A wall-climbing robot body-based maritime operation adsorption positioning method

Abstract

The present application relates to underwater operation robot technical field, especially to a kind of based on wall-climbing robot body's maritime operation adsorption positioning method, comprising: the structural feature map of target ship is obtained and operation route is planned;Wall-climbing robot is combined with propeller robot after entering water and is adsorbed to ship body and initializes positioning parameter;Collect underwater environment data enhancement processing and adsorption force data;Control wall-climbing robot to travel along route and obtain mileage, inertia, enhanced environment and adsorption force data;Based on enhanced environment data, structural feature of ship body is identified and structural feature map is matched to obtain visual positioning information, adsorption force data is matched to obtain tactile positioning information with structural feature map;Fusion mileage, visual positioning and tactile positioning information determine real-time position;Fusion enhanced environment data, inertia data and adsorption force distribution data solve attitude angle;When attitude angle is over threshold value, adjust motion state or start propeller correction.

Description

Technical Field

[0001] This invention relates to the field of underwater operation robot technology, and in particular to a method for adsorption and positioning in marine operations based on a wall-climbing robot body. Background Technology

[0002] Currently, underwater cleaning and inspection of ship hulls are mainly carried out by underwater robots. Existing underwater robots are mainly divided into two categories: propeller-type and magnetic adsorption-type. Propeller-type robots are powered by multiple propellers, swimming in the water and working close to the ship hull. However, they have difficulty maintaining their posture in wave-like or swell-like environments, resulting in poor contact with the ship's surface and difficulty in achieving precise positioning. Magnetic adsorption-type robots use magnetic force to adhere themselves to the surface of the ship's steel plates. Although this solves the contact problem, their positioning method is relatively simple, usually relying solely on encoders to record the distance traveled for position estimation. However, when traveling for extended periods on complex curved surfaces such as the underside of the ship, the encoder is prone to cumulative errors due to factors such as drive wheel slippage and changes in the ship's curvature. Furthermore, the lack of external absolute position calibration methods leads to a significant decrease in positioning accuracy as the operation time increases, failing to meet the requirements of high-precision operations. Summary of the Invention

[0003] To overcome the above shortcomings, this invention provides a maritime operation adsorption positioning method based on a wall-climbing robot body, aiming to improve the technical problem that existing magnetic adsorption underwater robots rely solely on single mileage data for positioning, and the positioning error accumulates with the travel distance due to the lack of multi-source absolute position calibration.

[0004] This invention provides the following technical solution: a maritime operation adsorption and positioning method based on a wall-climbing robot body. The method is applied to a maritime operation robot, which includes a wall-climbing robot and a thruster robot. The adsorption and positioning method includes the following steps:

[0005] S1. Obtain a structural feature map of the target vessel and plan the operation route;

[0006] S2. After combining the wall-climbing robot and the thruster robot, they are submerged in water, adsorbed onto the surface of the hull, and the positioning parameters are initialized.

[0007] S3. Collect underwater environmental data and perform enhancement processing, and at the same time collect the adsorption force data of each adsorption unit of the wall-climbing robot;

[0008] S4. Control the wall-climbing robot to move along the work route and acquire mileage data, inertial data, enhanced environmental data and adsorption force data;

[0009] S5. Identify the structural features of the ship based on the enhanced environmental data, match them with the structural feature map to obtain visual positioning information, and simultaneously match the adsorption force data with the structural feature map to obtain tactile positioning information.

[0010] S6. By integrating the mileage data, visual positioning information, and tactile positioning information, the real-time position of the wall-climbing robot is determined;

[0011] S7. By integrating the enhanced environmental data, inertial data, and adsorption force distribution data of each adsorption unit, the attitude angle of the wall-climbing robot is calculated.

[0012] S8. When the attitude angle exceeds the preset threshold, adjust the motion state of the wall-climbing robot or start the thruster robot to correct the attitude.

[0013] Preferably, in step S1, the steps of obtaining the structural feature map of the target vessel and planning the operation route specifically include:

[0014] Obtain the general arrangement drawing and coordinate data of the target vessel, and establish a three-dimensional model of the target vessel;

[0015] Structural feature information is extracted from the three-dimensional model, and the extracted structural feature information is annotated in the three-dimensional model to generate the structural feature map;

[0016] The work area is set based on the three-dimensional model, and the work route is deployed automatically or semi-automatically.

[0017] Preferably, in step S2, the step of combining the wall-climbing robot and the propeller robot, entering the water, adhering to the surface of the hull, and initializing the positioning parameters specifically includes:

[0018] The wall-climbing robot and the thruster robot are combined using a detachable connection structure to form a combined unit;

[0019] The assembly is placed in the water, and the thruster robot provides the thrust to help the assembly approach the working surface of the target ship hull.

[0020] When the distance between the assembly and the working surface reaches the adsorption trigger condition, the magnetic adsorption unit of the wall-climbing robot is activated to adsorb the wall-climbing robot onto the working surface.

[0021] The encoder of the wall-climbing robot is initialized to zero, and each sensor is activated to establish an initial state reference.

[0022] Preferably, in step S3, the steps of collecting underwater environmental data and performing enhancement processing, and simultaneously collecting the adsorption force data of each adsorption unit of the wall-climbing robot, specifically include:

[0023] Activate the image acquisition device mounted on the wall-climbing robot to acquire raw underwater images;

[0024] The original underwater image is enhanced to obtain the enhanced environmental data.

[0025] At the same time, the adsorption force sensors installed on each adsorption unit are activated to collect real-time adsorption force data of each adsorption unit and establish an adsorption force distribution baseline.

[0026] Preferably, in step S4, the step of controlling the wall-climbing robot to move along the work route and acquiring mileage data, inertial data, enhanced environmental data, and adhesion force data specifically includes:

[0027] Motion control commands are generated based on the work route to drive the drive wheel assembly of the wall-climbing robot to move along the work route;

[0028] During the journey, the mileage data is acquired in real time through the encoder mounted on the wall-climbing robot, and the inertial data is acquired in real time through the inertial measurement unit.

[0029] Simultaneously, it continuously receives the enhanced environmental data after the enhancement process, and continuously receives the adsorption force data fed back in real time from each adsorption unit.

[0030] Preferably, the method further includes pre-adjusting the adsorption force parameters of each adsorption unit of the wall-climbing robot, specifically including:

[0031] Based on the real-time position of the wall-climbing robot and the structural feature map, determine whether the wall-climbing robot has traveled to a preset front range of any pre-marked structural feature point;

[0032] When it is determined that the traveler has entered the preset forward range, the curvature information of the structural feature point is extracted from the structural feature map;

[0033] Based on the curvature information, the target adsorption force parameters of each adsorption unit are determined, and adjustment commands are sent to each adsorption unit to adjust the adsorption force to match the curvature of the ship ahead in advance.

[0034] Preferably, in step S5, the steps of obtaining visual positioning information and tactile positioning information specifically include:

[0035] The enhanced environmental data is input into the visual feature extraction module to identify the ship's structural features.

[0036] The identified hull structure features are matched with the pre-marked structure feature information in the structure feature map. When the match is successful, the coordinate data of the structure feature information in the structure feature map is obtained as the visual positioning information.

