Low-temperature cooling water cooling air-cooled island cooling robot path planning method and system

Abstract

The application discloses a low-temperature cooling water cooling air-cooled island cleaning robot path planning method and system, and belongs to the technical field of robot control, comprising: based on the air-cooled island fin layout drawing, initializing the workspace grid, setting the reference parameter; driving the cleaning robot to collect multi-dimensional data, calculating the actual friction coefficient of the measured grid through the main friction coefficient and the torque auxiliary friction coefficient, for the unmeasured grid, linear extrapolation is carried out through the gradient to construct the real-time friction field; obtaining the ideal contact force vector, constructing the kinematics model to obtain the joint required torque, and calculating the joint angle adjustment amount, generating the control instruction and executing through the feedforward friction compensation and PID feedback control, constructing the wear amount estimation model, obtaining the current estimated wear amount, and calculating the core index, generating the rating report, continuously improving the operation efficiency, realizing the unity of the efficiency and equipment safety of the air-cooled island cleaning.

Description

Technical Field

[0001] This invention relates to a path planning method and system for a cleaning robot in an air-cooled island using low-temperature cooling water, and belongs to the field of robot control technology. Background Technology

[0002] As the core heat dissipation device of the direct air-cooling system in thermal power plants, the air-cooled island plays a crucial role in condensing steam from the turbine exhaust into water. It consists of a large-scale array of finned tube bundles, fan groups, and supporting frames, and is exposed to the complex outdoor environment for a long time. Its equipment condition directly affects the safe and efficient operation of the power plant. Currently, the industry has explored a mobile robot cleaning solution based on cavitation bubble technology, which uses the cavitation effect to remove dirt from the surface of the fins. In actual operation, the robot needs to contact or maintain close contact with the fins through the cleaning device to transfer cavitation energy. This inevitably creates friction between the robot and the fins. At the same time, appropriate friction can help the robot maintain a close fit and suppress abnormal slippage.

[0003] However, existing technologies do not consider the impact of friction. Specifically, the thin and light fin structure makes it susceptible to scratches, deformation, or even breakage on the fin surface due to excessive friction, which damages the heat dissipation structure, reduces the heat exchange efficiency of the air-cooled island, and increases the energy consumption of the robot drive, shortens the endurance, and leads to task interruption. At the same time, the cleaning device components in contact with the fins will also wear down due to friction, which not only shortens their own lifespan but also interferes with the generation and effect of cavitation bubbles due to structural changes, indirectly reducing the cleaning quality. In addition, the nonlinear characteristics of friction can also interfere with the precision of robot motion control, making it difficult to accurately control the relative position and trajectory with the fins, which can easily lead to inadequate cleaning or excessive contact damage. These problems together restrict the efficiency of cleaning operations. Due to the lack of systematic control of friction, it is impossible to effectively avoid its damage to the fins and equipment itself and its adverse effects on energy consumption and control precision. It is also difficult to fully adapt to the dynamic changes in friction characteristics in complex outdoor environments, which greatly restricts the efficiency of cleaning operations and the safety protection of air-cooled island equipment. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a path planning method and system for an air-cooled island cleaning robot that uses low-temperature cooling water. The method constructs a cleaning network based on fin drawings and sets benchmark parameters to solve the problems of initial positioning deviation and improper contact force. It also constructs a real-time friction field through multi-sensor fusion to adapt to outdoor dynamic friction and dynamically adjusts the contact force and speed.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A path planning method for a cleaning robot operating in an air-cooled island with low-temperature cooling water includes:

[0007] Based on the air-cooled island fin layout drawing, initialize the workspace grid and set the baseline parameters;

[0008] The cleaning robot collects multi-dimensional data, calculates the actual friction coefficient of the measured grid by using the dominant friction coefficient and torque-assisted friction coefficient, and constructs a real-time friction field by linear extrapolation of the gradient for the unmeasured grid.

[0009] Obtain the ideal contact force vector, construct a kinematic model to obtain the joint torque requirement, and calculate the joint angle adjustment. Through feedforward compensation and PID feedback control, generate and execute control commands, construct a wear estimation model, obtain the current estimated wear amount, calculate core indicators, and generate a rating report.

[0010] Specifically, the steps for setting the baseline parameters include:

[0011] Obtain the vector information of the air-cooled island fins and the core parameters of the nozzle, and construct a fin vector information table using a coordinate system algorithm;

[0012] Based on the fin row spacing and nozzle coverage width, spatial uniform discretization is adopted to divide the working space into grids and assign semantic labels to divide the cleaning area, welding point area, and non-cleaning boundary to generate a cleaning grid.

[0013] The friction coefficient under rated driving pressure and clean fin scenario was obtained through a wear test platform, and the initial friction coefficient was calculated.

[0014] A baseline control set is generated by combining the set nozzle wear threshold and initial spray parameters.

