistency verification processing, invalid result data is further eliminated to ensure that the final output monitoring results are accurate, complete and consistent.

[0022] Secondly, a distributed control-based UAV painting quality monitoring system is provided, comprising: The data acquisition module is used to collect spraying execution data, flight status data and work environment data in real time, and to perform time synchronization and validity processing on the collected data to generate local spraying quality observation data. The local prediction module is used to perform a local prediction process using distributed Kalman filtering based on local spraying quality observation data to obtain local spraying quality prediction status data. The candidate building module is used to construct local update candidate data for filtering and updating based on local spraying quality prediction status data and local spraying quality observation data. The event determination module is used to embed the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter. It takes local update candidate data and historical published data as input to generate an event triggering determination result on whether to trigger a communication update. The communication control module is used to control the update process of the distributed Kalman filter based on the event trigger determination result. When the preset event trigger conditions are met, it publishes local update candidate data to neighboring UAVs and receives neighboring update data to form a neighboring update data set. The consistency fusion module is used to perform a distributed Kalman filter consistency fusion update on the local spraying quality prediction status data based on the neighborhood update data set, so as to obtain the fused spraying quality status estimation data. The result output module is used to output the spraying quality status estimation data as the online consistency monitoring result of multi-UAV collaborative spraying operation.

[0023] This technical solution utilizes the cooperation of a local prediction module and a consistency fusion module. After each UAV completes its local state prediction, the communication control module and the consistency fusion module enable controlled interaction and fusion of neighborhood information. Finally, the result output module generates a unified quality monitoring result, supporting consistent quality monitoring in multi-UAV distributed collaborative operation scenarios. The event determination module embeds an NN-ETM safe neural network event triggering mechanism. The communication control module only sends local data to the neighborhood when a communication update is triggered; otherwise, it only initializes the neighborhood data set. This significantly reduces unnecessary communication, lowers the communication bandwidth usage and energy consumption of the distributed system, and avoids network congestion caused by fixed-period broadcasts. The data acquisition module performs time synchronization, validity verification, missing data filling, and normalization. The candidate construction module performs time index consistency verification and field mapping. The event determination module performs time index and field integrity verification. The communication control module outputs the verified neighborhood update data set. The consistency fusion module performs time index alignment, field alignment, and fusion participation verification. The result output module performs consistency verification again. Multi-level verification processing runs through the entire system process, effectively eliminating abnormal data and ensuring the correctness of the fused input and the reliability of the output results. The consistency fusion module performs consistent fusion updates using distributed Kalman filtering based on local predicted state data and the verified neighborhood update data set. By fusing effective information from multiple UAVs, it suppresses state estimation bias caused by sensor noise or local environmental disturbances in a single UAV, thereby enhancing the continuity, stability, and anti-interference capability of the coating quality state estimation.

[0024] As a preferred technical approach, candidate building modules include: The time verification unit is used to perform time index consistency verification between the local spraying quality prediction status data and the local spraying quality observation data to obtain data input pairs with consistent time. The field mapping unit is used to perform field mapping processing on the state components and observation components in the data input pair, and establish a one-to-one correspondence between the state components and the observation components. The difference calculation unit is used to calculate the difference information between the predicted state component and the corresponding observed component item by item, forming a difference information set. The consistency filtering unit is used to identify and remove differential components that do not meet the preset differential constraints, and obtain a set of filtered differential information. The encapsulation unit is used to structurally encapsulate the field-aligned predicted state data, the field-aligned observation data, and the filtered set of difference information, and generate local update candidate data in a fixed order of state data, observation data, and difference information.

[0025] This technical solution ensures the accuracy and standardization of the local update candidate data construction process through the design of dedicated units such as time verification units and field mapping units.

[0026] Beneficial effects: 1. This invention employs distributed Kalman filtering to perform local prediction and consistency fusion of spraying execution data, flight status data, and operational environment data, thereby achieving continuous estimation and collaborative updating of the spraying quality status of multiple UAVs and effectively improving the temporal consistency and state stability of spraying quality monitoring results.

[0027] 2. This invention introduces a secure neural network event triggering mechanism into the communication update process to control the execution of the update process, reduce unnecessary data interaction, and optimize the communication load and update efficiency in distributed collaborative scenarios.

[0028] 3. This invention constructs local update candidate data and performs time indexing and field consistency verification on neighboring update data to achieve structured and constrained processing of fused input data, thereby enhancing the reliability of the distributed fusion update process.

