An unmanned aerial vehicle paint spraying system and control method based on hierarchical task planning and adaptive flow control

Abstract

The present application relates to a kind of unmanned plane paint spraying system and control method based on layered task planning and adaptive flow control, belong to unmanned plane spraying technical field.The system includes unmanned plane platform, task control module, paint spraying device (including adjustable fixture, integrated pressure detection and electromagnetic valve regulation flow control module), with obstacle avoidance perception and visual guidance function's depth camera;Method uses layered task planning, combined with global flight path and local operation path, target and surface abnormal area are identified by depth camera, linkage real-time feed pressure, spray margin and visual result, adaptive flow regulation is realized using feedforward-feedback compound control strategy, applicable to high-altitude narrow scene such as electric power tower, steel structure facade.The system can improve the adaptability of different sizes paint tank, realize high-altitude slender target fine spraying, improve the accuracy of abnormal area re-spraying and spraying consistency.

Description

Technical Field

[0001] This invention belongs to the field of drone painting technology, specifically relating to a drone painting system and control method based on hierarchical task planning and adaptive flow control. Background Technology

[0002] In scenarios such as power towers, steel structure facades, narrow high-altitude components, and lightning protection tower tops, the target areas are located at high altitudes, with limited space and difficult access for manual spraying. Traditional manual spraying methods suffer from high labor intensity, high construction risks, poor coating consistency, and low work efficiency. With the improvement of the load capacity and airborne perception capabilities of drone platforms, using drones equipped with spraying devices for high-altitude spraying operations has become an important development direction.

[0003] Existing drone painting solutions mainly fall into two categories. One type uses a ground-based material supply method, delivering paint to the airborne spray gun via a hose. While this method reduces the drone's load, its operating radius is limited by the length of the delivery pipe, pipeline resistance, and ground deployment conditions, and it is also susceptible to reduced flight stability due to traction forces. The other type uses an airborne material tank and pressurized atomization device, allowing the drone to carry the paint itself to complete the operation. This method offers higher mobility, but is constrained by load capacity and range, and the stability of the spray flow rate is difficult to guarantee as the spray tank pressure decreases and the remaining volume changes.

[0004] Furthermore, existing painting drones mostly rely on manual remote control or preset flight paths to approach the target area, lacking a mechanism for hierarchical planning of global flight paths and local operation paths. After arriving at the work area, they often still rely on manual visual judgment of the painting area, and a closed-loop painting control system based on visual feedback has not yet been formed. Especially in scenarios where the target size is small and the shape is slender, such as lightning protection tower tips, lightning rods, and tower connectors, the lack of close-range visual recognition and local trajectory replanning mechanisms can easily lead to problems such as unstable painting distance, repeated painting, or missed painting.

[0005] Furthermore, existing airborne spraying controls typically execute spraying based solely on preset on / off values ​​or fixed valve openings, failing to adjust the spray volume in real time based on can pressure, remaining paint volume, and the condition of defects on the target surface. When the target surface has areas of paint peeling, damage, or localized thin coatings, the fixed spray volume control method struggles to meet the differentiated coating needs of normal and abnormal areas, leading to insufficient touch-up spraying or localized buildup, thus affecting coating quality. Summary of the Invention

[0006] The purpose of this invention is to address the problems of limited operating radius of existing drone painting operations, reliance on manual target recognition, lack of hierarchical path planning, and inability to adjust spray volume in real time, by providing a drone painting system and control method based on hierarchical task planning and adaptive flow control.

[0007] To achieve the above objectives, the technical solution of the present invention is: a drone painting system based on hierarchical task planning and adaptive flow control, comprising a drone platform, a task control module, a painting device, and a depth camera; the painting device is installed on the lower part or lower side of the drone platform, and the task control module is communicatively connected to the drone platform, the painting device, and the depth camera; the task control module deploys a global path planning algorithm, a local operation path planning algorithm, a visual recognition algorithm, and a painting decision algorithm; the painting device includes an adjustable clamp, a spray can, a flow control module, and a nozzle; the adjustable clamp includes a fixed base, a fixed can clamp, and a vertical telescopic base, wherein the fixed base is fixed... For the UAV platform, the fixed can clamp is used to horizontally clamp spray cans of different diameters, and the vertical telescopic base is used to adjust the installation height of the spray cans; the flow control module includes a module housing, a feeding channel, a three-way connector, a solenoid valve, a pressure sensor, and a discharge channel. The feeding channel is connected to the output end of the spray can, the forward branch of the three-way connector is connected to the solenoid valve, the outlet of the solenoid valve is connected to the discharge channel, the discharge channel is connected to the nozzle, and the lateral branch of the three-way connector is connected to the pressure sensor. The pressure sensor is used to detect the feeding pressure at the front end of the solenoid valve and send it to the task control module. The solenoid valve is used to adjust the spray output flow according to the control command of the task control module.

[0008] Furthermore, the global path planning algorithm is used to generate a global flight path based on the takeoff point and the target work point. The global flight path includes an approach path from the takeoff point to the target work point, as well as a return path or transfer path. The local work path planning algorithm is used to generate a work path based on the target spatial location and surface morphology obtained by the depth camera after the UAV arrives near the target work point. The visual recognition algorithm is used to identify the images acquired by the depth camera and obtain the category labels, detection confidence scores, and pixel-level defect masks for paint chipping areas, damaged areas, or coating abnormal areas on the target surface. The spraying decision algorithm is used to output adaptive flow control commands based on the recognition results.

[0009] Furthermore, the depth camera is positioned at the nose of the UAV platform to perform obstacle avoidance perception during flight, and to identify the target and acquire the target surface depth and surface state information after reaching the work area.

[0010] Furthermore, the flow control module is bonded and fixed to the UAV platform by a foam adhesive layer of preset thickness. The foam adhesive layer is used to fix the flow control module and provide a buffering and vibration reduction effect.

[0011] Furthermore, the fixed can clamp is connected to the fixed base, the vertical telescopic base is connected to the fixed base, and the adjustable clamp is used to accommodate paint cans of different heights and diameters.