[0015] Specifically, the steps for assigning semantic tags include:

[0016] For each workspace grid, calculate the boundary distance. Safe distance And set a safe distance threshold. Each workspace grid is assigned a semantic label.

[0017] like Marked as the area to be cleaned, with semantic tags as ;

[0018] like Marked as the welding point area, with the semantic tag as ;

[0019] Otherwise, it is marked as an unwashed boundary, with the semantic label being... .

[0020] Specifically, the steps for calculating the actual friction coefficient of the measured mesh include:

[0021] Collect multi-dimensional data, perform preprocessing and timestamp alignment to generate a dynamic dataset, and calculate the dominant friction coefficient and torque-assisted friction coefficient.

[0022] Using a fouling identification model, the type of fouling on the fins is identified. Combined with a coefficient correction table, the friction correction coefficient is obtained, and the torque-assisted friction coefficient is corrected.

[0023] The dominant friction coefficient and the corrected torque-assisted friction coefficient are weighted and calculated to determine the actual friction coefficient and the friction deviation, so as to determine whether the actual friction coefficient is effective.

[0024] If the friction deviation exceeds the friction comparison threshold, data re-sampling is triggered, and continuous data acquisition continues. Next, if it exists If the friction deviation is greater than the friction comparison threshold, the actual friction coefficient is determined to be valid; otherwise, the sensor is determined to be abnormal.

[0025] If the friction deviation is less than or equal to the friction comparison threshold, the actual friction coefficient is determined to be valid.

[0026] Specifically, the steps for calculating the actual friction coefficient of the measured mesh also include:

[0027] The actual friction coefficient is assigned to the corresponding grid, and high-sensitivity areas are selected. The high-sensitivity friction coefficient is calculated using a damage weighting factor. The high-sensitivity area is the area where the height difference is greater than the damage threshold.

[0028] Obtain the dirt coverage, filter out the grids whose dirt coverage is greater than the standard coverage, and correct the actual friction coefficient using the friction correction coefficient;

[0029] Set a secondary adaptation threshold to determine the environmental category of the current air-cooled island based on ambient humidity, and configure the diffusion coefficient;

[0030] The actual friction coefficient is smoothed to construct the measured grid friction field, and the friction fluctuation value is calculated.

[0031] If the friction fluctuation value is greater than the fluctuation limit, the diffusion coefficient is reduced by adjusting the diffusion coefficient, and the measured grid friction field is updated.

[0032] If the friction fluctuation value is not greater than the lower limit of fluctuation, the diffusion coefficient is increased by the diffusion adjustment coefficient, and the measured grid friction field is updated.

[0033] Specifically, the steps for constructing a real-time friction field include:

[0034] Traverse the cleaned grid, divide it into measured grid and unmeasured grid, filter out the nearest measured grid of the unmeasured grid by Euclidean distance, and obtain the nearest Euclidean distance;

[0035] Calculate the friction coefficient gradient of the most recently measured grid, and combine it with the coordinate difference to calculate the extrapolation coefficient;

[0036] Based on the actual friction coefficient of the recently measured grid, calculate the extrapolated friction coefficient of the unmeasured grid;

[0037] Once the nearest Euclidean distance exceeds the reliable extrapolation range, the extrapolated friction coefficient is corrected based on the initial friction coefficient to generate a real-time friction field.

[0038] Calculate the mean and standard deviation of the friction coefficient of all grids in the real-time friction field, and compare them with the reasonable range of friction coefficient;

[0039] If the mean friction coefficient exceeds the preset reasonable range, the multi-source fusion weight readjustment is triggered; if the standard deviation of the friction coefficient is greater than the upper limit of friction fluctuation, the missing high-sensitivity areas are supplemented and marked; otherwise, the real-time friction field is globally consistent.

[0040] Specifically, the steps for constructing a real-time friction field also include:

[0041] Once the cleaning robot moves to the extrapolated, unmeasured grid, it obtains the actual friction coefficient and calculates the local measurement error for local verification.

[0042] If the local measured error is not greater than the extrapolated accuracy, then the real-time friction field is deemed valid.

[0043] If the local measured error is greater than the extrapolation accuracy, the friction coefficient gradient is corrected by configuring gradient weights, and extrapolation is re-performed on other unmeasured grids.

[0044] Specifically, the steps for generating control commands include:

[0045] Obtain the friction coefficients of the current contact mesh and its surrounding adjacent meshes, and calculate the local friction mean.

[0046] Configure the constraint range of the contact force and calculate the ideal contact force vector;

[0047] Construct a positive kinematics model and calculate the Jacobian matrix based on the current robot joint angles to generate the required joint torque and calculate the joint angle adjustment amount;

[0048] Get the robot's current cleaning path and the area in front. The friction coefficient sequence of each grid is used to calculate the forward sliding mean and standard deviation, calculate the velocity adjustment factor, obtain the initial velocity benchmark, and calculate the target velocity.

[0049] Obtain the frictional resistance that is currently expected to be offset, and generate feedforward compensation torque;

[0050] The target and threshold of feedback control are set, and based on the reference control set, the parameters of the feedback controller are set using the PID control algorithm to generate control commands.