[0029] 4. The present invention ultimately produces online consistent monitoring results that can be directly output, effectively supporting the continuous monitoring and unified evaluation of the spraying quality status during multi-UAV collaborative spraying operations. Attached Figure Description

[0030] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the process of the present invention.

[0031] Figure 2 This is a schematic diagram of the event-triggered distributed consistency update mechanism of the present invention. Detailed Implementation

[0032] The present invention will be further described in detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention and showing the main components related to the invention.

[0033] Example 1 This embodiment provides a method for monitoring the quality of drone painting based on distributed control, such as... Figure 1 , Figure 2 As shown, it includes the following steps: S1: Real-time acquisition of spraying execution data, flight status data, and operational environment data; time synchronization and validity processing of the acquired data; generating local spraying quality observation data; specifically including the following steps: S11: Collect spraying execution data, flight status data, and work environment data according to a unified sampling rhythm, and write a collection time identifier for each collection record to obtain raw data with collection time identifier; S12: Based on the acquisition time identifier, perform time synchronization, validity verification, missing data filling, and normalization on the raw data to obtain standardized observation feature data; S13: The standardized observation feature data is spliced ​​and encapsulated according to the field order of spraying execution data, flight status data, and operation environment data to generate local spraying quality observation data.

[0034] S2: Perform a local prediction process using distributed Kalman filtering based on local spraying quality observation data to obtain local spraying quality prediction status data; specifically including the following steps: S21: Based on local spraying quality observation data, read the spraying execution data, flight status data, and operation environment data corresponding to the current moment, and construct the prediction calculation input data for distributed Kalman filtering; S22: Based on the input data for prediction calculation, generate local initial prediction data for distributed Kalman filtering to obtain local initial prediction data; S23: Based on the local predicted initial value data, perform distributed Kalman filtering state-time recursive processing, write the motion change information represented by the flight state data into the state recursive process, write the spraying process information represented by the spraying execution data into the state recursive process, and obtain the local recursive predicted state data. S24: Perform distributed Kalman filtering prediction consistency verification on the local recursive prediction state data based on the work environment data, identify the state components that are inconsistent with the work environment data, and perform replacement and update processing on the inconsistent components to obtain the verified prediction state data. S25: Perform observation consistency pre-check processing of distributed Kalman filtering on the verified predicted state data, calculate the difference information between the verified predicted state data and the local spraying quality observation data, and write the difference information into the predicted state data to obtain the predicted state data carrying the difference information. S26: Perform numerical stabilization processing of the predicted state data carrying difference information using distributed Kalman filtering to complete outlier removal and range constraint processing, and obtain stable local spraying quality predicted state data. S27: Output stable local spraying quality prediction status data.

[0035] S3: Based on the local spraying quality prediction status data and the local spraying quality observation data, construct local update candidate data for filtering updates; specifically including the following steps: S31: Based on local spraying quality prediction status data and local spraying quality observation data, and performing time index consistency verification on the two types of data, a data input pair with consistent time is obtained; S32: Based on time-consistent data input pairs, perform field mapping processing on the state components in the local spraying quality prediction state data and the observation components in the local spraying quality observation data to establish a one-to-one correspondence between the state components and the observation components, and obtain the field-aligned prediction state data and the field-aligned observation data. S33: Based on the field-aligned predicted state data and the field-aligned observed data, calculate the difference information between the predicted state component and the corresponding observed component item by item to form a set of difference information; S34: Perform consistency filtering on the set of difference information, identify the difference components that do not meet the constraints according to the preset difference constraints, and remove the difference components that do not meet the constraints to obtain the filtered set of difference information. S35: The field-aligned predicted state data, field-aligned observation data, and filtered difference information set are structured and encapsulated, and local update candidate data is generated in a fixed order of state data, observation data, and difference information. S36: Output local update candidate data and use the local update candidate data as input data for the event trigger determination in step S4.

[0036] S4: Embed the NN-ETM (Neural Network Event-Triggered Mechanism) secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter. Using local update candidate data and historical published data as input, generate an event triggering determination result to determine whether to trigger a communication update. Specifically, this includes the following steps: S41: Introduce the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter, and set the NN-ETM secure neural network event triggering mechanism as the determination method for communication update execution; S42: Use locally updated candidate data as input data for the event triggering determination method; S43: Perform time index consistency check and field integrity check on the locally updated candidate data to obtain candidate judgment data; S44: Based on the candidate decision data, according to the preset input rules in the NN-ETM secure neural network event triggering mechanism, perform decision input construction processing on the candidate decision data to generate event trigger decision input data; S45: Based on the NN-ETM secure neural network event triggering mechanism, perform event triggering judgment processing on the event triggering judgment input data to obtain the communication update triggering judgment result; S46: Output the communication update trigger determination result.