[0012] Furthermore, the visual recognition algorithm performs distortion correction, brightness normalization, and target region cropping on the RGB image acquired by the depth camera, and then inputs the processed image into a pre-trained defect detection model. The pre-trained defect detection model outputs defect category, detection confidence, and pixel-level defect mask, and retains only defect results with a detection confidence of not less than 0.5. The task control module calculates the area of ​​the abnormal region by combining the depth data corresponding to the pixel-level defect mask.

[0013] Furthermore, the local job path is generated using the following formula:

[0014]

[0015] in, For the first in the local job path A trajectory point, The coordinates of the target center point, The radius of the spraying operation. For the surround angle, The height step size between adjacent trajectory layers; when When the local operation path is a circular operation path around the target; when When the target is a lightning protection tower top or lightning rod, the task control module determines the target center point coordinates, operation radius and height step size based on the target identification results.

[0016] This invention also provides a control method for a drone painting system based on hierarchical task planning and adaptive flow control, applied to the system described above, comprising:

[0017] Step S1: Receive the task information of the target to be painted. The task information includes at least the coordinates of the takeoff point, the coordinates of the target work point, the target category, and the desired painting parameters.

[0018] Step S2: The mission control module performs global flight path planning based on the takeoff point and the target work point, generates an approach path for the UAV platform to fly from the takeoff point to the target work point, and presets a return path or transfer path.

[0019] Step S3: When the UAV platform enters the identification trigger area around the target work point, the depth camera collects RGB images and depth data of the target area. The task control module performs distortion correction, brightness normalization and target area cropping on the RGB images and inputs the processed images into the pre-trained defect detection model to identify the spatial position, posture, shape boundary of the work target and the paint peeling area, damaged area or coating abnormal area on the target surface.

[0020] Step S4: The task control module generates a local operation path based on the recognition results, enabling the UAV platform to perform circling flight, spiral flight, or segmented close-to flight along the outer perimeter of the target, while maintaining the relative distance and spray angle between the nozzle and the target surface within a set range during the flight; during the execution of the local operation path in step S4, steps S5 and S6 are executed continuously or cyclically according to a set period.

[0021] Step S5: The flow control module sends the real-time feeding pressure detected by the pressure sensor to the task control module. The task control module calculates the basic spray volume based on the standard flow rate, the real-time feeding pressure, and the spray volume margin. When the visual recognition algorithm does not detect any abnormal areas, the basic spray volume is used as the target spray volume for the current execution. When the visual recognition algorithm detects paint peeling, damage, or coating abnormal areas, the task control module calculates the area of ​​the abnormal area based on the pixel-level defect mask and depth data output by the pre-trained defect detection model, then calculates the compensation flow rate based on the size of the abnormal area, and obtains the visually compensated target spray volume.

[0022] Step S6: The task control module adopts a feedforward-feedback control strategy. It calculates the reference pressure and the basic control duty cycle based on the target spray volume, and uses the basic control duty cycle as the feedforward reference quantity. At the same time, it calculates the pressure deviation and proportional-integral-derivative feedback correction quantity based on the real-time feeding pressure, and adds the feedback correction quantity to the feedforward reference quantity after scaling to obtain the corrected control duty cycle after amplitude limiting. The corrected control duty cycle is then sent to the flow control module to dynamically compensate the spray output flow rate.

[0023] Step S7: After the drone platform completes the local operation path, it leaves the operation area along the return path and ends the task, or switches to the next target operation point to continue the painting operation.

[0024] Furthermore, in step S6, the reference pressure is converted using the following formula:

[0025]

[0026] in, To determine the target spray volume The reference pressure obtained from the mapping This is the conversion factor between injection volume and pressure. This is the initial pressure bias.

[0027] The basic control duty cycle is calculated using the following formula:

[0028]

[0029] in, Based on the control duty cycle, and These are the minimum and maximum allowable duty cycles, respectively. This represents the maximum allowable output volume of the system.

[0030] Pressure deviation is calculated using the following formula:

[0031]

[0032] in, For a moment Feed pressure deviation, The real-time feeding pressure detected by the pressure sensor;

[0033] The proportional-integral-derivative feedback correction is calculated using the following formula:

[0034]

[0035] in, This is the correction amount for the material supply pressure feedback. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients;

[0036] The corrected control duty cycle is determined by the following formula:

[0037]

[0038] in, To correct the control duty cycle, This is the scaling factor for the feed pressure feedback correction amount to the duty cycle increment.

[0039] Furthermore, the spraying allowance is estimated using the following formula:

[0040]

[0041] in, For a moment The remaining amount of sprayed material. This is the initial spray amount. For a moment The actual spray volume;

[0042] The The pressure-volume-per-unit-time fitting curve is estimated by the task control module and satisfies the following:

[0043]

[0044] in, For a moment The execution duty cycle, For pressure The injection rate per unit time is obtained by fitting the pressure-injection rate per unit time sample obtained from the timed injection calibration. For at any time Real-time pressure of reading.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. This invention integrates a drone platform, a mission control module, a painting device, and a depth camera into one unit. It uses a spray can to provide paint and initial pressure, eliminating the need for external long-distance material delivery pipelines. This improves the maneuverability and deployment convenience of drones operating in confined high-altitude areas.

[0047] 2. The present invention adopts a hierarchical task planning method, which separates the global flight path planning from takeoff point A to target operation point B from the local operation path planning near the target. The global flight path planning includes at least the approach path and may further include the return path and / or transfer path, so as to meet the long-distance flight navigation requirements and achieve refined operation close to the target based on depth vision after arriving at the operation area.

[0048] 3. This invention incorporates the material supply pressure, spraying allowance, and visual recognition results into the spraying volume control process. In normal areas, spraying is performed according to the basic spraying volume, while in areas with paint peeling or damage, compensatory spraying is performed. This can improve the uniformity of spraying and the effect of defect repair, and reduce the risk of missed spraying and over-spraying.

[0049] 4. The depth camera in this invention combines obstacle avoidance perception and visual guidance functions, enabling the system to have higher autonomy and safety when approaching slender targets, flying around them, and identifying local defects.

[0050] 5. Based on the target spray volume calculation, this invention further introduces proportional-integral-derivative compensation adjustment based on the material supply pressure feedback, which can reduce the impact of pre-valve material supply pressure fluctuations and execution process delays on the spraying effect, thereby improving the spray volume tracking accuracy and spraying consistency. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the overall structure of the drone painting system of the present invention.