[0037] Simultaneously, the adsorption force fluctuation feature sequence generated by passing through the ship's structural features is extracted from the adsorption force data. The adsorption force fluctuation feature sequence is then matched with the structural feature information at the corresponding position in the structural feature map. When the pattern matching is successful, the coordinate data of the corresponding structural feature information in the structural feature map is obtained as the tactile positioning information.

[0038] Preferably, in step S6, the step of determining the real-time position of the wall-climbing robot specifically includes:

[0039] Acquire the mileage data, the visual positioning information, and the tactile positioning information at the same time.

[0040] Determine the confidence weights of the mileage data, the visual positioning information, and the tactile positioning information respectively;

[0041] Based on the confidence weights, the mileage data, the visual positioning information, and the tactile positioning information are weighted and fused to calculate the real-time position of the wall-climbing robot.

[0042] Preferably, in step S7, the step of calculating the attitude angle of the wall-climbing robot specifically includes:

[0043] Acquire the enhanced environmental data, the inertial data, and the adsorption force distribution data of each adsorption unit at the same time.

[0044] Based on the enhanced environmental data, the relative positional relationship between the wall-climbing robot and the surface of the ship is analyzed to obtain visual posture reference information;

[0045] Based on the inertial data, the acceleration and angular velocity of each axis of the wall-climbing robot are analyzed to obtain inertial attitude reference information;

[0046] Based on the adsorption force distribution data of each adsorption unit, the differences in adsorption force on the left and right sides and the differences in adsorption force at the front and rear ends are analyzed to obtain tactile posture reference information.

[0047] By integrating the visual posture reference information, the inertial posture reference information, and the tactile posture reference information, the three-dimensional posture angles of the wall-climbing robot are calculated.

[0048] Preferably, in step S8, the step of adjusting the motion state of the wall-climbing robot or activating the thruster robot for attitude correction specifically includes:

[0049] The real-time three-dimensional attitude angles of the wall-climbing robot are obtained, and the real-time three-dimensional attitude angles are compared with a preset attitude angle threshold.

[0050] When any axial angle in the real-time three-dimensional attitude angle exceeds the corresponding attitude angle threshold, the attitude deviation direction and deviation amount are determined.

[0051] Based on the direction and amount of the attitude deviation, a correction strategy is determined. The correction strategy includes: adjusting the differential speed of the left and right drive wheel sets of the wall-climbing robot to generate a reverse correction torque, and / or activating the thruster robot to provide auxiliary thrust or auxiliary torque.

[0052] The real-time three-dimensional attitude angle of the wall-climbing robot is continuously monitored until it returns to the preset attitude angle threshold range.

[0053] The present invention has the following beneficial effects:

[0054] 1. In this invention, multi-source collaborative positioning is achieved by integrating mileage data, visual positioning information and tactile positioning information. Visual recognition of ship structure features and adsorption force fluctuation feature sequence matching are used as absolute position calibration methods. This overcomes the defect of existing magnetic adsorption robots that rely solely on a single data source, the encoder, which causes positioning errors to accumulate with the distance traveled. This enables the robot to maintain high-precision positioning capability during long-term operation on the ship surface.

[0055] 2. In this invention, by pre-aiming at the curvature change of the hull ahead through the structural feature map, the adsorption force parameters of each adsorption unit are adjusted in advance before traveling to structural feature points such as ribs and welds, so that the adsorption force is pre-matched with the curvature of the hull ahead, avoiding adsorption instability or detachment due to sudden curvature changes, which significantly improves the robot's smoothness of passage and operational safety in complex curved areas of the hull.

[0056] 3. In this invention, by integrating enhanced environmental data, inertial data, and adsorption force distribution data of each adsorption unit, the robot's posture angle is comprehensively calculated from three dimensions: vision, inertia, and touch. This upgrades posture perception from traditional manual visual judgment to closed-loop control driven by quantitative data. When the posture is abnormal, it can automatically trigger correction actions, greatly improving the robot's posture stability during operation.

[0057] 4. In this invention, multiple sensing methods such as underwater image enhancement processing, visual feature recognition, adsorption force sensing, and inertial measurement are organically integrated. Each sensing channel can complement each other under different working conditions. When a certain channel is disturbed by the environment, the other channels can still provide effective positioning or attitude information, so that the whole system can maintain reliable positioning and attitude control capabilities under various adverse conditions such as turbid water and complex ship structure. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating a maritime operation adsorption and positioning method based on a wall-climbing robot body proposed in this invention. Detailed Implementation

[0059] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] This invention provides a method for adsorption and positioning in maritime operations based on a wall-climbing robot. Figure 1 As shown, the method is applied to a marine operation robot, which includes a wall-climbing robot and a thruster robot. The adsorption and positioning method includes the following steps:

[0061] S1. Obtain a structural feature map of the target vessel and plan the operation route.

[0062] Furthermore, in S1, the steps of acquiring the structural feature map of the target vessel and planning the operation route specifically include:

[0063] Obtain the general arrangement drawing and coordinate data of the target vessel, and establish a three-dimensional model of the target vessel;

[0064] Structural feature information is extracted from the 3D model, and the extracted structural feature information is annotated in the 3D model to generate a structural feature map;

[0065] The work area is set based on a 3D model, and the work route is deployed automatically or semi-automatically.

[0066] Specifically, the general arrangement drawing and coordinate data of the target vessel are obtained. The general arrangement drawing is a digital drawing generated during the ship design phase, containing geometric information such as the hull shape, rib positions, and bulkhead boundaries. The coordinate data consists of the three-dimensional coordinates of various feature points on the hull surface in the hull coordinate system. The general arrangement drawing and coordinate data are imported into a 3D modeling system to create a 3D model of the target vessel. Structural feature information is extracted from the 3D model. Structural feature information includes rib positions, weld orientations, and bulkhead boundaries, and may also include geometric features with positioning reference value such as bilge bends and keel lines. Extraction methods include directly reading the reserved feature annotation layers in the 3D model, or automatically identifying curvature abrupt changes as structural feature points through curvature analysis of the 3D model.

[0067] The specific method of curvature analysis is as follows: for any vertex on the 3D model Calculate its mean curvature :

[0068] ;

[0069] in, The maximum principal curvature at that vertex. This represents the minimum principal curvature at that vertex. The maximum and minimum principal curvatures are obtained by fitting a quadratic surface to the changes in the normal vectors of the triangular faces within the neighborhood of that vertex. When the vertex... The average curvature satisfies At that time, the vertex is marked as a structural feature point, where The preset curvature threshold is set to 0.05 to 0.15 for rib locations and 0.03 to 0.10 for weld orientation. All vertices on the 3D model are traversed, and the average curvature is calculated for each vertex and compared with the curvature threshold. Vertices meeting the conditions are clustered to form a set of structural feature points, and the type and coordinate data of each structural feature point are recorded.

[0070] The extracted structural feature information is annotated in the 3D model to generate a structural feature map. The structural feature map is a 3D model carrying markers for the location of structural features. Each marker point records the type of the structural feature and its coordinate data in the ship's coordinate system. The work area is set based on the 3D model. The work area can be a section or side of the ship's hull. The target area can be selected by the operator on the 3D model interface, or it can be automatically divided by the system according to preset task parameters. After the work area is set, the system automatically or semi-automatically deploys the work route. In automatic deployment, the system generates a path trajectory covering the entire work area according to the surface shape of the work area and a preset line spacing. In semi-automatic deployment, the operator can make local adjustments or confirm the automatically generated path. The path trajectory consists of a series of sequentially arranged path points. Each path point contains its 3D coordinates in the ship's coordinate system and the direction of the surface normal vector at that point.