[0037] S5: Based on the event trigger determination result, the update process of the distributed Kalman filter is executed in a controlled manner. When the preset event trigger conditions are met, local update candidate data is published to neighboring drones and neighboring update data is received, forming a neighboring update data set; specifically including the following steps: S51: Based on the communication update trigger determination result and local update candidate data, form communication update control input data; S52: Based on the communication update control input data, perform communication update condition judgment processing; if the communication update trigger judgment result indicates that communication update is triggered, generate communication update execution data; if the communication update trigger judgment result indicates that communication update is not triggered, generate the neighborhood update data set initialization result. S53: After the communication update execution data is generated, send local update candidate data to neighboring UAVs and generate a sending completion record to obtain the published local update candidate data; S54: Based on the published local update candidate data, receive neighborhood update data from the neighboring drones and perform reception registration processing on the neighborhood update data to obtain a neighborhood update data set; S55: Perform time index consistency check and field integrity check on the neighborhood update data set, delete data records that do not meet the time index consistency requirement and delete data records that do not meet the field integrity requirement, and obtain the verified neighborhood update data set; S56: Output the set of neighborhood update data after verification.

[0038] S6: Based on the neighborhood update dataset, perform a distributed Kalman filter consensus fusion update on the local spraying quality prediction status data to obtain the fused spraying quality status estimate data; specifically including the following steps: S61: Based on the local spraying quality prediction status data and the verified neighborhood update data set, a consistent fusion update input data is formed; S62: Perform time index alignment and field alignment processing on the consistent fusion update input data to obtain the aligned local prediction state data and the aligned neighborhood update data set; S63: Perform fusion participation verification processing based on the aligned neighborhood update data set, identify data records that do not meet the fusion participation conditions and perform elimination processing. The fusion participation conditions are preset according to the input data to obtain an effective neighborhood update data set. S64: Based on the effective neighborhood update data set, generate fusion update constraint data according to the distributed Kalman filter consensus fusion update rule; S65: Based on the fusion update constraint data, the effective neighborhood update data set is written into the aligned local prediction state data fusion update process, and distributed Kalman filter consistency fusion update processing is performed to obtain the fused spraying quality state estimation data. S66: Output the estimated spray quality status data after fusion.

[0039] S7: Output the estimated spraying quality status data as the online consistency monitoring result of multi-UAV collaborative spraying operation, specifically including the following steps: S71: Based on the fused spraying quality status estimation data, generate input data to form a consistent monitoring result; S72: Generate input data for consistency monitoring results, perform time index consistency verification and field integrity verification, and generate candidate data from the results; S73: Generate candidate data based on the results, perform output field mapping processing on the state components in the generated candidate data, establish a one-to-one correspondence between the state components and the output fields of the online consistency monitoring results, and obtain the field-aligned result data; S74: Perform structured encapsulation processing on the field-aligned result data, and generate online consistency monitoring result output data according to the fixed order of time index and output fields; S75: Perform output consistency verification on the online consistency monitoring result output data, delete data records that do not meet the time index consistency requirements and data records that do not meet the field integrity requirements, and obtain the verified online consistency monitoring result output data.

[0040] To verify the feasibility of this invention in practice, a verification was conducted in a scenario of online consistency monitoring of spraying quality during multi-UAV collaborative spraying operations. During the operation, multiple UAVs in a preset formation spray the same work surface. Spraying execution data continuously reflects process variables such as spraying valve opening, spraying flow rate, spraying pressure, spraying width, and spraying speed. Flight status data continuously reflects motion variables such as position, speed, attitude angle, and attitude angular velocity. Environmental data continuously reflects environmental variables such as wind speed, wind direction, temperature, and humidity. In this embodiment, spraying execution data, flight status data, and environmental data are collected in real time at the edge of each UAV. The collected data undergoes time synchronization and validity processing to generate local spraying quality observation data. Controlled execution of communication updates is achieved through a distributed Kalman filter method combined with an NN-ETM secure neural network event triggering mechanism. Then, a consistent fusion update using distributed Kalman filtering is used to form fused spraying quality state estimation data. Finally, the spraying quality state estimation data is output as the online consistency monitoring result of the multi-UAV collaborative spraying operation.