[0052] Figure 2 This is a block diagram showing the control relationship between the task control module, the painting device, and the depth camera of the present invention.

[0053] Figure 3 This is a schematic diagram of the painting device of the present invention.

[0054] Figure 4 This is a schematic diagram of the adjustable clamp of the present invention.

[0055] Figure 5 This is a schematic diagram of the flow control module of the present invention.

[0056] Figure 6 This is a flowchart of the overall process for the drone painting control method of the present invention.

[0057] Explanation of reference numerals in the attached drawings: 1. Unmanned aerial vehicle platform; 2. Mission control module; 3. Painting device; 4. Depth camera; 31. Adjustable clamp; 311. Fixed base; 312. Fixed can clamp; 313. Vertical telescopic base; 32. Spray can; 33. Flow control module; 331. Module housing; 332. Feed channel; 333. T-connector; 334. Solenoid valve; 335. Pressure sensor; 336. Discharge channel; 34. Nozzle. Detailed Implementation

[0058] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0059] This invention provides a drone painting system based on hierarchical task planning and adaptive flow control, comprising a drone platform, a task control module, a painting device, and a depth camera. The painting device is installed on the lower part or lower side of the drone platform. The task control module is communicatively connected to the drone platform, the painting device, and the depth camera. The task control module deploys a global path planning algorithm, a local operation path planning algorithm, a visual recognition algorithm, and a painting decision algorithm. The painting device includes an adjustable clamp, a spray can, a flow control module, and a nozzle. The adjustable clamp includes a fixed base, a fixed can clamp, and a vertical telescopic base, and the fixed base is fixed to the drone platform. The fixed can clamp is used to horizontally clamp spray cans of different diameters, and the vertical telescopic base is used to adjust the installation height of the spray cans. The flow control module includes a module housing, a feed channel, a three-way connector, a solenoid valve, a pressure sensor, and a discharge channel. The feed channel is connected to the output end of the spray can, the forward branch of the three-way connector is connected to the solenoid valve, the outlet of the solenoid valve is connected to the discharge channel, the discharge channel is connected to the nozzle, and the lateral branch of the three-way connector is connected to the pressure sensor. The pressure sensor is used to detect the feed pressure at the front end of the solenoid valve and send it to the task control module. The solenoid valve is used to adjust the spray output flow according to the control command of the task control module.

[0060] This invention also provides a control method for a drone painting system based on hierarchical task planning and adaptive flow control, applied to the system described above, comprising:

[0061] Step S1: Receive the task information of the target to be painted. The task information includes at least the coordinates of the takeoff point, the coordinates of the target work point, the target category, and the desired painting parameters.

[0062] Step S2: The mission control module performs global flight path planning based on the takeoff point and the target work point, generates an approach path for the UAV platform to fly from the takeoff point to the target work point, and presets a return path or transfer path.

[0063] Step S3: When the UAV platform enters the identification trigger area around the target work point, the depth camera collects RGB images and depth data of the target area. The task control module performs distortion correction, brightness normalization and target area cropping on the RGB images and inputs the processed images into the pre-trained defect detection model to identify the spatial position, posture, shape boundary of the work target and the paint peeling area, damaged area or coating abnormal area on the target surface.

[0064] Step S4: The task control module generates a local operation path based on the recognition results, enabling the UAV platform to perform circling flight, spiral flight, or segmented close-to flight along the outer perimeter of the target, while maintaining the relative distance and spray angle between the nozzle and the target surface within a set range during the flight; during the execution of the local operation path in step S4, steps S5 and S6 are executed continuously or cyclically according to a set period.

[0065] Step S5: The flow control module sends the real-time feeding pressure detected by the pressure sensor to the task control module. The task control module calculates the basic spray volume based on the standard flow rate, the real-time feeding pressure, and the spray volume margin. When the visual recognition algorithm does not detect any abnormal areas, the basic spray volume is used as the target spray volume for the current execution. When the visual recognition algorithm detects paint peeling, damage, or coating abnormal areas, the task control module calculates the area of ​​the abnormal area based on the pixel-level defect mask and depth data output by the pre-trained defect detection model, then calculates the compensation flow rate based on the size of the abnormal area, and obtains the visually compensated target spray volume.

[0066] Step S6: The task control module adopts a feedforward-feedback control strategy. It calculates the reference pressure and the basic control duty cycle based on the target spray volume, and uses the basic control duty cycle as the feedforward reference quantity. At the same time, it calculates the pressure deviation and proportional-integral-derivative feedback correction quantity based on the real-time feeding pressure, and adds the feedback correction quantity to the feedforward reference quantity after scaling to obtain the corrected control duty cycle after amplitude limiting. The corrected control duty cycle is then sent to the flow control module to dynamically compensate the spray output flow rate.

[0067] Step S7: After the drone platform completes the local operation path, it leaves the operation area along the return path and ends the task, or switches to the next target operation point to continue the painting operation.

[0068] The following is a detailed implementation process of the present invention.

[0069] like Figures 1 to 5As shown, this embodiment provides a drone painting system based on hierarchical task planning and adaptive flow control, including a drone platform 1, a task control module 2, a painting device 3, and a depth camera 4. The painting device 3 is fixed to the bottom of the drone platform 1, and the depth camera 4 is installed at the nose of the drone platform 1 so that the forward field of view of the depth camera 4 covers the flight direction of the drone and the target work area.

[0070] In this embodiment, the UAV platform 1 is preferably a multi-rotor UAV platform, which itself has a flight controller, a positioning and navigation unit, and an attitude stabilization unit. The mission control module 2 is preferably an onboard computer with computing power, which is communicatively connected to the flight controller of the UAV platform 1, and is used to issue flight path, speed commands, and operational status commands to the flight controller; wherein, the computing power is used to perform inference calculations for the pre-trained defect detection model.