[0051] Specifically, the steps for generating a rating report include:

[0052] A wear estimation model is constructed to obtain the estimated wear at the current moment. Once the estimated wear exceeds the nozzle wear threshold, a shutdown maintenance is triggered.

[0053] Calculate the cleanliness change rate; if the cleanliness change rate is less than the minimum cleaning effect threshold, the cavitation effect is deemed insufficient.

[0054] Real-time calculation of cleaning energy consumption per unit area, prediction of remaining cleanable area, and charging planning;

[0055] Calculate cleaning quality indicators, equipment protection indicators, and energy efficiency indicators, set rating standards for each indicator, and generate rating reports.

[0056] A path planning system for a cleaning robot in an air-cooled island using low-temperature cooling water includes: an initialization module, a friction assessment module, a control module, and an assessment module.

[0057] The initialization module is used to construct a cleaning network with semantic labels, set a reference control set, calibrate a reference fin column to determine the robot's initial pose, activate all sensors, and set the synchronous acquisition frequency.

[0058] The friction evaluation module is used to calculate the actual friction coefficient and perform smoothing, calculate the extrapolated friction coefficient of the unmeasured grid based on the neighbor gradient, and construct the real-time friction field.

[0059] The control module is used to extract the average local friction value, obtain the joint required torque and joint angle adjustment amount, and integrate feedforward compensation and PID feedback to generate control commands.

[0060] The evaluation module is used to parse and execute control commands, construct a wear estimation model, calculate the cleanliness change rate to determine the cavitation effect, predict the remaining washable area for charging planning, and generate a rating report based on cleaning quality indicators, equipment protection indicators, and energy efficiency indicators.

[0061] The beneficial effects of this invention are:

[0062] By constructing a cleaning network through drawing analysis and coordinate unification, and setting benchmark parameters through material experiments, the system avoids robots accidentally touching non-cleaning areas due to spatial confusion, and provides safety control anchors during the startup phase. This prevents excessive contact force from scratching the fins or insufficient force from affecting the cleaning effect, while also avoiding ineffective energy consumption and ensuring that the initial control is within a safe and efficient range. By constructing a real-time friction field and calculating the actual friction coefficient through multi-source sensor fusion, combined with environmental dynamic smoothing and extrapolation, the system effectively adapts to dynamic variables such as rainfall and scaling in outdoor environments, solving the problem that initial values ​​cannot cover complex environments. When generating control commands, the system dynamically adjusts the contact force and speed based on the local mean of the friction field, with feedforward compensation offsetting friction energy consumption and PID feedback correcting deviations to avoid inadequate cleaning or excessive contact. During the execution evaluation phase, wear monitoring and cavitation optimization prevent the decline in cleaning quality caused by component wear, endurance management avoids task interruption, and performance index rating achieves closed-loop optimization, continuously improving operational efficiency and achieving a balance between high efficiency and equipment safety in air-cooled island cleaning. Attached Figure Description

[0063] Figure 1 A flowchart of a path planning method for an air-cooled island cleaning robot to cool low-temperature cooling water;

[0064] Figure 2 This is a flowchart of setting the reference parameters in this invention;

[0065] Figure 3 This is a flowchart of the process for constructing a real-time friction field in this invention;

[0066] Figure 4 This is a flowchart of the process for generating control instructions in this invention. Detailed Implementation

[0067] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0068] Example 1:

[0069] refer to Figures 1 to 4 As shown in the figure, this embodiment introduces a path planning method for a cleaning robot in an air-cooled island using low-temperature cooling water, including:

[0070] Step S1: Obtain the air-cooled island fin layout drawing, clarify the fin row position, spacing, and welding point distribution to initialize the workspace grid of the cleaning robot, avoiding accidental contact with non-cleaning areas due to unclear spatial perception. Since the robot has not yet received real-time sensor data during the startup phase, there is no reference to the actual friction and contact state. Based on the fin material and cleaning mode, set the reference parameters as the initial control anchor point to avoid excessive contact force that scratches the fins or insufficient contact force that affects the cleaning effect due to the lack of parameter basis. At the same time, prevent energy waste and ensure that the control parameters are within a safe and efficient range during startup, and drive the cleaning robot to perform cleaning operations. Calibrate the reference fin row to determine the robot's initial pose relative to the reference fin row, avoiding trajectory deviation caused by initial positioning deviation. Activate all sensors deployed on the cleaning robot and set the sensor data synchronous acquisition frequency to achieve synchronous acquisition of multi-source data.