[0041] In the same scenario, existing methods often employ a combination of single-machine edge filtering and fixed-period broadcasting for collaborative monitoring. Fixed-period broadcasting is prone to communication congestion and latency jitter when the working environment changes abruptly or spraying execution fluctuates significantly, leading to inconsistencies in time indices and missing fields in the neighborhood update data set, thus disrupting the input conditions for the fusion update process. Simultaneously, single-machine edge filtering lacks consistency constraints when neighborhood information is unavailable or quality is unstable, easily resulting in accumulated quality estimation biases from different UAVs for the same work surface, making it difficult to generate output results directly usable for online consistency monitoring. This invention addresses these issues by embedding an NN-ETM safe neural network event triggering mechanism into the communication update process of a distributed Kalman filter. Local update candidate data drives the event triggering determination of whether to trigger a communication update. When the communication update triggering determination indicates that a communication update is to be triggered, local update candidate data is published, and neighborhood update data is received to form a verified neighborhood update data set. Then, a consistent fusion update rule is used to perform a fusion update on the local spraying quality prediction status data, ensuring that multi-UAV edge estimations are aligned at the time index and field level, can be filtered at the fusion participation condition level, and are consistent at the fusion constraint level.

[0042] During implementation, a unified sampling rate of 10 samples per second was set. The collected records were marked with a sampling time identifier to ensure time synchronization and validity verification. Missing records were supplemented using linear interpolation of two adjacent valid records, with supplementary value markers retained. Normalization was performed using a sliding window statistic to transform each field to zero mean and unit variance. The sliding window length was set to 60 sampling points to obtain standardized observation feature data. This data was then concatenated and packaged according to the field order of spraying execution data, flight status data, and operational environment data to generate local spraying quality observation data. In this embodiment, the field set of the local spraying quality observation data was limited to spraying valve opening, spraying flow rate, spraying pressure, spraying speed, position, velocity, attitude angle, wind speed, wind direction, temperature, and humidity, ensuring a unique input for subsequent field mapping and alignment processes.

[0043] The local prediction process of distributed Kalman filtering takes local spraying quality observation data as input. It first reads the spraying execution data, flight status data, and operational environment data corresponding to the current moment to construct the prediction calculation input data, and then regenerates the initial local prediction data. In this embodiment, the initial local prediction data consists of a state vector, state covariance, process noise covariance, and observation noise covariance. The state vector is set to 8 dimensions, including the implicit state component of spraying quality and the motion-coupled state component. The initial state covariance is set as a diagonal matrix with diagonal elements of 0.25, the process noise covariance is set as a diagonal matrix with diagonal elements of 0.01, and the observation noise covariance is set as a diagonal matrix with diagonal elements of 0.02. Subsequently, a state-time recursive processing is performed, writing the motion change information represented by the flight status data into the state recursive process, and writing the spraying process information represented by the spraying execution data into the state recursive process, thus obtaining the local recursive prediction state data. In this embodiment, the state transition matrix is ​​discretized in the state recursion with a sampling interval of 0.1s. The spraying process information is written into the control input channel as a joint term of spraying flow rate and spraying pressure. Then, a prediction consistency check is performed based on the operating environment data. State components inconsistent with the operating environment data are identified, and replacement and update processing is performed on these inconsistent components. In this embodiment, the drift term caused by wind speed and direction is used as the environmental consistency constraint object. When the deviation between the drift term and the corresponding field of the operating environment data exceeds three times the standard deviation, a replacement and update is triggered. Further observation consistency pre-check processing is performed, calculating the difference information between the checked predicted state data and the local spraying quality observation data and writing it into the predicted state data. The difference information is expressed using a field-by-field residual vector. Finally, numerical stabilization processing is performed to complete outlier removal and range constraint processing. Outlier removal uses the median absolute deviation criterion, and range constraints limit the spraying quality-related state components to a physically feasible range, resulting in stable local spraying quality predicted state data.

[0044] When constructing local update candidate data, a time index consistency check is first performed on the local spraying quality prediction state data and the local spraying quality observation data to obtain time-consistent data input pairs. Then, a field mapping process is performed on the state components and observation components to establish a one-to-one correspondence, resulting in field-aligned prediction state data and field-aligned observation data. In this embodiment, the mapping table is limited to the combined observation components of spraying valve opening, spraying flow rate, spraying pressure, and spraying speed corresponding to the spraying quality-related state components, and the motion-coupled state components corresponding to the position, velocity, and attitude angle observation components. Subsequently, the difference information between the prediction state components and the corresponding observation components is calculated item by item to form a difference information set. The consistency screening process identifies the difference components that do not meet the constraints based on preset difference constraints and performs elimination processing. In this embodiment, the difference constraint is set to a residual normalization value not exceeding 2.5. The filtered difference information set, the field-aligned prediction state data, and the field-aligned observation data are structured and encapsulated, and local update candidate data is generated in a fixed order.