[0071] like Figure 3 and Figure 4 As shown, the painting device 3 includes an adjustable clamp 31, a spray can 32, a flow control module 33, and a nozzle 34. The adjustable clamp 31 is used to connect the drone platform 1 and the spray can 32. In this embodiment, the adjustable clamp 31 consists of a fixed base 311, a fixed can clamp 312, and a vertical telescopic base 313; wherein, the fixed base 311 is fixed to the drone platform 1, the fixed can clamp 312 is connected to the fixed base 311 and is used to horizontally clamp the spray can 32 to accommodate spray cans 32 of different diameters, and the vertical telescopic base 313 is connected to the fixed base 311 and is used to precisely adjust the installation height of the spray can 32 according to its height. Thus, the main function of the adjustable clamp 31 is to enable the same drone platform 1 to accommodate spray cans of different sizes.

[0072] The spray can 32 is filled with anti-corrosion paint, lightning protection paint, or other functional spraying materials, and serves as a paint storage unit and an initial pressure supply unit. The output end of the spray can 32 is connected to the inlet end of the flow control module 33, and the outlet end of the flow control module 33 is connected to the nozzle 34. The nozzle 34 can be a fan-shaped nozzle, a cone-shaped nozzle, or a fine mist nozzle suitable for local repairs.

[0073] like Figure 5 As shown, the flow control module 33 includes a module housing 331, a feed channel 332, a tee connector 333, a solenoid valve 334, a pressure sensor 335, and a discharge channel 336. It is bonded to the UAV platform 1 using a foam adhesive layer of preset thickness. The foam adhesive layer is used to fix the flow control module 33 and provide cushioning and vibration reduction. One end of the feed channel 332 is connected to the output end of the spray can 32, and the other end is connected to the tee connector 333.

[0074] The three-way connector 333 is used to divide the paint pressure from the spray can 32 into a forward main flow path and a lateral pressure measuring branch. The forward main flow path is connected to the inlet of the solenoid valve 334, the outlet of the solenoid valve 334 is connected to the discharge channel 336, and the discharge channel 336 is further connected to the nozzle 34. The lateral pressure measuring branch is connected to the pressure sensor 335, which is used to detect the feed pressure at the front end of the solenoid valve 334 and send the pressure signal to the task control module 2.

[0075] During operation, the paint is output from the spray tank 32 and enters the three-way connector 333 through the feed channel 332. The mainstream paint flows along the forward mainstream path into the solenoid valve 334, and after adjustment by the solenoid valve 334, it is sprayed out through the discharge channel 336 and the nozzle 34. Simultaneously, the pressure in the lateral branch of the three-way connector 333 is transmitted to the pressure sensor 335, so that the pressure sensor 335 can collect the real-time feed pressure at the front end of the solenoid valve 334. Since the pressure sensor 335 is arranged in a lateral position in the mainstream path, the pressure detection structure does not directly occupy the mainstream spray channel, which helps to reduce the impact on the spray delivery and improve the stability of pressure acquisition.

[0076] To achieve adaptive flow control, the flow control module 33 integrates pressure monitoring and flow regulation functions and is communicatively connected to the task control module 2. During operation, the flow control module 33 sends the feeding pressure detected by the pressure sensor 335 to the task control module 2 and receives the control duty cycle command output by the task control module 2 to drive the solenoid valve 334 to regulate the output flow of the sprayed material.

[0077] In a preferred embodiment, process parameters and control coefficients are calibrated before formal operation. During calibration, a standard test panel of the same or similar material as the actual target material is selected. The spraying distance is controlled at 0.20m–0.40m, the relative operating speed at 0.20m / s–0.80m / s, and the ambient wind speed at no more than 3m / s. The spraying is repeated 3–5 times under rated reference pressure. The target spraying amount per unit area is obtained by weighing the test panel before and after spraying and dividing the difference by the effective coverage area. For lightning protection paint applications, The preferred value is 0.12 kg / m² to 0.18 kg / m².

[0078] Under the above calibration conditions, the measured deposition mass on the test plate and the actual coating mass sprayed from the spray can were recorded respectively, and the spray deposition efficiency was determined by the ratio of the two. For the scenario of drones spraying paint close to the top of a tower, The preferred value is 0.45 to 0.65; when changing the nozzle type, paint viscosity, or spraying distance, the value should be measured again. .

[0079] Furthermore, under the condition that the spray width, flight speed, and spraying distance remain unchanged, select It offers three pressure levels: 1.0, 0.8, and 0.6. Cross-spraying experiments were conducted at three margin levels: 1.0, 0.5, and 0.2. The preferred method is to use the real-time feeding pressure detected by the pressure sensor 335; measure the deviation between the actual spraying amount and the target spraying amount under each working condition, and obtain the result using least squares fitting. and For lightning protection paint spraying operations, Preferably, it is 0.35 to 0.65. The preferred value is 0.15 to 0.35.

[0080] In a preferred embodiment, the pressure-volume-per-unit-time curve is also pre-calibrated. During calibration, under the conditions of fixed nozzle type, paint type, and spraying distance, at least four different inlet pressure settings are selected, such as 100%, 85%, 70%, and 55% of the rated reference pressure, and continuous spraying is maintained for a fixed duration at each pressure setting. The preferred time is 1s to 3s, and the mass of the corresponding sprayed paint is weighed. Thus, a sample of spray volume per unit time is obtained. Subsequently, under pressure The independent variable is the spray volume per unit time. As the dependent variable, a quadratic polynomial fitting or piecewise linear fitting is used to obtain the pressure-per-unit-time injection volume function. For the quadratic polynomial fitting case, its expression can be written as:

[0081]

[0082] in, , and Obtained by least squares fitting.

[0083] For the abnormal area compensation coefficient The calibration can be performed on artificially prepared defect samples. Preferably, defect samples with abnormal area ratios of 0.05, 0.20, and 0.40 are prepared respectively, and re-spraying is performed under the same spraying parameters. The criterion is to establish a coating thickness that is close to the target thickness after re-spraying without over-spraying. Comparison table. For localized touch-up spraying of lightning protection paint, The preferred value is 0.50 to 1.00.

[0084] For the proportional-integral-differential feedback coefficient , , The critical oscillation tuning method is preferred for determining the tuning. During tuning, first set... , Slowly increase The critical gain is recorded until the feed pressure response collected by pressure sensor 335 shows continuous oscillation. and critical oscillation period , and according to , , Initial values ​​are given, and then fine-tuning of the overshoot and settling time is performed under conditions of a step change in spray volume. For the painting apparatus of this embodiment, Preferably, it is 0.20 to 1.50. Preferably 0.01 ~0.30 , The preferred time is 0.001s to 0.08s.