[0071] Specifically, the steps for setting the baseline parameters include:

[0072] Obtain the fin layout drawing of the current air-cooled island, and use analytical tools to extract the vector information of the fins, including the coordinates of the fin column boundary, fin spacing, and welding point center coordinates. Obtain the effective spray radius and optimal working distance from the nozzle technical manual to generate the nozzle core parameters. Since the drawing uses the design coordinate system, the nozzle core parameters are based on the nozzle local coordinate system, and the robot motion is based on the driving base coordinate system, a homogeneous coordinate transformation method is introduced. Based on the transformation matrix, all fin vector information and nozzle core parameters are uniformly mapped to the driving base coordinate system to ensure that the position data benchmark is consistent. This generates a fin vector information table unified to the driving base coordinate system, including fin column boundary, spacing, welding point coordinates, and nozzle core parameters.

[0073] Based on the fin vector information table, and using the fin column spacing and nozzle coverage width as criteria, spatial uniform discretization is employed to divide the workspace into grids. This ensures that the size of each grid matches the effective cleaning range of a single nozzle stroke, improving cleaning efficiency. Simultaneously, each workspace grid is assigned a semantic label to delineate the cleaning area, welding point area, and non-cleaning boundary, ensuring the robot can identify hazardous areas and avoid accidental contact with welding points or fin edges. This includes calculating the boundary distance for each workspace grid using the Euclidean distance from the grid center to the nearest fin edge and the Euclidean distance from the grid center to the nearest welding point. Safe distance And set a safe distance threshold. Each workspace grid is assigned a semantic label, ultimately generating a spatially discretized cleaning network with semantic labels; if If the mesh is simultaneously far from both the weld point and the fin edge, then the workspace mesh is marked as the area to be cleaned, with the semantic label being... The cleaning head will not touch the welding points or fin edges in this area, ensuring equipment safety and cleaning efficiency; if If the mesh is close to the welding point, then the workspace mesh is marked as the welding point area, and the semantic label is... Otherwise, mark the workspace mesh as an unwashed boundary, with the semantic label as follows: ;

[0074] Using a wear test platform, we simulated robot driving scenarios, set different driving pressures and different fin surface conditions, tested the friction coefficient between the driving wheel and the fin surface, extracted the friction coefficient data under rated driving pressure and clean fin scenarios, and calculated the initial friction coefficient using the arithmetic mean formula to ensure that the friction reference for driving control in the initial stage conforms to normal working conditions and avoid excessive friction of the driving wheel leading to fin scratches or wheel wear.

[0075] Since nozzle orifice enlargement and nozzle deformation can lead to a decrease in spray pressure and a shift in spray range, affecting the cleaning effect, a nozzle wear threshold is set, including the maximum wear amount of the orifice and the upper limit of the spray pressure attenuation. The initial spray flow rate of the nozzle is calculated using the orifice outflow formula, and the minimum effective spray pressure is set in combination with the cleaning process requirements to generate initial spray parameters. If the nozzle is a contact cleaning type, the upper limit of the contact force between the nozzle and the fin is calculated using the material mechanics stress formula, with the fin yield strength as a constraint, to avoid excessive contact pressure that accelerates nozzle wear and fin deformation. The initial friction coefficient, nozzle wear threshold, and initial spray parameters are integrated to generate a baseline control set to ensure that the control parameters during the start-up phase cover the three core requirements of friction, nozzle, and contact safety.

[0076] Based on the workspace grid, an initial action position is randomly selected and defined as the reference fin column. The robot is driven to move along the workspace grid path to the vicinity of the reference fin column, ensuring that the robot's vision sensor can clearly capture the edge features of the reference fin column. The robot's onboard vision sensor and IMU are then activated. The vision sensor acquires images of the reference fin column, and the two-dimensional feature coordinates of the reference fin column are extracted using an edge detection algorithm. The IMU acquires the robot's current posture data, and a vision-IMU joint calibration algorithm is used to fuse the two-dimensional feature coordinates of the vision features with the posture data of the IMU to calculate the robot's initial pose matrix relative to the reference fin column, including position coordinates and attitude angles, thus avoiding initial positioning deviations that could lead to subsequent cleaning trajectory offsets.

[0077] Step S2: During the robot's movement, all sensors are driven to collect multi-dimensional data in real time. The actual friction coefficient at the contact point is calculated using the dominant friction coefficient and torque-assisted friction coefficient. Based on the robot's real-time pose, the actual friction coefficient is assigned to the corresponding grid. Using the discretized diffusion equation and combined with the ambient humidity, the measured grid is smoothed. The diffusion coefficient is dynamically adjusted to balance noise reduction and the true trend, resulting in a smooth friction field of the measured grid. For the unmeasured grid, linear extrapolation is performed based on the gradient of the actual friction coefficient value of the measured grid to supplement the friction coefficient value of the unmeasured grid, so as to construct a real-time friction field covering the current working area and ensure the accuracy and comprehensiveness of friction perception.