[0071] After the above steps are completed, the structural feature map and the operation route are downloaded to the integrated control system of the wall-climbing robot for subsequent travel and positioning.

[0072] S2. After combining the wall-climbing robot and the propulsion robot, the robot enters the water, adheres to the surface of the hull, and initializes the positioning parameters.

[0073] Furthermore, in S2, after the wall-climbing robot and the propulsion robot are combined and submerged in water, adhering to the surface of the hull, the steps for initializing the positioning parameters specifically include:

[0074] The wall-climbing robot and the thruster robot are combined using a detachable connection structure to form a combined unit;

[0075] The combined structure is placed in the water, and the thruster robot provides the thrust to help the combined structure approach the working surface of the target ship hull;

[0076] When the distance between the combined body and the working surface reaches the adsorption trigger condition, the magnetic adsorption unit of the wall-climbing robot is activated, and the wall-climbing robot is adsorbed onto the working surface.

[0077] The encoder of the wall-climbing robot is initialized to zero, and each sensor is activated to establish an initial state reference.

[0078] Specifically, the wall-climbing robot and the thruster robot are combined using a detachable connection structure to form a combined unit. The detachable connection structure is located on the back or side of the wall-climbing robot and mates with the corresponding interface of the thruster robot. The detachable connection structure can employ a combination of limiting slots and quick-release screws. The limiting slots provide mechanical restraint for the wall-climbing robot and the thruster robot in the front-back and left-right directions, while the quick-release screws secure them vertically. After assembly, the wall-climbing robot and the thruster robot form a rigid mechanical unit, and electrically, control signals and power are transmitted through a wiring assembly.

[0079] The assembly is deployed into the water. Deployment can be achieved by using a crane to lift the assembly from the deck and lower it smoothly into the water via lifting rings, or by using a deployment and recovery device to slide the assembly into the water along a track. After the assembly enters the water, the propulsion robot activates, providing thrust to assist the assembly in moving towards the target hull's working surface according to remote operator commands or a preset approach path. As the assembly gradually approaches the working surface, the distance sensor on the wall-climbing robot continuously monitors the distance between the assembly and the working surface. When the distance decreases to within the threshold range set by the adsorption trigger condition, an adsorption action is triggered. The threshold range for the adsorption trigger condition can be set from 5 cm to 15 cm. After triggering, the wall-climbing robot's magnetic adsorption unit is activated, generating an adsorption force to adhere the wall-climbing robot and the assembly together to the working surface.

[0080] The magnetic adsorption unit consists of an array of multiple independent magnet units, each a permanent magnet. The specific activation method for the magnetic adsorption unit is as follows: the wall-climbing robot uses a drive mechanism to push the magnet unit from its storage position to its working position, bringing the working surface of the magnet unit close to the surface of the ship's steel plate. Magnetic lines of force pass through the steel plate, forming a closed magnetic circuit and generating an adsorption force. After adsorption is complete, the propeller robot can reduce its thrust output or enter standby mode to reduce energy consumption. The encoder of the wall-climbing robot is initialized to zero. The encoder is located on the output shaft of the drive motor or the drive wheel shaft of the wall-climbing robot. Zero initialization means resetting the encoder pulse count value at the current moment to zero, so that subsequent count values ​​directly reflect the distance traveled from the current position. Simultaneously, the sensors on the wall-climbing robot are activated to establish an initial state reference. The sensors that need to be activated include the inertial measurement unit, the adsorption force sensors on each magnetic adsorption unit, and the image acquisition device. The initial state reference of the inertial measurement unit is established by allowing it to remain stationary for a preset time in a stable adsorption state. The system collects triaxial acceleration and triaxial angular velocity data output by the inertial measurement unit during this period, with a sampling frequency of [frequency missing]. The total number of sampling points is Calculate the zero bias compensation value using the following formula:

[0081] ;

[0082] in, For the first The triaxial acceleration vector of each sampling point , For the first The triaxial angular velocity vector of each sampling point , For the zero-bias compensation vector of acceleration, This is the zero-bias compensation vector for angular velocity. During subsequent use, the acceleration and angular velocity at each sampling moment are subtracted from the corresponding zero-bias compensation value to obtain the compensated inertial data: .

[0083] The initial state reference of the adsorption force sensor is established as follows: Assume the magnetic adsorption unit contains a total of Each magnet unit is equipped with an adsorption force sensor, which, under stable adsorption conditions, measures the adsorption force at a sampling frequency. collection The sampling point, the first Baseline of adsorption force of each magnet unit Calculate using the following formula:

[0084] ;

[0085] in, For the first The magnet unit in the first The adsorption force value at each sampling point .all The adsorption force baselines of each magnet unit together constitute the adsorption force distribution baseline: Once the baseline for the adsorption force distribution is established, it will be used for comparative analysis of the real-time adsorption force data of each magnet unit during subsequent movement.

[0086] Once activated, the image acquisition device begins acquiring raw underwater images for subsequent enhancement processing. After initialization, the wall-climbing robot and its sensor system have established clear initial baselines and are capable of traveling along the work route and continuously acquiring various types of data required for positioning.

[0087] S3. Collect underwater environmental data and perform enhancement processing, while simultaneously collecting adsorption force data of each adsorption unit of the wall-climbing robot.

[0088] Furthermore, in S3, the steps of collecting and enhancing underwater environmental data, and simultaneously collecting adsorption force data of each adsorption unit of the wall-climbing robot, specifically include:

[0089] Activate the image acquisition device on the wall-climbing robot to obtain raw underwater images;

[0090] The original underwater image is enhanced to obtain enhanced environmental data;

[0091] At the same time, the adsorption force sensors installed on each adsorption unit are activated to collect real-time adsorption force data of each adsorption unit and establish an adsorption force distribution baseline.

[0092] Specifically, the image acquisition devices on the wall-climbing robot are activated to acquire raw underwater images. These devices include a turbidity camera and a selfie stick camera. The turbidity camera is mounted at the front of the wall-climbing robot, facing the work surface, and is used to acquire close-up images of the work surface. The selfie stick camera is mounted on the side or rear of the wall-climbing robot via a retractable pole, acquiring a global image of the robot's overall position relative to the hull surface from a side-top or rear view. Once activated, the turbidity camera and selfie stick camera continuously output raw underwater images at a frame rate of 25 frames per second. The raw underwater images are then enhanced to obtain enhanced environmental data. This enhancement processing is performed by an enhancement module, which embeds a turbidity underwater image enhancement system. The turbidity underwater image enhancement system employs a dual-pipeline architecture that runs in parallel with a traditional enhancement pipeline and a deep learning enhancement model. By default, it executes in the order of traditional enhancement followed by deep learning, but can also be switched to a deep learning-only mode.