[0085] Depth camera 4 is positioned at the nose of the UAV platform 1 and performs two functions. First, while the UAV platform 1 is flying towards the target work point B, depth camera 4 identifies obstacles ahead of the flight path to assist the mission control module 2 in adjusting the flight trajectory. Second, after the UAV platform 1 enters the work area, depth camera 4 outputs RGB images and depth information of the target area for the mission control module 2 to identify the location, outline, and surface anomalies of the specific work target.

[0086] The task control module 2 is equipped with a global path planning algorithm, a local operation path planning algorithm, a visual recognition algorithm, and a spraying decision algorithm. The global path planning algorithm generates a global flight path based on the takeoff point A and the target operation point B. The global flight path includes at least an approach path from takeoff point A to target operation point B, and may further include a return path after the operation is completed and / or a transfer path to the next target operation point. Takeoff point A is the starting flight position of the UAV mission. The local operation path planning algorithm generates an operation path based on the target spatial position and surface morphology obtained by the depth camera 4 after the UAV reaches the vicinity of target operation point B. The visual recognition algorithm is deployed in the task control module 2 and uses a pre-trained defect detection model to identify the images acquired by the depth camera 4 to obtain category labels, detection confidence, and pixel-level defect masks for paint peeling areas, damaged areas, or coating abnormal areas on the target surface. The spraying decision algorithm outputs adaptive flow control commands based on the identification results. The above algorithms are implemented by the processor of the task control module 2 executing program instructions in the memory.

[0087] In this embodiment, the pre-trained defect detection model in the visual recognition algorithm is deployed in the task control module 2, and inference is performed using the computing power provided by the task control module 2. The pre-trained defect detection model is preferably a conventional target detection and pixel-level instance segmentation network in the art. In a preferred embodiment, a ResNet-50 backbone network is used for feature extraction, combined with a feature pyramid network for multi-scale feature fusion, and defect categories, detection boxes, and pixel-level defect masks are output through classification / regression branches and mask branches, respectively. The network structure, training process, and deployment method can all be implemented using techniques known in the art.

[0088] During the offline training phase, an annotated image dataset containing lightning protection paint peeling, substrate damage, coating loss, and local abnormal patches was constructed. The annotations included at least defect category labels and pixel-level masks. Supervised learning was employed during training, using a weighted sum of classification loss, bounding box regression loss, and mask segmentation loss as the objective loss function to iteratively optimize the network parameters. After training, the model parameters were solidified and deployed to task control module 2.

[0089] During the operation, the task control module 2 first performs distortion correction and brightness normalization on the RGB image acquired by the depth camera 4, then crops the target surface area according to the depth boundary and target contour, and then scales the cropped target area image to the preset input size before sending it to the pre-trained defect detection model for inference.

[0090] The pre-trained defect detection model outputs defect categories, detection confidence scores, and pixel-level defect masks. Task control module 2 filters the model output results, retaining only defect results with a detection confidence score of not less than 0.5 and located within the target surface area; when multiple detection results overlap, an overlap suppression rule with an overlap ratio threshold of not more than 0.3 is used to retain the results with higher confidence scores.

[0091] like Figure 6 As shown, the drone painting control method in this embodiment includes steps such as task information reception, global path planning, target recognition, local operation path planning, adaptive flow control, and task completion or transition. The hierarchical task planning includes two levels: global path planning and local operation path planning. Global path planning addresses the navigation problem of the drone platform 1 flying from takeoff point A to target operation point B and leaving the target operation area or switching to the next target operation point after completion. Local operation path planning addresses how the drone platform 1 performs precise operations around the specific painting target near target operation point B.

[0092] In this embodiment, before takeoff, the ground mission terminal sends mission information to the mission control module 2. The mission information includes at least the spatial location of the target work point B, the target type, and preset spraying parameters. The mission control module 2 generates a global flight path based on the current pose information of the UAV platform 1. The global flight path includes at least the approach path from the takeoff point A to the target work point B, and may further include the return path after the operation is completed or the transfer path to the next target work point. The mission control module 2 controls the UAV platform 1 to fly to the vicinity of the target work point B according to the approach path.

[0093] When the UAV platform 1 arrives at the identification trigger area around the target operation point B, the mission control module 2 switches to local operation mode and controls the depth camera 4 to scan the target area. For slender targets such as lightning protection tower tops and lightning rods, the depth camera 4 can identify the center position of the top of the target and the axial direction, and transmit the identification results to the mission control module 2.

[0094] Task control module 2 generates a circular or spiral operation path. Specifically, it uses the target center point as the reference point. Establish a local coordinate system centered on the center and set the spraying operation radius. and height step Then according to different angles Generate multiple operation trajectory points around the target. Its expression can be written as:

[0095] Formula 1;

[0096] in, For the first in the local job path A trajectory point; when At that time, all trajectory points are located at the same height level, corresponding to the surrounding operation path; when During this process, each trajectory point changes layer by layer along the height direction, corresponding to a spiral operation path. When the UAV platform 1 executes this local operation path, the nozzle 34 faces the target surface to maintain a relatively stable spraying distance and spraying posture.

[0097] In this embodiment, the execution of the local operation path is the outer layer process, while the flow calculation and feedback compensation are the inner layer loop processes. That is, during the flight of the UAV platform 1 along the local operation path corresponding to Formula 1, the task control module 2 continuously or periodically triggers visual recognition, abnormal area calculation, target spray volume update, and control duty cycle update, and sends the updated control commands to the flow control module 33 for execution; therefore, the spray volume compensation and material supply pressure feedback compensation described later are not performed once after the local operation path is completed, but are executed in real time during the local operation process.

[0098] exist Figure 6 In the overall process shown, adaptive flow control includes four stages: standard flow determination, pressure and margin compensation, visual anomaly compensation, and feed pressure feedback correction. First, based on the preset target spraying volume per unit area... Effective spray width Relative operating speed and spray deposition efficiency Standard flow It can be represented as:

[0099] Formula 2;

[0100] in, For standard flow, The target spraying amount per unit area. For effective spray width, The operating speed of the unmanned aerial vehicle platform (1) relative to the target surface, For spray deposition efficiency.