[0078] Specifically, the steps for constructing a real-time friction field include:

[0079] Since the initial values ​​only represent ideal working conditions, and the outdoor environment of the air-cooled island has a large number of variables, such as sudden rainfall causing the fins to become slippery, uneven local scaling thickness, and sudden temperature changes affecting the frictional properties of the material, the initial values ​​cannot cover these dynamic changes. Based on the activated full range of sensors, multi-dimensional data is collected, including: force vector and torque vector of the current contact point are collected based on the end six-dimensional force sensor to obtain contact force data; real-time torque and speed of the driving motor and cleaning arm motor are collected based on the motor current sensor and torque sensor to obtain drive load data; fin surface flatness is collected using the laser displacement sensor; fin surface image is collected using the vision sensor to obtain fin state data; and current ambient humidity and temperature are collected using the temperature and humidity sensor to obtain environmental data. The collected multi-dimensional data is preprocessed and timestamps are aligned simultaneously to avoid fusion errors caused by time sequence misalignment, thereby generating a dynamic dataset.

[0080] Based on the pre-processed contact force, the tangential and normal components are separated to calculate the dominant friction coefficient. According to the motor torque balance relationship, the friction resistance is inversely calculated. Combined with the normal force of the motor drive, the torque-assisted friction coefficient is calculated based on the ratio of friction resistance to normal force. Among them, the friction resistance is the difference between the motor output torque and the load torque.

[0081] Using the constructed dirt recognition model, the collected image data is processed to identify the type of dirt on the fins, such as dust and scale. A preset coefficient correction table is consulted to obtain the friction correction coefficient of the current contact point, and the torque-assisted friction coefficient is corrected. The coefficient correction table is used to store the friction correction coefficients corresponding to different dirt types, which are obtained through a wear test platform.

[0082] Considering the differences in sensor reliability under different working conditions, such as the motor torque data being more reliable in clean areas and the visual correction being more critical in dirty areas, the dominant friction coefficient and the corrected torque-assisted friction coefficient are weighted and fused to calculate the actual friction coefficient. The friction deviation between the actual friction coefficient and the initial friction coefficient is also calculated to determine whether the actual friction coefficient is effective.

[0083] If the friction deviation exceeds the friction comparison threshold, data resampling is triggered, and continuous data acquisition continues. Next, if it exists If the friction deviation exceeds the friction comparison threshold, the actual friction coefficient is considered valid. In this case, it only indicates severe wear. If no friction deviation exists... If the friction deviation exceeds the friction comparison threshold, the sensor is deemed malfunctioning; if the friction deviation is less than or equal to the friction comparison threshold, the actual friction coefficient is deemed valid. ;

[0084] Based on the initial pose calibrated by vision-IMU and the encoder data update, the robot's real-time pose is obtained, the coordinates of the current contact point in the cleaning network are determined, and the calculated actual friction coefficient is assigned to the corresponding mesh to overwrite the default friction coefficient at initialization, thereby realizing the dynamic update of the mesh friction properties.

[0085] Based on the preprocessed laser displacement sensor data, the surface flatness of the fins is analyzed. If the height difference in a local area is greater than the damage threshold, it is determined that there are scratches / deformations. The corresponding grid is marked as a high-sensitivity area. A damage weighting factor is introduced. The high-sensitivity friction coefficient is obtained by multiplying the damage weighting factor with the actual friction coefficient and adding the actual friction coefficient. The high-sensitivity area is marked in red on the grid mapping map. At the same time, the coordinate range of the high-sensitivity area is recorded for subsequent trajectory planning to avoid or reduce the speed.

[0086] Based on the fin surface image, the dirt coverage is obtained, and the grids with dirt coverage greater than the standard coverage are screened out. The actual friction coefficient is corrected using the friction correction coefficient. The product of the friction correction coefficient and the actual friction coefficient is superimposed on the actual friction coefficient to ensure that the friction coefficient is consistent with the actual surface state of the fin.

[0087] Measurement data from a single grid contains noise, such as random sensor errors. Smoothing is used to eliminate fluctuations and make the friction field more consistent with physical continuity. A secondary adaptation threshold for environmental judgment is set, and the environmental category of the current air-cooled island is determined based on the collected ambient humidity, including dry, moderate humidity, and humid environments. Different diffusion coefficients are configured for each type of environment to obtain the diffusion coefficient under the current environment. A physics-based diffusion equation is used to smooth the actual friction coefficient, eliminating friction field fluctuations caused by noise from single-grid measurements. Since the cleaning grid is discrete, the continuous diffusion equation is discretized using the finite difference method. The smoothed friction coefficient is iteratively calculated in the time and spatial domains and mapped to the cleaning grid to obtain a smooth and realistic measured grid friction field, avoiding subsequent control misjudgments caused by noise.