[0045] After embedding the NN-ETM secure neural network event triggering mechanism into the communication update process, the local update candidate data is used as input data for the event triggering determination method. This input data undergoes time index consistency verification and field integrity verification to obtain candidate determination data. Then, the event triggering determination input data is constructed according to the preset input rules in the NN-ETM secure neural network event triggering mechanism. In this embodiment, the input rules are limited to a concatenated vector of state data, observation data, and difference information, supplemented with a summary feature of the published record from the previous time step. The summary feature of the published record consists of the residual mean, residual variance, and publication interval from the previous time step. The NN-ETM secure neural network event triggering mechanism adopts a 3-layer feedforward structure, with hidden layer widths set to 32 and 16, a hyperbolic tangent function as the activation function, and a binary classification output to generate the communication update triggering determination result. The mechanism training was completed using historical offline data. In this embodiment, 60,000 candidate decision data were used for training, the training rounds were set to 120, the batch size was set to 256, the learning rate was set to 0.001, the optimizer used adaptive moment estimation, and the event trigger decision threshold was determined by the criterion of minimizing the comprehensive index of communication trigger rate and fusion error on the validation set, resulting in a threshold of 0.55.

[0046] After the communication update trigger determination result is output, the communication update condition judgment process is entered. When the communication update trigger determination result indicates that a communication update has been triggered, communication update execution data is generated and local update candidate data is published to neighboring UAVs. At the same time, a transmission completion record is generated to obtain the published local update candidate data. When the communication update trigger determination result indicates that a communication update has not been triggered, the initialization result of the neighboring update data set is generated and the publication of this cycle is terminated.

[0047] After receiving the published data, the neighboring drone returns neighboring update data. In this embodiment, the neighboring update data is received and registered to form a neighboring update data set. Then, time index consistency and field integrity checks are performed on the neighboring update data set. Data records that do not meet the time index consistency requirement are deleted, and data records that do not meet the field integrity requirement are also deleted, resulting in a verified neighboring update data set. In the consistency fusion update stage, the local spraying quality prediction status data and the verified neighboring update data set are first used to form the consistency fusion update input data. Then, time index alignment and field alignment are performed to obtain aligned local prediction status data and aligned neighboring update data sets. Subsequently, fusion participation verification processing is performed to identify data records that do not meet the fusion participation conditions and remove them to obtain a valid neighboring update data set. In this embodiment, the fusion participation conditions are limited to a publication interval of no more than 2 seconds, a residual normalized value of no more than 3.0, and the field set being consistent with the local mapping table. Based on the effective neighborhood update data set, fusion update constraint data is generated according to the distributed Kalman filter consistency fusion update rule. The fusion update constraint data consists of a neighborhood weight set and a covariance consistency upper bound. The neighborhood weight set is obtained by normalizing the inverse of the neighborhood residual variance, and the covariance consistency upper bound is set to 0.8. Finally, the effective neighborhood update data set is written into the fusion update process of the aligned local predicted state data to perform consistency fusion update processing, resulting in fused spraying quality state estimation data. The fused spraying quality state estimation data is then input into the consistency monitoring result generation process to complete time index consistency verification, field integrity verification, output field mapping and structured encapsulation, and finally, output consistency verification is performed to obtain the online consistency monitoring result output data.

[0048] This embodiment uses a fixed-period communication update method as a control, with the fixed period set to broadcast once every 0.5 seconds, without using the NN-ETM secure neural network event triggering mechanism. Evaluation metrics include the mean absolute error, root mean square error, relative percentage error, cross-UAV consistency deviation, and communication trigger rate of the estimated spray quality status data relative to the measured spray quality observations. The cross-UAV consistency deviation is defined as the average difference between the maximum and minimum values ​​of the output fields of different UAVs under the same time index. Experimental results show that during periods of heightened environmental disturbance, the communication trigger rate of the proposed method is approximately 0.42, significantly lower than the equivalent communication trigger rate of 1.00 for the fixed-period method. Simultaneously, the cross-UAV consistency deviation decreases from 0.08 to 0.03, and the mean absolute error decreases from 0.11 to 0.06. This demonstrates that controlled communication updates and consistency fusion updates can maintain the consistency and stability of the spray quality estimation while reducing the frequency of communication updates.