[0101] Preferably, the target spraying amount per unit area According to the target wet film thickness With coating density Confirmed, satisfied Alternatively, it can be determined by the ratio of the mass increment after spraying a standard test panel to the covered area. For lightning protection paint or anti-corrosion paint applications, A suitable range is 0.08 kg / m² to 0.25 kg / m², preferably 0.12 kg / m² to 0.18 kg / m². Spray deposition efficiency. The measured deposition quality can be obtained through standard test plates. Compared with the actual spray quality of the spray can ratio Confirmed; assuming consistent nozzle type, spraying distance, operating speed, and ambient wind speed. The value can be 0.35 to 0.75, and preferably 0.45 to 0.65.

[0102] When no abnormal areas are detected on the target surface As a standard flow rate, the basic spray volume is calculated based on the feed pressure and spray margin. .

[0103] Since the feed pressure before the valve decreases as the spray can pressure decays during continuous operation, and the remaining spray volume decreases as the sprayed material is consumed, this embodiment further estimates the remaining spray volume. Its expression is:

[0104] Formula 3;

[0105] in, For a moment The remaining amount of sprayed material. This is the initial spray amount. For a moment The actual spray volume.

[0106] In this embodiment, the actual spray volume in Formula 3 It is not directly measured by the mass flow meter, but rather by the task control module 2 based on a pre-calibrated pressure-per-time injection volume function. The online estimation is obtained. Specifically, the task control module 2 acquires the feeding pressure detected by the pressure sensor 335 in real time. , and by The injection volume per unit time corresponding to the current pressure is obtained by interpolation or substitution calculation; in a preferred embodiment, It can be written as:

[0107] Formula 4;

[0108] in, For pressure The spray volume per unit time is as follows , , These are the fitting coefficients;

[0109] Combined with the current execution duty cycle Estimated time The actual spray volume satisfies:

[0110] Formula 5;

[0111] in, For a moment The execution duty cycle, if the system has not yet entered the feeding pressure feedback correction stage, then Take the basic control duty cycle If the system has already performed a feed pressure feedback correction, then Take the corrected control duty cycle Therefore, we can... Perform discrete or continuous integration to obtain the spray margin. .

[0112] Based on this, the basic injection volume It can be represented as:

[0113] Formula Six;

[0114] in, Based on the base spray volume, For pressure sensor 335 at time The real-time feeding pressure obtained from the detection. The pressure is the rated reference pressure under the corresponding operating conditions for spray can 32, and the pressure compensation coefficient is... The margin compensation coefficient is used to characterize the impact of pre-valve feed pressure attenuation on the spraying effect. This is used to characterize the degree of correction that changes in the spraying allowance affect the spraying volume control. Through this compensation process, even if the supply pressure at the front end of the solenoid valve 334 gradually decreases, the task control module 2 can still output a basic spraying volume command after pressure and allowance compensation. The flow control module 33 adjusts the spraying output flow accordingly to keep the spraying thickness within the target range.

[0115] Preferably, The value can be 0.20 to 0.80, preferably 0.35 to 0.65; The value can be 0.10 to 0.50, and preferably 0.15 to 0.35. and This can be determined through bench calibration: [The following is a separate, unrelated statement:] In terms of spray width... Work speed Under the condition of keeping the spraying distance constant, select at least three pressure ratios. Operating conditions and at least three margin ratios Under operating conditions, a standard test panel is sprayed, and the deviation between the actual spray coverage and the target spray coverage is measured. Linear fitting or least squares fitting is used to minimize the spray coverage error in order to obtain the coating system. and .

[0116] During normal spraying, depth camera 4 continuously acquires images of the target surface. The visual recognition algorithm in task control module 2 uses a pre-trained defect detection model to identify areas of paint peeling, damage, or coating abnormalities on the target surface, and extracts the area of ​​the corresponding abnormal area. Simultaneously, task control module 2 obtains the working area corresponding to the current spraying field of view. abnormal area ratio It can be represented as:

[0117] Formula 7;

[0118] In this embodiment, if the set of defective pixels is denoted as The set of pixels in the working region is denoted as Then, task control module 2 calculates the area of ​​the abnormal region based on the depth map. and work area area Under the condition that the target surface is approximately frontally viewed and has a small local curvature, for pixels... Its corresponding projected area unit is approximately ,in This is the depth value of the pixel. and This is the focal length intrinsic parameter of depth camera 4. and The pixel spacing is defined as follows: If the target surface has a large tilt angle or local curvature changes, the task control module 2 fits a local plane based on the pixel neighborhood depth points and obtains the surface normal vector, then calculates the angle between the surface normal vector and the camera optical axis. and adopt Make corrections. In actual calculations, the set... All valid pixels By summing, we can obtain For sets All valid pixels By summing, we can obtain .

[0119] When the visual recognition algorithm does not detect abnormal areas, that is The system uses a base spray volume The target spray volume is used for current execution; when the visual recognition algorithm detects an abnormal area, i.e. Target spray volume after visual compensation It can be represented as:

[0120] Formula 8;

[0121] in, The maximum spray volume allowed by the system. This is the abnormal area compensation coefficient; preferably, The value can be 0.30 to 1.50, and preferably 0.50 to 1.00. Based on the proportion of defect area The type of coating is selected from a preset parameter table or determined through a standard defect sample repainting experiment; in a preferred embodiment, when hour, Take 0.30 to 0.60; when hour, Take 0.60 to 1.00; when hour, Take a value of 1.00 to 1.50.

[0122] The larger the proportion of the abnormal area, the larger the corresponding compensation spray volume, but it will not exceed the maximum spray volume allowed by the system. .

[0123] To obtain the target spray volume Then, task control module 2 first maps the target jet volume to reference pressure. Its expression is:

[0124] Formula Nine;

[0125] in, To determine the target spray volume The reference pressure obtained from the mapping This is the conversion factor between injection volume and pressure. This is the initial pressure bias.

[0126] Then, task control module 2 calculates the basic control duty cycle. Its expression is:

[0127] Formula 10;

[0128] in, Based on the control duty cycle, and These are the minimum and maximum allowable duty cycles, respectively.