[0088] Friction fluctuation values ​​are calculated by taking the absolute difference between the friction coefficients before and after smoothing the same grid. If the friction fluctuation value is greater than the upper limit of fluctuation, the fluctuation is too large, indicating oversmoothing. In this case, the diffusion coefficient is reduced using a preset diffusion adjustment coefficient, and the product of the diffusion adjustment coefficient and the original diffusion coefficient is used to replace the original diffusion coefficient. The measured grid friction field is then updated. If the friction fluctuation value is not greater than the lower limit of fluctuation, the fluctuation is too small, indicating ineffective noise reduction. In this case, the diffusion coefficient is increased using the diffusion adjustment coefficient, and the ratio of the original diffusion coefficient to the diffusion adjustment coefficient is used to replace the original diffusion coefficient. The measured grid friction field is then updated to ensure that the smoothed data is both noise-free and retains the true friction change trend. If the friction fluctuation value is between the lower and upper limits of fluctuation, the original diffusion coefficient is maintained. The diffusion adjustment coefficient is less than 1.

[0089] Friction coefficients cannot be directly measured in areas not reached by the robot. For these unmeasured areas, a linear extrapolation method based on the gradient of neighboring grids is used to predict the friction coefficients, forming a complete friction field. The cleaning grid is traversed, and grids not touched by the robot and without actual friction coefficient settings are marked (i.e., unmeasured grids), representing the cleaning areas not yet reached by the robot. Based on the fin layout diagram, unmeasured grids are classified according to the same fin column and adjacent fin columns. The regularity of the fin structure is used to improve extrapolation accuracy. For each unmeasured grid, its four neighboring measured grids are searched, and the friction coefficient gradient of the measured grids is calculated, including the gradient along the fin column direction. Gradient in the direction perpendicular to the fin row Based on the grid center coordinates, the Euclidean distance between the unmeasured grid and each neighboring grid is calculated, the nearest measured grid is selected, and the coordinate difference between the unmeasured grid and the nearest measured grid is calculated. , The extrapolation coefficients are obtained by multiplying the coordinate differences by their corresponding gradients and then summing the results. Based on the sum of the friction coefficient and extrapolation coefficient of the recently measured mesh, the extrapolation friction coefficient of the unmeasured mesh is calculated.

[0090] Simultaneously, extrapolation accuracy constraints are implemented. If the distance between the unmeasured grid and the nearest measured grid exceeds the reliable extrapolation range, the extrapolated friction coefficient of the unmeasured grid is corrected by a weighted algorithm based on the initial friction coefficient to avoid excessive errors caused by excessive extrapolation distance, and finally a complete real-time friction field is generated.

[0091] The mean and standard deviation of the friction coefficient of all grids in the real-time friction field are calculated and compared with the reasonable range of friction coefficient. If the mean friction coefficient exceeds the preset reasonable range, it is determined that there is a systematic deviation in the real-time friction field, triggering the multi-source fusion weight readjustment, such as increasing the motor torque weight, and checking whether the force sensor is offset. If the standard deviation of the friction coefficient is greater than the upper limit of friction fluctuation, the friction field fluctuation is too large, and there are unmarked high-sensitivity areas. In this case, damage detection is re-executed to supplement and mark the missed high-sensitivity areas. Otherwise, the real-time friction field is globally consistent.

[0092] Since the cleaning robot will move to the next grid according to the cleaning path after completing the cleaning of the current grid, once the cleaning robot moves to the extrapolated unmeasured grid, it will obtain the actual friction coefficient. Based on the absolute difference between the actual friction coefficient and the extrapolated friction coefficient, it will compare with the actual friction coefficient to calculate the local measurement error for local verification.

[0093] If the local measured error is not greater than the extrapolation accuracy, the real-time friction field is deemed valid; if the local measured error is greater than the extrapolation accuracy and the extrapolation accuracy is insufficient, the gradient of the friction coefficient is corrected by configuring gradient weights, and extrapolation is re-executed on other unmeasured grids.

[0094] Step S3: Extract the friction coefficients of the current contact grid and adjacent grids to calculate the local mean. Combine the contact force constraint range to determine the ideal contact force vector, avoid misjudgment due to single data fluctuations, and balance fin protection and cleaning efficiency. Construct a kinematic model, map the ideal contact force vector to the joint demand torque through the Jacobian matrix, and calculate the joint angle adjustment. Generate control commands through feedforward compensation and PID feedback control to ensure stable operation of the actuator and avoid fin scratches and inadequate cleaning.

[0095] Specifically, the steps for generating control commands include:

[0096] From the real-time friction field, the friction coefficients of the robot's current contact grid and surrounding adjacent grids are extracted, and the local friction mean is calculated to avoid misjudgment caused by fluctuations in data from a single grid.

[0097] The upper limit of fin contact force and the minimum cleaning tangential force are called from the reference control set. Combined with the local friction average, the lower limit of contact force constraint is calculated. Finally, the constraint range of contact force is determined, including the upper limit of contact force constraint and the lower limit of contact force constraint, to ensure that the contact force adjustment is always within a safe and effective range.

[0098] Traverse the real-time friction field and read the maximum and minimum friction coefficients as boundary references for adjusting the ideal contact force;

[0099] Using the difference between the maximum friction coefficient and the local average friction value as the numerator and the difference between the maximum friction coefficient and the minimum friction coefficient as the denominator, the ideal normal correction coefficient is calculated. Combined with the upper limit of the contact force constraint, the ideal normal force is calculated to achieve dynamic adaptation of reducing the normal force in the high friction zone to protect the fins and increasing the normal force in the low friction zone to protect the cleaning process.