[0049] Table 1 Comparison of Key Indicators of Online Consistency Monitoring Results

[0050] As shown in Table 1, the predicted and measured values ​​of film thickness deviation maintained a consistent trend across the five samples, with a maximum single-point difference of 0.01. This indicates that the local prediction process of the distributed Kalman filter can form stable local spraying quality prediction data after incorporating flight status data and spraying execution data, and maintains the traceability of film thickness deviation after consistent fusion update. The predicted values ​​of coverage and uniformity showed a high degree of fit with the measured values. In Sample 4, the changes in coverage and uniformity, corresponding to the increase in missed spray rate, were synchronously reflected, indicating that after the fusion participation verification and removal of neighboring records that did not meet the conditions, the participation of the effective neighboring update data set in the fusion update can suppress the disturbance of inconsistent neighboring information to the estimation. The predicted and measured values ​​of missed spray rate and overspray rate showed good consistency, reflecting that the controlled communication update driven by the event trigger judgment result can still ensure the continuity of online monitoring and the stability of consistent output of key quality fields even when the communication trigger rate decreases. Based on calculations using five samples, the mean absolute error of this embodiment is 0.057, the root mean square error is 0.072, the relative percentage error is 0.63, and the cross-UAV consistency deviation is 0.03, which meets the requirements of online consistency monitoring for stable output.

[0051] Example 2 This embodiment provides a drone painting quality monitoring system based on distributed control, which includes the following modules: The data acquisition module is used to collect spraying execution data, flight status data and work environment data in real time, and to perform time synchronization and validity processing on the collected data to generate local spraying quality observation data. The local prediction module is used to perform a local prediction process using distributed Kalman filtering based on local spraying quality observation data to obtain local spraying quality prediction status data. The candidate building module is used to construct local update candidate data for filtering and updating based on local spraying quality prediction status data and local spraying quality observation data. The event determination module is used to embed the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter. It takes local update candidate data and historical published data as input to generate an event triggering determination result on whether to trigger a communication update. The communication control module is used to control the update process of the distributed Kalman filter based on the event trigger determination result. When the preset event trigger conditions are met, it publishes local update candidate data to neighboring UAVs and receives neighboring update data to form a neighboring update data set. The consistency fusion module is used to perform a distributed Kalman filter consistency fusion update on the local spraying quality prediction status data based on the neighborhood update data set, so as to obtain the fused spraying quality status estimation data. The result output module is used to output the spraying quality status estimation data as the online consistency monitoring result of multi-UAV collaborative spraying operation.

[0052] The candidate construction module includes: a time verification unit, used to perform time index consistency verification on the local spraying quality prediction status data and the local spraying quality observation data to obtain time-consistent data input pairs; a field mapping unit, used to perform field mapping processing on the state components and observation components in the data input pairs to establish a one-to-one correspondence between the state components and observation components; a difference calculation unit, used to calculate the difference information between the predicted state components and the corresponding observation components item by item to form a difference information set; a consistency filtering unit, used to identify and remove difference components that do not meet the preset difference constraints according to the preset difference constraints to obtain a filtered difference information set; and an encapsulation unit, used to structurally encapsulate the field-aligned predicted state data, the field-aligned observation data, and the filtered difference information set, and generate local update candidate data in a fixed order of state data, observation data, and difference information.

[0053] Furthermore, the event determination module may also include: a trigger mechanism embedding unit, used to introduce the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter; an input construction unit, used to perform time index consistency verification and field integrity verification on the local update candidate data, and construct event trigger determination input data according to the preset input rules of NN-ETM; and a determination execution unit, used to perform event trigger determination processing on the determination input data according to NN-ETM, and output the communication update trigger determination result. Through the collaborative work of the trigger mechanism embedding unit, the input construction unit, and the determination execution unit, the complete and reliable execution of the NN-ETM secure neural network event triggering mechanism is achieved, ensuring the accuracy and robustness of the communication update determination.

[0054] This embodiment achieves online consistency monitoring of multi-UAV spraying quality within a distributed framework, improving system scalability and ease of engineering deployment. It is understood that the detailed functional implementation of the above modules can be found in the descriptions in the foregoing method embodiments, and will not be elaborated further here.

[0055] The above are specific embodiments of the present invention, which demonstrate the substantial features and progress of the present invention. Equivalent modifications can be made to them according to actual usage needs, under the guidance of the present invention, and all such modifications are within the scope of protection of this solution.