[0129] Next, task control module 2 constructs a feeding pressure feedback compensation based on the deviation between the reference pressure and the real-time feeding pressure, and the pressure deviation... It can be represented as:

[0130] Formula 11;

[0131] Proportional-integral-derivative feedback correction It can be represented as:

[0132] , Formula 12;

[0133] in, This is the feed pressure feedback correction amount, with dimensions consistent with pressure deviation. Consistent; This is the proportionality coefficient. The integral coefficient is... is the differential coefficient.

[0134] Finally, the corrected control duty cycle It can be represented as:

[0135] Formula Thirteen;

[0136] in, This is the scaling factor for the feed pressure feedback correction amount to the duty cycle increment. Task control module 2 is based on... A control command is generated and sent to the flow control module 33, which then performs feedback compensation and correction on the output flow of the sprayed material.

[0137] Preferably, This can be obtained through a feed pressure feedback compensation calibration experiment: Under fixed nozzle type, spraying distance, and paint type conditions, gradually change the duty cycle and record the steady-state feed pressure change and the corresponding duty cycle change. Take the fitting result of the ratio of the two as the result. When pressure is expressed in MPa, 0.03 is acceptable. ~0.30 Preferably 0.05 ~0.15 .

[0138] Preferably, A value of 0.20 to 1.50 is acceptable. 0.01 is acceptable. ~0.30 , The range can be 0.001s to 0.08s. The proportional coefficient, integral coefficient, and differential coefficient can be obtained through a proportional-integral-differential tuning process: first, let... , Gradually increase Record the critical gain until the real-time feeding pressure response exhibits continuous oscillation. and critical oscillation period Then set the initial coefficients to , , Then, fine-tuning is performed based on the pressure overshoot and settling time to ensure that the pressure overshoot is no more than 10% and the settling time is no more than 2 seconds.

[0139] In this embodiment, formulas nine through thirteen together constitute a feedforward-feedback control strategy. Specifically, the task control module 2 first determines the target spray volume... The reference pressure is obtained from Formula 9. The basic control duty cycle is obtained from Formula 10. Among them, the basic control duty cycle As a feedforward reference quantity, it is used to enable the ejector output to quickly track changes in the target ejection quantity. Subsequently, task control module 2 uses formulas eleven and twelve to calculate the feedback correction amount of the real-time feed pressure relative to the reference pressure. And according to Formula Thirteen, through the scaling factor The feedback correction is mapped to a duty cycle increment, and compared with the feedforward reference quantity. The final execution duty cycle is obtained by superimposing the results. Because Formula 13 includes upper and lower limit constraints, when the superposition result exceeds the allowable range, the final execution duty cycle is limited to [specific value]. and This allows for over-adjustment and the suppression of control dead zones and overshoot risks.

[0140] In a preferred embodiment, the target is the lightning protection tower tip near the power transmission line. After the UAV platform 1 takes off from the ground work point, the mission control module 2 first plans the flight path from takeoff point A to the vicinity of lightning protection tower No. 3, and controls the UAV platform 1 to fly along the planned path. After arriving near the target work point B, the depth camera 4 identifies the specific location of the lightning protection tower tip and the lightning rod. Based on the identification results, the mission control module 2 generates a surrounding work path around the lightning rod; when continuous coverage along the height direction is required, a non-zero depth camera can also be used. Generate a spiral operation path.

[0141] As the drone platform 1 flies around the described orbital path, nozzle 34 performs painting operations towards the surface of the lightning protection tower tip. If the depth camera 4 detects a region where the lightning protection paint has peeled off, the task control module 2 includes the area of ​​this peeled-off area in the abnormal area count. The compensation spray volume is then calculated using formulas seven and eight. Subsequently, task control module 2 generates a corrected control duty cycle command for the area based on formulas nine to thirteen and sends it to flow control module 33 to increase the spray output flow rate of the area; if the surface condition of the corresponding orientation is normal, the coating is applied according to the basic spray volume.

[0142] Through the above methods, this embodiment achieves a complete closed-loop control process from "reaching the target area" to "identifying the specific target" and then to "fine spraying around the target." Especially for scenarios where there are local defects on the target surface, this invention can dynamically increase the spray volume according to the size of the defect area, and reduce the impact of pressure fluctuations by using proportional-integral-derivative adjustment based on material supply pressure feedback, thereby improving the accuracy of re-spraying and the consistency of operations.

[0143] It should be noted that the adjustment amount corresponding to the flow control module 33 in this invention can be reflected as duty cycle, control voltage, pulse width or other equivalent control parameters; the depth camera 4 can also be replaced by a binocular vision module or a structured light vision module with depth perception capability. As long as it can achieve target recognition, depth perception and surface anomaly recognition, it is within the protection scope of this invention.

[0144] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A drone painting system based on hierarchical task planning and adaptive flow control, characterized in that, The system includes a drone platform, a mission control module, a painting device, and a depth camera. The painting device is installed at the lower part or lower side of the drone platform. The mission control module is communicatively connected to the drone platform, the painting device, and the depth camera. The mission control module is equipped with a global path planning algorithm, a local operation path planning algorithm, a visual recognition algorithm, and a painting decision algorithm. The painting device includes an adjustable clamp, a spray can, a flow control module, and nozzles. The adjustable clamp includes a fixed base, a fixed can clamp, and a vertical telescopic base. The fixed base is fixed to the drone platform, and the fixed can clamp is used to laterally hold different diameters. The spray can is equipped with a vertical telescopic base for adjusting its installation height. The flow control module includes a module housing, a feed channel, a three-way connector, a solenoid valve, a pressure sensor, and a discharge channel. The feed channel is connected to the output end of the spray can. The forward branch of the three-way connector is connected to the solenoid valve. The outlet of the solenoid valve is connected to the discharge channel, which is connected to the nozzle. The lateral branch of the three-way connector is connected to the pressure sensor. The pressure sensor is used to detect the feed pressure at the front end of the solenoid valve and send it to the task control module. The solenoid valve is used to adjust the spray output flow rate according to the control command of the task control module.