[0100] Based on the product of minimum cleaning tangential force, average local friction and ideal normal force, the maximum value is selected as the ideal tangential force, which balances cleaning effect and avoids ineffective energy consumption.

[0101] The fin flatness detected by the laser displacement sensor is used to determine the normal unit vector of the fin surface, and the tangential unit vector is determined by combining it with the robot's preset cleaning motion direction, and the ideal contact force vector is synthesized.

[0102] Load the robot DH parameter table for initial pose calculation, including link length, joint offset, and joint angle range. Construct a forward kinematics model of the cleaning arm, determine the position mapping relationship of the end effector (spray head) relative to the robot base coordinate system, and calculate the Jacobian matrix of the cleaning arm based on the robot kinematic formula according to the current robot joint angles to describe the mapping relationship between end effector force and joint torque. By transposing the Jacobian matrix, the ideal contact force vector is mapped to the joint torque requirement in joint space to ensure that the joint torque matches the contact force requirement. The joint angles are obtained by updating the initial pose combined with encoder data.

[0103] The stiffness matrix of the robot mechanism in the wear test platform is obtained. To avoid vibration or overtravel caused by excessive joint adjustment, the joint torque requirement is converted into joint angle adjustment amount through the stiffness matrix. At the same time, joint angle limit constraint is introduced. If the joint angle adjustment amount causes the joint to exceed the limit constraint, the adjustment amount is reduced proportionally until the limit constraint is met.

[0104] Based on the forward kinematics model, the position deviation of the adjusted end effector is calculated. If the position deviation is less than the adjustment accuracy, the joint angle adjustment is confirmed to be effective. Otherwise, the Jacobian matrix is ​​recalculated to correct the joint angle error until the position deviation meets the standard, so as to avoid the end position offset leading to inadequate cleaning or excessive contact.

[0105] Based on the cleaning network, obtain the path of the robot currently cleaning ahead. The friction coefficient sequence of each grid is used, while grids semantically labeled as weld points or uncleaned boundaries are excluded. The look-ahead moving average of the friction coefficient along the forward path is then calculated. and standard deviation Based on the initial friction coefficient With forward moving average The ratio is used to calculate the forward correction factor, and the volatility attenuation coefficient is used to apply it. with standard deviation The product of the two factors constructs the index adjustment factor, and the velocity adjustment factor is calculated by multiplying the forward correction factor and the index adjustment factor. To obtain an initial velocity reference, calculate the target velocity to ensure that the velocity matches the friction characteristics ahead, balancing cleaning efficiency and equipment safety; the expression is as follows:

[0106]

[0107] In the formula, This is a redundant term to prevent division by zero.

[0108] The actual friction coefficient of the current grid is extracted from the real-time friction field and multiplied with the ideal normal force to obtain the friction resistance that is expected to be offset. The friction resistance is then converted into the feedforward compensation torque of the motor drive end and directly injected into the motor driver to offset friction interference in advance, thereby reducing drive energy consumption and trajectory deviation.

[0109] By calling sensor data, setting the target and threshold of feedback control, and using a PID control algorithm based on the reference control set, the parameters of the feedback controller are set. The joint angle adjustment, target speed, feedforward compensation torque and feedback fine-tuning are integrated to generate the final control command.

[0110] Step S4: Execute control commands, build a wear estimation model, obtain the current estimated wear amount, and immediately trigger a shutdown maintenance once the estimated wear amount exceeds the nozzle wear threshold to effectively avoid excessive wear of parts affecting operations. Calculate core indicators, generate a rating report, and provide a basis for operational efficiency evaluation.

[0111] Specifically, the steps for generating a rating report include:

[0112] The system receives control commands and uses the robot's underlying drive protocol to parse joint angle commands into pulse signals for servo motors, speed commands into motor speed setpoints, and torque commands into current control signals, ensuring that the actuators accurately receive action commands.

[0113] The rate of change of friction coefficient is calculated based on historical data of actual friction coefficient and real-time friction coefficient at the current moment.

[0114] Based on the initial friction coefficient A wear estimation model is constructed to obtain the estimated wear amount at the current moment, and a nozzle wear threshold is obtained. Once the estimated wear amount exceeds the nozzle wear threshold, a shutdown maintenance is triggered. The expression is as follows:

[0115]

[0116] In the formula, for Estimated wear and tear at any given time. This is the initial wear amount. The wear coefficient is... , for Real friction coefficient and real speed at all times;

[0117] Images of the fin surface after cleaning are collected by a visual sensor. Based on the cleanliness before and after cleaning, the cleanliness change rate is calculated. If the cleanliness change rate is less than the minimum cleaning effect threshold, the cavitation effect is deemed insufficient.