[0093] Traditional enhancement links include cascaded dark channel prior dehazing modules and The contrast enhancement module includes an adaptive parameter adjustment unit. The dark channel prior dehazing module performs dehazing on the original underwater image based on an atmospheric scattering model. The atmospheric scattering model expression is:

[0094] ;

[0095] In the formula, For the original underwater image in pixels Pixel value at that location, To obtain a clear image after dehazing, This represents the global atmospheric light value. Transmittance. Through dark channel prior estimation, the dark channel is defined as a pixel-level channel. Centered on, window size is The minimum value of each color channel within a local region:

[0096] ;

[0097] The formula for estimating transmittance is:

[0098] ;

[0099] In the formula, This is a parameter for defogging intensity, with a value ranging from 0 to 1. To prevent excessively low transmittance from amplifying noise, a lower limit for transmittance is set. The typical value is 0.1, and the final transmittance is taken as... The restored clear image is as follows:

[0100] ;

[0101] The adaptive parameter adjustment unit analyzes the degree of fogging in each frame of the original underwater image in real time. and contrast loss Atomization level Defined as the ratio of the dark channel mean to 255:

[0102] ;

[0103] Contrast loss Defined as the grayscale entropy normalized value:

[0104] ;

[0105] in, The grayscale entropy of the original underwater image. This represents the theoretical maximum grayscale entropy for an image of the same size. The optimal dehazing intensity parameters are dynamically calculated. and Contrast enhancement threshold parameter :

[0106] ;

[0107] ;

[0108] To ensure smooth inter-frame updates, an exponential moving average is used to update the parameters:

[0109] ;

[0110] ;

[0111] in, The smoothing factor is set to 0.7. and The dehazing intensity parameters used for the current frame and Threshold parameter, and These are the corresponding parameters used in the previous frame.

[0112] The contrast enhancement module converts the dehazed image to... Color space, for The luminance channel implements limited contrast adaptive histogram equalization, with the histogram height limited to a threshold. The excess portion is redistributed, and the cumulative distribution is calculated block-by-block by the transformation function. Block artifacts are eliminated through bilinear interpolation, resulting in the output image of the traditional enhancement link. The deep learning enhancement model employs an encoder-decoder network structure, which includes wavelet transform enhancement blocks and gradient-guided feature blocks. The wavelet transform enhancement blocks enhance the input feature map... use Wavelet basis downsampling decomposition:

[0113] ;

[0114] After splicing the four sub-bands... Convolutional fusion, followed by reconstruction using depthwise separable convolutions, results in a composite image. Features of the same scale are added to the input feature residuals:

[0115] ;

[0116] Gradient-guided feature block utilization Operator calculates gradient magnitude of feature map :

[0117] ;

[0118] in, and These represent the feature maps in the horizontal and vertical directions, respectively. Gradient. The magnitude of the gradient is... After gating, multiply with the original feature:

[0119] ;

[0120] Then send in two in sequence Each The residual structure is formed by depthwise separable convolution and channel splitting normalization.

[0121] ;

[0122] The encoder-decoder employs three levels of downsampling and upsampling, with each level containing embedded wavelet transform enhancement blocks and gradient-guided feature blocks. The model incorporates skip connections. The output layer maps the features back to 3 channels and adds them to the residual of the input image to obtain the output image of the deep learning augmentation model.

[0123] Deep learning augmentation models require pre-training before implementation. The training dataset consists of paired clear underwater images with corresponding turbid underwater images. The loss function is the mean squared error function, and the optimizer is... The optimizer was configured with an initial learning rate of 0.001, a batch size of 8, and 200 training epochs. After training, the model weights were converted to... The format is deployed in the enhanced processing module, and the inference engine can adopt it. or ,accomplish or Accelerate inference. The output images of the traditional enhancement pipeline and the deep learning enhancement model are processed sequentially in a dual pipeline mode. That is, the output of the traditional enhancement pipeline is used as the input of the deep learning enhancement model, and the final output is the enhanced environmental data.

[0124] Simultaneously, the adsorption force sensors installed on each adsorption unit are activated to collect real-time adsorption force data. The adsorption force sensors are located on the back of the working surface of each magnet unit and are used to detect the actual adsorption force generated by that magnet unit on the surface of the ship's steel plate in real time. Each adsorption force sensor continuously collects adsorption force data at a preset sampling frequency, which can be set from 10 Hz to 50 Hz. The collected real-time adsorption force data from each adsorption unit is transmitted to the integrated control system for subsequent positioning and attitude calculation.

[0125] The enhanced environmental data in this step is used for the subsequent steps of S4 (travel data acquisition), S5 (visual recognition and positioning), and S7 (posture fusion). The real-time adsorption force data is used for the subsequent steps of pre-aiming adjustment, S5 (tactile positioning), and S7 (posture fusion).

[0126] S4. Control the wall-climbing robot to move along the working route and acquire mileage data, inertial data, enhanced environmental data, and adhesion force data.

[0127] Furthermore, in S4, the steps of controlling the wall-climbing robot to move along the work path and acquiring mileage data, inertial data, enhanced environmental data, and adhesion force data specifically include:

[0128] Motion control commands are generated based on the work route to drive the drive wheel assembly of the wall-climbing robot to move along the work route;

[0129] During the journey, the robot acquires mileage data in real time through the encoder and inertial data in real time through the inertial measurement unit.

[0130] Simultaneously, it continuously receives enhanced environmental data after enhancement processing, and continuously receives adsorption force data fed back in real time from each adsorption unit.

[0131] Specifically, based on the work route generated in step S1, the integrated control system converts the sequence of path points in the work route into motion control commands. The work route consists of a series of sequentially arranged path points, each containing three-dimensional coordinates and the direction of the surface normal vector in the ship's coordinate system. The integrated control system reads the coordinates of the current path point and the coordinates of the next path point, calculates the travel direction vector, and combines it with the surface normal vector to generate the rotation speed command for the drive wheel sets, driving the left and right drive wheel sets of the wall-climbing robot to travel along the work route. The drive wheel sets include a left drive wheel set and a right drive wheel set, and differential control enables the wall-climbing robot to travel in a straight line and turn on the ship's surface.

[0132] During the movement, the climbing robot acquires mileage data in real time via an encoder mounted on it. The encoder is mounted on the output shaft of the drive motor or the drive wheel shaft, outputting pulse signals as the drive wheel rotates. The pulse count per unit time is proportional to the number of rotations of the drive wheel. Combined with the diameter parameter of the drive wheel, the pulse count is converted into the distance traveled, forming mileage data. Mileage data reflects the displacement increment of the climbing robot on the work surface since the last sampling moment. Simultaneously, inertial data is acquired in real time via an inertial measurement unit (IMU). The IMU includes a three-axis accelerometer and a three-axis gyroscope, outputting three-axis acceleration data and three-axis angular velocity data respectively at a preset sampling frequency. This inertial data has been compensated for according to the zero-bias compensation value established in step S2 during acquisition; the compensated inertial data is... and ,in and These are the original acceleration vector and the angular velocity vector, respectively. and This is the zero-bias compensation vector calculated in step S2. The purpose of zero-point compensation is to eliminate the adverse effects of the sensor's own zero bias on the accuracy of subsequent attitude calculation.

[0133] Simultaneously, it continuously receives enhanced environmental data output from the enhancement processing module in step S3. This enhanced environmental data includes a close-up enhanced image of the work surface from the perspective of the muddy water camera and a global enhanced image from the perspective of the selfie stick camera. Simultaneously, it continuously receives real-time adsorption force data from each adsorption unit. Adsorption force sensors installed on each adsorption unit transmit real-time adsorption force values ​​to the integrated control system.