Claims

1. A method for monitoring the quality of UAV painting based on distributed control, characterized in that, Includes the following steps: S1: Real-time acquisition of spraying execution data, flight status data, and operational environment data; time synchronization and validity processing of the acquired data; and generation of local spraying quality observation data. S2: Perform a local prediction process using distributed Kalman filtering based on local spraying quality observation data to obtain local spraying quality prediction status data; S3: Based on the local spraying quality prediction status data and the local spraying quality observation data, construct local update candidate data for filtering and updating; S4: Embed the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter, using local update candidate data and historical published data as input to generate an event triggering determination result for whether to trigger a communication update; S5: Based on the event trigger determination result, the update process of the distributed Kalman filter is executed in a controlled manner. When the preset event trigger conditions are met, local update candidate data is published to the neighboring UAVs and neighboring update data is received to form a neighboring update data set. S6: Based on the neighborhood update dataset, perform a distributed Kalman filter to perform a consistent fusion update on the local spraying quality prediction status data to obtain the fused spraying quality status estimation data; S7: Output the estimated spraying quality status data as the online consistency monitoring result of multi-UAV collaborative spraying operation.

2. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S1 specifically includes: S11: Collect spraying execution data, flight status data, and work environment data according to a unified sampling rhythm, and write a collection time identifier for each collection record to obtain raw data with collection time identifier; S12: Based on the acquisition time identifier, perform time synchronization, validity verification, missing data filling, and normalization on the raw data to obtain standardized observation feature data; S13: The standardized observation feature data is spliced ​​and encapsulated according to the field order of spraying execution data, flight status data, and operation environment data to generate local spraying quality observation data.

3. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S2 specifically includes: S21: Based on local spraying quality observation data, read the spraying execution data, flight status data, and operation environment data corresponding to the current moment, and construct the prediction calculation input data for distributed Kalman filtering; S22: Based on the input data for prediction calculation, generate local initial prediction data for distributed Kalman filtering to obtain local initial prediction data; S23: Based on the local predicted initial value data, perform distributed Kalman filtering state-time recursive processing, write the motion change information represented by the flight state data into the state recursive process, write the spraying process information represented by the spraying execution data into the state recursive process, and obtain the local recursive predicted state data. S24: Perform distributed Kalman filtering prediction consistency verification on the local recursive prediction state data based on the work environment data, identify the state components that are inconsistent with the work environment data, and perform replacement and update processing on the inconsistent components to obtain the verified prediction state data. S25: Perform observation consistency pre-check processing of distributed Kalman filtering on the verified predicted state data, calculate the difference information between the verified predicted state data and the local spraying quality observation data, and write the difference information into the predicted state data to obtain the predicted state data carrying the difference information. S26: Perform numerical stabilization processing of the predicted state data carrying difference information using distributed Kalman filtering to complete outlier removal and range constraint processing, and obtain stable local spraying quality predicted state data. S27: Output stable local spraying quality prediction status data.

4. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S3 specifically includes: S31: Based on local spraying quality prediction status data and local spraying quality observation data, and performing time index consistency verification on the two types of data, a data input pair with consistent time is obtained; S32: Based on time-consistent data input pairs, perform field mapping processing on the state components in the local spraying quality prediction state data and the observation components in the local spraying quality observation data to establish a one-to-one correspondence between the state components and the observation components, and obtain the field-aligned prediction state data and the field-aligned observation data. S33: Based on the field-aligned predicted state data and the field-aligned observed data, calculate the difference information between the predicted state component and the corresponding observed component item by item to form a set of difference information; S34: Perform consistency filtering on the set of difference information, identify the difference components that do not meet the constraints according to the preset difference constraints, and remove the difference components that do not meet the constraints to obtain the filtered set of difference information. S35: The field-aligned predicted state data, field-aligned observation data, and filtered difference information set are structured and encapsulated, and local update candidate data is generated in a fixed order of state data, observation data, and difference information. S36: Output local update candidate data and use the local update candidate data as input data for the event trigger determination in step S4.

5. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S4 specifically includes: S41: Introduce the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter, and set the NN-ETM secure neural network event triggering mechanism as the determination method for communication update execution; S42: Use locally updated candidate data as input data for the event triggering determination method; S43: Perform time index consistency check and field integrity check on the locally updated candidate data to obtain candidate judgment data; S44: Based on the candidate decision data, according to the preset input rules in the NN-ETM secure neural network event triggering mechanism, perform decision input construction processing on the candidate decision data to generate event trigger decision input data; S45: Based on the NN-ETM secure neural network event triggering mechanism, perform event triggering judgment processing on the event triggering judgment input data to obtain the communication update triggering judgment result; S46: Output the communication update trigger determination result.

6. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S5 specifically includes: S51: Based on the communication update trigger determination result and local update candidate data, form communication update control input data; S52: Based on the communication update control input data, perform communication update condition judgment processing; if the communication update trigger judgment result indicates that communication update is triggered, generate communication update execution data; if the communication update trigger judgment result indicates that communication update is not triggered, generate the neighborhood update data set initialization result. S53: After the communication update execution data is generated, send local update candidate data to neighboring UAVs and generate a sending completion record to obtain the published local update candidate data; S54: Based on the published local update candidate data, receive neighborhood update data from the neighboring drones and perform reception registration processing on the neighborhood update data to obtain a neighborhood update data set; S55: Perform time index consistency check and field integrity check on the neighborhood update data set, delete data records that do not meet the time index consistency requirement and delete data records that do not meet the field integrity requirement, and obtain the verified neighborhood update data set; S56: Output the set of neighborhood update data after verification.

7. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S6 specifically includes: S61: Based on the local spraying quality prediction status data and the verified neighborhood update data set, a consistent fusion update input data is formed; S62: Perform time index alignment and field alignment processing on the consistent fusion update input data to obtain the aligned local prediction state data and the aligned neighborhood update data set; S63: Perform fusion participation verification processing based on the aligned neighborhood update data set, identify data records that do not meet the fusion participation conditions and perform removal processing to obtain a valid neighborhood update data set; S64: Based on the effective neighborhood update data set, generate fusion update constraint data according to the distributed Kalman filter consensus fusion update rule; S65: Based on the fusion update constraint data, the effective neighborhood update data set is written into the aligned local prediction state data fusion update process, and distributed Kalman filter consistency fusion update processing is performed to obtain the fused spraying quality state estimation data. S66: Output the estimated spray quality status data after fusion.

8. The method for monitoring the quality of UAV painting based on distributed control according to claim 1, characterized in that, Step S7 specifically includes: S71: Based on the fused spraying quality status estimation data, generate input data to form a consistent monitoring result; S72: Generate input data for consistency monitoring results, perform time index consistency verification and field integrity verification, and generate candidate data from the results; S73: Generate candidate data based on the results, perform output field mapping processing on the state components in the generated candidate data, establish a one-to-one correspondence between the state components and the output fields of the online consistency monitoring results, and obtain the field-aligned result data; S74: Perform structured encapsulation processing on the field-aligned result data, and generate online consistency monitoring result output data according to the fixed order of time index and output fields; S75: Perform output consistency verification on the online consistency monitoring result output data, delete data records that do not meet the time index consistency requirements and data records that do not meet the field integrity requirements, and obtain the verified online consistency monitoring result output data.

9. A UAV painting quality monitoring system based on distributed control, characterized in that, include: The data acquisition module is used to collect spraying execution data, flight status data and work environment data in real time, and to perform time synchronization and validity processing on the collected data to generate local spraying quality observation data. The local prediction module is used to perform a local prediction process using distributed Kalman filtering based on local spraying quality observation data to obtain local spraying quality prediction status data. The candidate building module is used to construct local update candidate data for filtering and updating based on local spraying quality prediction status data and local spraying quality observation data. The event determination module is used to embed the NN-ETM secure neural network event triggering mechanism into the communication update process of the distributed Kalman filter. It takes local update candidate data and historical published data as input to generate an event triggering determination result on whether to trigger a communication update. The communication control module is used to control the update process of the distributed Kalman filter based on the event trigger determination result. When the preset event trigger conditions are met, it publishes local update candidate data to neighboring UAVs and receives neighboring update data to form a neighboring update data set. The consistency fusion module is used to perform a distributed Kalman filter consistency fusion update on the local spraying quality prediction status data based on the neighborhood update data set, so as to obtain the fused spraying quality status estimation data. The result output module is used to output the spraying quality status estimation data as the online consistency monitoring result of multi-UAV collaborative spraying operation.

10. A UAV painting quality monitoring system based on distributed control according to claim 9, characterized in that, Candidate building modules include: The time verification unit is used to perform time index consistency verification between the local spraying quality prediction status data and the local spraying quality observation data to obtain data input pairs with consistent time. The field mapping unit is used to perform field mapping processing on the state components and observation components in the data input pair, and establish a one-to-one correspondence between the state components and the observation components. The difference calculation unit is used to calculate the difference information between the predicted state component and the corresponding observed component item by item, forming a difference information set. The consistency filtering unit is used to identify and remove differential components that do not meet the preset differential constraints, and obtain a set of filtered differential information. The encapsulation unit is used to structurally encapsulate the field-aligned predicted state data, the field-aligned observation data, and the filtered set of difference information, and generate local update candidate data in a fixed order of state data, observation data, and difference information.

Images