2. The UAV painting system based on hierarchical task planning and adaptive flow control according to claim 1, characterized in that, The global path planning algorithm is used to generate a global flight path based on the takeoff point and the target work point. The global flight path includes an approach path from the takeoff point to the target work point, as well as a return path or transfer path. The local work path planning algorithm is used to generate a work path based on the target spatial location and surface morphology obtained by the depth camera after the UAV arrives near the target work point. The visual recognition algorithm is used to identify images acquired by a depth camera and obtain category labels, detection confidence scores, and pixel-level defect masks for paint chipping areas, damaged areas, or coating abnormal areas on the target surface. The spraying decision algorithm is used to output adaptive flow control commands based on the recognition results.

3. The UAV painting system based on hierarchical task planning and adaptive flow control according to claim 1, characterized in that, The depth camera is located at the nose of the UAV platform and is used to perform obstacle avoidance perception during flight, and to identify the target and obtain the target surface depth and surface condition information after reaching the work area.

4. The UAV painting system based on hierarchical task planning and adaptive flow control according to claim 1, characterized in that, The flow control module is bonded and fixed to the UAV platform by a foam adhesive layer of preset thickness. The foam adhesive layer is used to fix the flow control module and provide a buffering and vibration reduction effect.

5. The UAV painting system based on hierarchical task planning and adaptive flow control according to claim 1, characterized in that, The fixed can clamp is connected to the fixed base, the vertical telescopic base is connected to the fixed base, and the adjustable clamp is used to accommodate paint cans of different heights and diameters.

6. The UAV painting system based on hierarchical task planning and adaptive flow control according to claim 2, characterized in that, The visual recognition algorithm performs distortion correction, brightness normalization, and target region cropping on the RGB image acquired by the depth camera, and then inputs the processed image into a pre-trained defect detection model. The pre-trained defect detection model outputs defect category, detection confidence, and pixel-level defect mask, and retains only defect results with a detection confidence of not less than 0.

5. The task control module calculates the area of ​​the abnormal region by combining the depth data corresponding to the pixel-level defect mask.

7. The UAV painting system based on hierarchical task planning and adaptive flow control according to claim 2, characterized in that, The local job path is generated using the following formula: in, For the first in the local job path A trajectory point, The coordinates of the target center point, The radius of the spraying operation. For the circumferential angle, The height step size between adjacent trajectory layers; when When the local operation path is a circular operation path around the target; when When the target is a lightning protection tower top or lightning rod, the task control module determines the target center point coordinates, operation radius and height step size based on the target identification results.

8. A control method for a drone painting system based on hierarchical task planning and adaptive flow control, applied to the system described in any one of claims 1 to 7, characterized in that, include: Step S1: Receive the task information of the target to be painted. The task information includes at least the coordinates of the takeoff point, the coordinates of the target work point, the target category, and the desired painting parameters. Step S2: The mission control module performs global flight path planning based on the takeoff point and the target work point, generates an approach path for the UAV platform to fly from the takeoff point to the target work point, and presets a return path or transfer path. Step S3: When the UAV platform enters the identification trigger area around the target work point, the depth camera collects RGB images and depth data of the target area. The task control module performs distortion correction, brightness normalization and target area cropping on the RGB images and inputs the processed images into the pre-trained defect detection model to identify the spatial position, posture, shape boundary of the work target and the paint peeling area, damaged area or coating abnormal area on the target surface. Step S4: The task control module generates a local operation path based on the recognition results, enabling the UAV platform to perform circling flight, spiral flight, or segmented close-to flight along the outer perimeter of the target, while maintaining the relative distance and spray angle between the nozzle and the target surface within a set range during the flight; during the execution of the local operation path in step S4, steps S5 and S6 are executed continuously or cyclically according to a set period. Step S5: The flow control module sends the real-time feeding pressure detected by the pressure sensor to the task control module. The task control module calculates the basic spray volume based on the standard flow rate, the real-time feeding pressure, and the spray volume margin. When the visual recognition algorithm does not detect an abnormal area, the base spray volume is used as the target spray volume for the current execution; when the visual recognition algorithm detects paint peeling, damage or coating abnormal areas, the task control module calculates the area of ​​the abnormal area based on the pixel-level defect mask and depth data output by the pre-trained defect detection model, then calculates the compensation flow based on the size of the abnormal area, and obtains the visually compensated target spray volume. Step S6: The task control module adopts a feedforward-feedback control strategy, calculates the reference pressure based on the target injection volume and calculates the basic control duty cycle, and uses the basic control duty cycle as the feedforward reference quantity. Simultaneously, the pressure deviation and proportional-integral-derivative feedback correction amount are calculated based on the real-time feeding pressure. The feedback correction amount is then scaled and superimposed on the feedforward reference amount to obtain the corrected control duty cycle after amplitude limiting. The corrected control duty cycle is then sent to the flow control module to dynamically compensate the output flow rate of the sprayed material. Step S7: After the drone platform completes the local operation path, it leaves the operation area along the return path and ends the task, or switches to the next target operation point to continue the painting operation.

9. A control method for a UAV painting system based on hierarchical task planning and adaptive flow control according to claim 8, characterized in that, In step S6, the reference pressure is converted using the following formula: in, To determine the target spray volume The reference pressure obtained from the mapping This is the conversion factor between injection volume and pressure. This is the initial pressure bias. The basic control duty cycle is calculated using the following formula: in, Based on the control duty cycle, and These are the minimum and maximum allowable duty cycles, respectively. This represents the maximum allowable output volume of the system. Pressure deviation is calculated using the following formula: in, For a moment The deviation in the material supply pressure, The real-time feeding pressure detected by the pressure sensor; The proportional-integral-derivative feedback correction is calculated using the following formula: in, This is the correction amount for the material supply pressure feedback. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients; The corrected control duty cycle is determined by the following formula: in, To correct the control duty cycle, This is the scaling factor for the feed pressure feedback correction amount to the duty cycle increment.

10. The control method for a UAV painting system based on hierarchical task planning and adaptive flow control according to claim 8, characterized in that, The spraying allowance is estimated using the following formula: in, For a moment The remaining amount of sprayed material. This is the initial spray amount. For a moment The actual spray volume; The The pressure-volume-per-unit-time fitting curve is estimated by the task control module and satisfies the following: in, For a moment The execution duty cycle, For pressure The injection rate per unit time is obtained by fitting the pressure-injection rate per unit time sample obtained from the timed injection calibration. For at any time Real-time pressure of reading.

Images