[0134] Mileage data, inertial data, enhanced environmental data, and adsorption force data are timestamped and aligned within the integrated control system. Data is correlated at the same sampling time to form a synchronized multi-source data frame, which is used for fusion positioning in step S6 and attitude calculation in step S7. Within the synchronized multi-source data frame, mileage data provides relative displacement increments, inertial data provides acceleration and angular velocity information, enhanced environmental data provides visual perception input, and adsorption force data provides the actual adsorption force value of each adsorption unit at the current time.

[0135] Furthermore, the method also includes pre-adjusting the adsorption force parameters of each adsorption unit of the wall-climbing robot, specifically including:

[0136] Based on the real-time position and structural feature map of the wall-climbing robot, determine whether the wall-climbing robot has traveled to the preset front range of any pre-marked structural feature point;

[0137] When it is determined that the travel has reached the preset forward range, the curvature information of the structural feature point is extracted from the structural feature map;

[0138] Based on curvature information, the target adsorption force parameters of each adsorption unit are determined, and adjustment commands are sent to each adsorption unit to adjust the adsorption force to match the curvature of the hull in advance.

[0139] Specifically, during the wall-climbing robot's movement along the work route, the integrated control system compares the robot's real-time position, obtained from multi-source fusion positioning in step S6, with the structural feature map generated in step S1 to determine whether the robot has moved into the preset forward range of any pre-marked structural feature point. The preset forward range refers to a distance interval extending forward along the direction of travel, with the structural feature point as the reference. This distance interval can be set to 10 cm to 30 cm. The specific determination method is as follows: calculate the arc length distance between the wall-climbing robot's real-time position and each structural feature point in the structural feature map on the work route. When this arc length distance is less than the threshold of the preset forward range, it is determined that the wall-climbing robot has moved into the preset forward range of that structural feature point.

[0140] When the ship is determined to have traveled to a predetermined forward range, the curvature information of that structural feature point is extracted from the structural feature map. The curvature information includes the radius of curvature of the hull surface at that structural feature point perpendicular to the direction of travel. radius of curvature In step S1, through the mean curvature The calculation shows that the relationship is: Based on radius of curvature Determine the target adsorption force parameters for each adsorption unit. Adsorption force of the magnetic adsorption unit. The gap between the working surface of the magnet and the surface of the steel plate The relationship between them is represented as follows:

[0141] ;

[0142] in, This represents the maximum adsorption force when there is zero gap between the working surface of the magnet and the surface of the steel plate. The magnetic attenuation coefficient is determined by the properties of the magnet material and the magnetic circuit structure. When the hull surface curvature is positive (i.e., the surface is convex), the gap between the magnet units increases, resulting in a change in the equivalent gap. With radius of curvature Width of the magnet unit along the direction of travel The relationship is:

[0143] ;

[0144] When the hull surface curvature is negative, i.e., the surface is concave, the equivalent clearance change is... Calculated using the same formula, the result is negative. The target adsorption force parameter is the magnet working gap that should be adjusted to. The calculation formula is:

[0145] ;

[0146] The integrated control system uses the calculated target adsorption force parameters The system sends adjustment commands to each adsorption unit. Upon receiving the adjustment command, each adsorption unit adjusts the working gap of the magnet unit to the target value via a servo mechanism. This process ensures that the adsorption force is pre-matched to the curvature of the hull in front, thus maintaining stable adsorption as the wall-climbing robot passes through the structural feature point. Adjustment can be achieved by adjusting the mechanical gap between the magnet unit and the robot body, or by adjusting the coil current when the magnet unit is equipped with an electromagnetic auxiliary coil to compensate for the adsorption force. After adjustment, the wall-climbing robot continues along the working route and passes through the structural feature point.

[0147] In this step, the extraction of curvature information depends on the structural feature map generated in step S1, the acquisition of real-time position depends on the fusion positioning result in step S6, and the adjusted adsorption force data is continuously collected in step S4 and used for tactile positioning in step S5 and posture calculation in step S7.

[0148] S5. Based on the enhanced environmental data, identify the structural features of the ship's hull and match them with the structural feature map to obtain visual positioning information. At the same time, match the adsorption force data with the structural feature map to obtain tactile positioning information.

[0149] Furthermore, in S5, the steps for obtaining visual positioning information and tactile positioning information specifically include:

[0150] The enhanced environmental data is input into the visual feature extraction module to identify the structural features of the ship's hull.

[0151] The identified hull structural features are matched with pre-labeled structural feature information in the structural feature map. When a match is successful, the coordinate data of the structural feature information in the structural feature map is obtained as visual positioning information.

[0152] Simultaneously, the adsorption force fluctuation feature sequence generated by passing through the ship's structural features is extracted from the adsorption force data. The adsorption force fluctuation feature sequence is then matched with the structural feature information at the corresponding location in the structural feature map. When the pattern match is successful, the coordinate data of the corresponding structural feature information in the structural feature map is obtained as tactile positioning information.

[0153] Specifically, step S5 executes two branches in parallel: visual positioning information acquisition and tactile positioning information acquisition. In the visual positioning information acquisition branch, the enhanced close-up image of the work surface from the enhanced environmental data acquired in step S4 is input into the visual feature extraction module. The visual feature extraction module is a pre-trained convolutional neural network model. This model uses a residual network as its backbone and is trained on an underwater hull image dataset containing labeled hull structural features, employing a cross-entropy loss function and a stochastic gradient descent optimizer during training. The visual feature extraction module infers from the input close-up enhanced image of the work surface and outputs the type of identified hull structural feature and its pixel coordinates in the image coordinate system. The identified hull structural feature types include at least one of welds, corrosion pits, and marine organism attachment areas. The pixel coordinates of the identified hull structural features are transformed to the hull coordinate system using the known geometric relationship between the climbing robot and the work surface, obtaining the three-dimensional coordinates of the feature in the hull coordinate system. The three-dimensional coordinates of the visually identified hull structural features in the hull coordinate system are then matched with the pre-labeled structural feature information in the structural feature map. Matching is determined using Euclidean distance as the metric, calculated as follows:

[0154] ;

[0155] in, The three-dimensional coordinates of the visual recognition features in the ship's coordinate system. These are the 3D coordinates of the pre-labeled features in the structural feature map. When the Euclidean distance... Less than the preset visual matching threshold When the match is successful, it is determined that the match is successful. Typical values ​​range from 5 mm to 10 mm. After a successful match, the coordinates of the structural feature information in the structural feature map are obtained as visual positioning information. The visual positioning information is the absolute coordinates in the ship's coordinate system.

[0156] In the tactile positioning information acquisition branch, the adsorption force fluctuation feature sequence caused by passing through the structural features of the hull is extracted from the adsorption force data acquired in step S4. When the wall-climbing robot travels through structural features such as ribs or welds, the working gap of the magnet unit changes due to the abrupt curvature of the hull surface, causing significant fluctuations in the adsorption force value collected by the adsorption force sensor, forming a fluctuation data segment with a characteristic pattern. This fluctuation data segment is extracted into an adsorption force fluctuation feature sequence F(k) in chronological order. , Let F(k) be the sequence length, and F(k) be the adsorption force value at the kth sampling point.

[0157] The adsorption force fluctuation feature sequence F(k) is pattern matched with the corresponding structural feature information in the structural feature map. Each structural feature type in the structural feature map has a pre-stored corresponding adsorption force fluctuation feature template M, which is a typical adsorption force fluctuation curve for that type of structural feature obtained through multiple field measurements or simulations. Pattern matching uses a cross-correlation function to calculate the similarity between the adsorption force fluctuation feature sequence and the pre-set fluctuation feature template.

[0158] ;

[0159] in, For the preset wave feature template at the position The value at that location, This represents the phase shift. The maximum value of the cross-correlation function is taken. As a similarity metric. When Exceeding the preset similarity threshold When the pattern match is successful, it is determined that the pattern is matched. The determination is made through calibration based on the characteristics of the magnetic adsorption unit and the type of ship structure.

[0160] After successful pattern matching, the coordinate data of the corresponding structural feature information in the structural feature map are obtained as tactile positioning information. The tactile positioning information is the absolute coordinate in the ship's coordinate system. After the visual positioning information and tactile positioning information are obtained, they are output together with the odometer data to step S6 for multi-source fusion positioning. The visual positioning information provides an absolute position reference based on visual feature recognition, and the tactile positioning information provides an absolute position reference based on the tactile perception of adsorption force. The two together constitute a calibration information source for the cumulative error of the encoder odometer data from different perceptual dimensions.

[0161] S6. By integrating mileage data, visual positioning information, and tactile positioning information, the real-time position of the wall-climbing robot is determined.

[0162] Furthermore, in S6, the steps for determining the real-time position of the wall-climbing robot specifically include:

[0163] Acquire mileage data, visual positioning information, and tactile positioning information at the same time.

[0164] Determine the confidence weights for mileage data, visual positioning information, and tactile positioning information respectively;

[0165] Based on confidence weights, the mileage data, visual positioning information, and tactile positioning information are weighted and fused to calculate the real-time position of the wall-climbing robot.

[0166] Specifically, it acquires mileage data, visual positioning information, and tactile positioning information at the same moment. This "same moment" refers to the sampling time corresponding to the synchronous multi-source data frame formed after timestamp alignment in step S4. The mileage data provides the displacement increment of the wall-climbing robot on the working surface since the previous sampling time. Based on the real-time location determined in the previous fusion cycle. The current location calculated from the mileage data is Visual positioning information and haptic positioning information These are the absolute coordinates in the ship's coordinate system output from step S5. The confidence weights for the mileage data, visual positioning information, and tactile positioning information are determined separately. The confidence weights are dynamically determined based on the data quality of each positioning information source at the current moment. The confidence weight for the mileage data... The detection method is based on whether a slippage event occurs in the drive wheels. The slippage event is detected by comparing the displacement obtained by double integration of the acceleration output from the inertial measurement unit with the mileage displacement output from the encoder. If the difference between the two exceeds a preset slippage threshold, the drive wheels are considered to be slipping. If no slippage is detected, ... Take a value between 0.2 and 0.4; when slippage is detected, Take a value between 0.05 and 0.1.

[0167] Confidence weight of visual positioning information Based on visual matching distance Sure. The Euclidean distance between the visual recognition features calculated in step S5 and the corresponding features in the structural feature map is given. The calculation formula is:

[0168] ;

[0169] in, This is the visual matching threshold in step S5. As shown in the formula, the smaller the matching distance, the higher the confidence weight of the visual positioning information.

[0170] Confidence weights of haptic positioning information Based on the cross-correlation peak value calculated in step S5 Sure. The calculation formula is:

[0171] ;

[0172] in, The similarity threshold in step S5, The peak value of the cross-correlation under ideal matching conditions is determined through calibration. As shown in the formula, the larger the peak value of the cross-correlation, the higher the confidence weight of the tactile localization information.

[0173] The three confidence weights mentioned above are normalized to satisfy the following:

[0174] ;

[0175] When visual or tactile positioning information becomes unavailable at a certain moment due to feature matching failure, its corresponding confidence weight is reset to zero, and the remaining available information sources are fused according to normalized weights. Based on the confidence weights, the position estimated from the odometer data, visual positioning information, and tactile positioning information are weighted and fused to obtain the real-time position of the wall-climbing robot at the current moment. :

[0176] ;

[0177] in, The real-time location determined in the previous fusion cycle. The displacement increment given for the current period of the mileage data. The absolute position given by visual positioning information. The absolute position given for haptic positioning information. All positions are represented in the ship's coordinate system.

[0178] This fusion calculation is performed once at the sampling time of each synchronous multi-source data frame, forming a continuous real-time position sequence. The resulting real-time position P is used for the adsorption force pre-aiming adjustment judgment in step S4 and the attitude calculation in step S7, and also serves as the basis for the next fusion cycle. Substitute the values ​​into the calculation.

[0179] S7. By fusing the enhanced environmental data, inertial data, and adsorption force distribution data of each adsorption unit, the attitude angle of the wall-climbing robot is calculated.

[0180] Furthermore, in S7, the specific steps for calculating the attitude angles of the wall-climbing robot include:

[0181] Acquire enhanced environmental data, inertial data, and adsorption force distribution data of each adsorption unit at the same time.

[0182] Based on the enhanced environmental data, the relative positional relationship between the wall-climbing robot and the surface of the ship is analyzed to obtain visual pose reference information;

[0183] Based on inertial data, the acceleration and angular velocity of each axis of the wall-climbing robot are analyzed to obtain inertial attitude reference information;

[0184] Based on the adsorption force distribution data of each adsorption unit, the differences in adsorption force on the left and right sides and the differences in adsorption force at the front and rear ends are analyzed to obtain tactile posture reference information.

[0185] By integrating visual attitude reference information, inertial attitude reference information, and tactile attitude reference information, the three-dimensional attitude angles of the wall-climbing robot are calculated.

[0186] Specifically, enhanced environmental data, inertial data, and adsorption force distribution data of each adsorption unit are acquired simultaneously. This simultaneous moment coincides with the sampling time of the synchronous multi-source data frame in step S6. The global enhanced image in the enhanced environmental data is acquired by the selfie stick camera and enhanced in step S3, providing a view of the relative position of the climbing robot as a whole and the hull surface. The inertial data is output by the inertial measurement unit, including compensated three-axis acceleration. and triaxial angular velocity The adsorption force distribution data consists of a set of data comprising the real-time adsorption force values ​​of each magnet unit at the current moment, totaling M adsorption force values, where M is the number of magnet units.

[0187] Visual pose reference information is analyzed based on enhanced environmental data. The global enhanced image is input into the image processing unit, which extracts the contour of the wall-climbing robot and identifies the extension direction of the hull surface in the image. The geometric angle between the contour of the wall-climbing robot and the extension direction of the hull surface is calculated to obtain the visual roll angle. and visual pitch angle The visual roll angle reflects the degree of tilt of the wall-climbing robot relative to the hull surface in the left-right direction, while the visual pitch angle reflects the degree of tilt of the wall-climbing robot relative to the hull surface in the forward-backward direction. Inertial attitude reference information is analyzed based on inertial data. Compensated three-axis acceleration is utilized. When the wall-climbing robot is in a quasi-static or low-dynamic condition, the inertial roll angle is calculated by the components of the gravity vector on each axis. and inertial pitch angle The calculation formula is:

[0188] ;

[0189] ;

[0190] When the wall-climbing robot is in a dynamic movement state, the accelerometer measurement includes a motion acceleration component. At this time, a complementary filtering algorithm is used to integrate and predict the attitude angle using the angular velocity data output by the gyroscope, and the attitude angle calculated by the accelerometer is used as a correction quantity to obtain the inertial roll angle and inertial pitch angle under dynamic conditions.

[0191] Tactile posture reference information is analyzed based on the adsorption force distribution data of each adsorption unit. Let the average adsorption force of all adsorption units on the left be... The average adsorption force of all adsorption units on the right is The average adsorption force of all adsorption units at the front end is The average adsorption force of all adsorption units at the back end is Tactile roll angle and tactile pitch angle The calculation formula is:

[0192] ;

[0193] ;

[0194] in, and This is the preset force-angle conversion factor, in degrees per Newton. and The calibration experiment determined that the corresponding difference in adsorption force was recorded on a standard inclined plane platform with a known tilt angle, and the conversion coefficient was obtained through linear fitting. When the adsorption force on the left side is greater than that on the right side, it indicates that the wall-climbing robot is tilted to the right, and the tactile roll angle is [not specified]. A positive value indicates that the wall-climbing robot is tilted forward when the front-end adhesion force is greater than the rear-end adhesion force, and the tactile pitch angle is positive. It is a positive value.

[0195] By fusing visual, inertial, and tactile attitude reference information, the three-dimensional attitude angles of the wall-climbing robot are calculated. These three-dimensional attitude angles include the roll angle. Pitch angle and yaw angle The roll and pitch angles are calculated using a weighted fusion method.

[0196] ;

[0197] ;

[0198] in, The roll angle fusion weighting coefficient satisfies ; For pitch angle fusion weighting coefficients, satisfying Each fusion weight coefficient is dynamically allocated based on the real-time reliability of the corresponding attitude reference information. The reliability of the inertial attitude reference information is determined by the degree of vibration interference from the accelerometer, the reliability of the visual attitude reference information is determined by the sharpness of the global enhanced image and the confidence of contour extraction, and the reliability of the tactile attitude reference information is determined by the noise level of the adsorption force data and the consistency of the adsorption force of each magnet unit.

[0199] Yaw angle Yaw angular velocity via inertial measurement unit Integrating the data and using the change in travel direction in the odometer data as a correction reference, the result is obtained after complementary filtering. The calculated real-time three-dimensional attitude angles are then obtained. The output is sent to step S8 for attitude deviation determination and correction. The three-dimensional attitude angle reflects the instantaneous spatial attitude of the wall-climbing robot on the hull surface and is a key basis for judging the adsorption stability and operational accuracy.

[0200] S8. When the attitude angle exceeds the preset threshold, adjust the motion state of the wall-climbing robot or start the thruster robot to correct the attitude.

[0201] Furthermore, in S8, the specific steps for adjusting the motion state of the wall-climbing robot or activating the thruster robot for attitude correction include:

[0202] The real-time three-dimensional attitude angles of the wall-climbing robot are obtained, and the real-time three-dimensional attitude angles are compared with the preset attitude angle thresholds.

[0203] When any axial angle in the real-time three-dimensional attitude angle exceeds the corresponding attitude angle threshold, determine the direction and amount of attitude deviation.

[0204] Based on the direction and amount of the attitude deviation, a correction strategy is determined. The correction strategy includes: adjusting the differential speed of the left and right drive wheel sets of the wall-climbing robot to generate a reverse correction torque, and / or activating the thruster robot to provide auxiliary thrust or auxiliary torque.

[0205] Continuously monitor the real-time three-dimensional attitude angles of the wall-climbing robot until they return to the preset attitude angle threshold range.

[0206] Specifically, the real-time three-dimensional attitude angles of the wall-climbing robot, including the roll angle, are obtained from step S7. Pitch angle and yaw angle The real-time 3D attitude angles along each axis are compared with preset attitude angle thresholds. The preset attitude angle thresholds include the roll angle threshold. Pitch angle threshold and yaw angle threshold The threshold values ​​are set based on the maximum tilt angle allowed for the wall-climbing robot's magnetic adsorption unit to maintain stable adhesion, including the roll angle threshold. and pitch angle threshold The yaw angle threshold can be set from 5 to 15 degrees. The setting can be from 10 to 20 degrees. When any axial angle in the real-time 3D attitude angle exceeds its corresponding attitude angle threshold, an attitude anomaly is determined, and the correction process begins. The determination condition for an attitude anomaly is that any of the following inequalities holds:

[0207] ;

[0208] in, The reference value for the expected yaw angle at this location is determined by the direction of the surface normal vector of the path point planned in step S1.

[0209] After determining an abnormal attitude, the direction and amount of the attitude deviation are assessed. The direction of the deviation is determined by the sign of the axial angle exceeding a threshold. Taking the roll angle as an example, when... When, it indicates that the wall-climbing robot is tilting to the right; when When the value is set to "tilt," it indicates that the wall-climbing robot is tilting to the left. The deviation is the portion exceeding the difference between the current angle and the corresponding threshold. Roll angle deviation. The calculation method is as follows:

[0210] ;

[0211] Pitch angle deviation and yaw angle deviation Calculate using the same method.

[0212] The correction strategy is determined based on the direction and magnitude of the attitude deviation. The selection of the correction strategy follows these principles: For roll and pitch deviations, the priority is to adjust the differential speed of the left and right drive wheel sets of the wall-climbing robot to generate a reverse corrective torque; when the differential speed correction effect is insufficient or the deviation exceeds the correction capability range of the drive wheel set differential speed, the thruster robot is activated to provide auxiliary thrust or auxiliary torque for collaborative correction; when the deviation is small and within the correctable range of the drive wheel set differential speed, only differential speed correction is used, and the thruster robot is not activated. When using drive wheel set differential speed correction, the speed difference between the left and right drive wheel sets... The relationship between the deviation and the amount of deviation is as follows:

[0213] ;

[0214] in, This is the proportional control coefficient. The differential control coefficient, This represents the rate of change of the roll angle. A positive value indicates that the left drive wheel is accelerating or the right drive wheel is decelerating. The opposite applies when the value is negative. Differential correction for pitch angle deviation follows the same method, adjusting the direction of the speed difference application to a front-to-rear drive relationship.

[0215] When the thruster robot is activated for auxiliary correction, it outputs auxiliary thrust or auxiliary torque based on the direction and amount of deviation. The magnitude of the auxiliary thrust... Determined by the deviation amount:

[0216] ;

[0217] in, This is the thrust ratio coefficient of the thruster. The direction of the auxiliary thrust is opposite to the direction of the attitude deviation, so that the corrective torque generated by the thruster pushes the wall-climbing robot back to its normal attitude range.

[0218] Once the correction strategy is determined, the integrated control system sends control commands to the drive wheel assembly and the propeller robot of the wall-climbing robot respectively to execute the correction actions. During the correction process, the real-time three-dimensional attitude angles calculated in step S7 are continuously monitored, and the corrected attitude angles are cyclically compared with preset attitude angle thresholds until... and If all conditions are met, the attitude is determined to have recovered to the preset attitude angle threshold range, and the correction is complete.

[0219] After the correction is completed, the wall-climbing robot resumes its normal movement along the work route, and the thruster robot reduces its thrust output or returns to standby mode to continue performing subsequent work tasks.

[0220] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for adsorption and positioning in maritime operations based on a wall-climbing robot, the method being applied to a maritime operation robot, the maritime operation robot comprising a wall-climbing robot and a thruster robot, characterized in that, The adsorption localization method includes the following steps: S1. Obtain a structural feature map of the target vessel and plan the operation route; S2. After combining the wall-climbing robot and the thruster robot, they are submerged in water, adsorbed onto the surface of the hull, and the positioning parameters are initialized. S3. Collect underwater environmental data and perform enhancement processing, and at the same time collect the adsorption force data of each adsorption unit of the wall-climbing robot; S4. Control the wall-climbing robot to move along the work route and acquire mileage data, inertial data, enhanced environmental data and adsorption force data; S5. Identify the structural features of the ship based on the enhanced environmental data, match them with the structural feature map to obtain visual positioning information, and simultaneously match the adsorption force data with the structural feature map to obtain tactile positioning information. S6. By integrating the mileage data, visual positioning information, and tactile positioning information, the real-time position of the wall-climbing robot is determined; S7. By integrating the enhanced environmental data, inertial data, and adsorption force distribution data of each adsorption unit, the attitude angle of the wall-climbing robot is calculated. S8. When the attitude angle exceeds the preset threshold, adjust the motion state of the wall-climbing robot or start the thruster robot to correct the attitude.

2. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S1, the steps of obtaining the structural feature map of the target vessel and planning the operation route specifically include: Obtain the general arrangement drawing and coordinate data of the target vessel, and establish a three-dimensional model of the target vessel; Structural feature information is extracted from the three-dimensional model, and the extracted structural feature information is annotated in the three-dimensional model to generate the structural feature map; The work area is set based on the three-dimensional model, and the work route is deployed automatically or semi-automatically.

3. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S2, the process of combining the wall-climbing robot and the propulsion robot, entering the water, adhering to the surface of the hull, and initializing the positioning parameters specifically includes: The wall-climbing robot and the thruster robot are combined using a detachable connection structure to form a combined unit; The assembly is placed in the water, and the thruster robot provides the thrust to help the assembly approach the working surface of the target ship hull. When the distance between the assembly and the working surface reaches the adsorption trigger condition, the magnetic adsorption unit of the wall-climbing robot is activated to adsorb the wall-climbing robot onto the working surface. The encoder of the wall-climbing robot is initialized to zero, and each sensor is activated to establish an initial state reference.

4. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S3, the steps of collecting underwater environmental data and performing enhancement processing, and simultaneously collecting the adsorption force data of each adsorption unit of the wall-climbing robot, specifically include: Activate the image acquisition device mounted on the wall-climbing robot to acquire raw underwater images; The original underwater image is enhanced to obtain the enhanced environmental data. At the same time, the adsorption force sensors installed on each adsorption unit are activated to collect real-time adsorption force data of each adsorption unit and establish an adsorption force distribution baseline.

5. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S4, the steps of controlling the wall-climbing robot to move along the work route and acquiring mileage data, inertial data, enhanced environmental data, and adhesion force data specifically include: Motion control commands are generated based on the work route to drive the drive wheel assembly of the wall-climbing robot to move along the work route; During the journey, the mileage data is acquired in real time through the encoder mounted on the wall-climbing robot, and the inertial data is acquired in real time through the inertial measurement unit. Simultaneously, it continuously receives the enhanced environmental data after the enhancement process, and continuously receives the adsorption force data fed back in real time from each adsorption unit.

6. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 5, characterized in that, The method also includes pre-adjusting the adsorption force parameters of each adsorption unit of the wall-climbing robot, specifically including: Based on the real-time position of the wall-climbing robot and the structural feature map, determine whether the wall-climbing robot has traveled to a preset front range of any pre-marked structural feature point; When it is determined that the traveler has entered the preset forward range, the curvature information of the structural feature point is extracted from the structural feature map; Based on the curvature information, the target adsorption force parameters of each adsorption unit are determined, and adjustment commands are sent to each adsorption unit to adjust the adsorption force to match the curvature of the ship ahead in advance.

7. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S5, the steps of obtaining visual positioning information and tactile positioning information specifically include: The enhanced environmental data is input into the visual feature extraction module to identify the ship's structural features. The identified hull structure features are matched with the pre-marked structure feature information in the structure feature map. When the match is successful, the coordinate data of the structure feature information in the structure feature map is obtained as the visual positioning information. Simultaneously, the adsorption force fluctuation feature sequence generated by passing through the ship's structural features is extracted from the adsorption force data. The adsorption force fluctuation feature sequence is then matched with the structural feature information at the corresponding position in the structural feature map. When the pattern matching is successful, the coordinate data of the corresponding structural feature information in the structural feature map is obtained as the tactile positioning information.

8. The method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S6, the step of determining the real-time position of the wall-climbing robot specifically includes: Acquire the mileage data, the visual positioning information, and the tactile positioning information at the same time. Determine the confidence weights of the mileage data, the visual positioning information, and the tactile positioning information respectively; Based on the confidence weights, the mileage data, the visual positioning information, and the tactile positioning information are weighted and fused to calculate the real-time position of the wall-climbing robot.

9. A method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S7, the step of calculating the attitude angles of the wall-climbing robot specifically includes: Acquire the enhanced environmental data, the inertial data, and the adsorption force distribution data of each adsorption unit at the same time. Based on the enhanced environmental data, the relative positional relationship between the wall-climbing robot and the surface of the ship is analyzed to obtain visual posture reference information; Based on the inertial data, the acceleration and angular velocity of each axis of the wall-climbing robot are analyzed to obtain inertial attitude reference information; Based on the adsorption force distribution data of each adsorption unit, the differences in adsorption force on the left and right sides and the differences in adsorption force at the front and rear ends are analyzed to obtain tactile posture reference information. By integrating the visual posture reference information, the inertial posture reference information, and the tactile posture reference information, the three-dimensional posture angles of the wall-climbing robot are calculated.

10. A method for adsorption and positioning in maritime operations based on a wall-climbing robot body according to claim 1, characterized in that, In step S8, the steps of adjusting the motion state of the wall-climbing robot or activating the thruster robot for attitude correction specifically include: The real-time three-dimensional attitude angles of the wall-climbing robot are obtained, and the real-time three-dimensional attitude angles are compared with a preset attitude angle threshold. When any axial angle in the real-time three-dimensional attitude angle exceeds the corresponding attitude angle threshold, the attitude deviation direction and deviation amount are determined. Based on the direction and amount of the attitude deviation, a correction strategy is determined. The correction strategy includes: adjusting the differential speed of the left and right drive wheel sets of the wall-climbing robot to generate a reverse correction torque, and / or activating the thruster robot to provide auxiliary thrust or auxiliary torque. The real-time three-dimensional attitude angle of the wall-climbing robot is continuously monitored until it returns to the preset attitude angle threshold range.

Images