[0118] Based on cumulative energy consumption and the area already cleaned, the cleaning energy consumption per unit area is calculated in real time. Combined with the remaining power and battery efficiency, the remaining cleanable area of ​​the cleaning robot is predicted. If the remaining cleanable area can only complete less than 50% of the cleaning of the current area, the operation is paused and a return charging path is planned to avoid task interruption.

[0119] Based on the number of cleaning grids and the cleanliness change rate of each grid, cleaning quality indicators are calculated. Based on the number of times the contact force exceeds the limit, the number of times the motor current exceeds the limit, and the total number of monitoring, equipment protection indicators are calculated. Energy efficiency indicators are obtained based on the cleaning energy consumption per unit area, and rating standards are set for each indicator to generate a rating report.

[0120] Example 2:

[0121] Another embodiment of the present invention provides: a path planning system for an air-cooled island cleaning robot with low-temperature cooling water cooling, comprising: an initialization module, a friction evaluation module, a control module, and an evaluation module;

[0122] The initialization module is used to acquire the air-cooled island fin layout drawing, extract vector information, and combine it with nozzle parameters to unify it into the driving base coordinate system through coordinate transformation. It constructs a cleaning network with semantic labels and sets the reference control set in conjunction with experiments. The reference fin column is calibrated to determine the robot's initial pose, activates all sensors and sets the synchronous acquisition frequency to avoid accidental contact with non-cleaning areas due to spatial perception ambiguity, prevents improper initial contact force from scratching the fins or affecting the cleaning effect, and provides a safe and efficient reference framework for subsequent operations.

[0123] The friction assessment module is used to drive sensors to collect contact force, load, fin condition, and environmental data, calculate the actual friction coefficient and dynamically update it to the corresponding grid. It smooths the actual friction coefficient of the measured grid by discretizing the diffusion equation, and extrapolates the extrapolated friction coefficient of the unmeasured grid based on the neighboring gradient to construct a complete real-time friction field. This adapts to the dynamic changes in outdoor temperature, humidity, and fouling, solves the limitations of static initial values, ensures the accuracy and comprehensiveness of friction sensing, and provides a precise basis for control decisions.

[0124] The control module is used to extract the local friction mean based on the real-time friction field, determine the ideal contact force vector by combining the reference parameters, map the contact force into the joint demand torque and joint angle adjustment through the kinematic model, dynamically plan the motion speed, and generate control commands by integrating feedforward compensation and PID feedback to achieve dynamic adaptation of contact force and speed, avoid fin scratches in high friction areas and energy waste in low friction areas, and ensure cleaning accuracy and equipment safety.

[0125] The evaluation module is used to parse and execute control commands, build a wear estimation model based on historical and real-time data of actual friction coefficients, calculate the cleanliness change rate through visual sensors, determine the cavitation effect, calculate energy consumption per unit area in real time, and predict the remaining cleanable area based on the remaining power for charging planning. It generates a rating report based on cleaning quality indicators, equipment protection indicators, and energy efficiency indicators to prevent excessive wear of components and task interruption, provide a basis for operation efficiency evaluation and subsequent optimization, and ensure that the cleaning process is efficient and stable.

[0126] Working principle and effects:

[0127] Vector information is extracted from the air-cooled island fin drawings and combined with nozzle parameters to construct a cleaning network. The initial pose of the robot is determined by calibrating the reference fin column and setting reference parameters to avoid accidental contact with non-cleaning areas due to spatial perception ambiguity, and to prevent improper initial contact force from scratching the fins or affecting cleaning. Sensors are activated simultaneously to achieve synchronous data acquisition. When the robot moves, multiple sensors collect data and fuse them to calculate the actual friction coefficient. By smoothing the measured grid and extrapolating the unmeasured grid, a real-time friction field is constructed to adapt to changes in the outdoor environment and solve the problem that static initial values ​​cannot cover dynamic friction, ensuring accurate and comprehensive friction perception. The ideal contact force is determined based on the local mean of the friction field. Control commands are generated by combining kinematic models and feedforward-feedback control to make the contact force and speed adapt to the friction state, avoiding scratches in high-friction areas and energy waste in low-friction areas, and ensuring control accuracy. When executing commands, wear, cleanliness, and energy consumption are monitored. If the threshold is exceeded, maintenance or work is suspended to prevent excessive wear of components and task interruption. Finally, a rating report is generated to achieve efficient and safe air-cooled island cleaning, taking into account fin protection, energy consumption reduction, and improved cleaning quality.

[0128] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for path planning of an air-cooled island cleaning robot for low-temperature cooling water cooling, characterized by, The method comprises the following steps: Based on the air cooling island fin layout drawing, initialize the workspace grid, set the reference parameters; Drive the cleaning robot to collect multi-dimensional data, calculate the actual friction coefficient of the measured grid by the dominant friction coefficient and the torque auxiliary friction coefficient, for the unmeasured grid, linear extrapolate by gradient to construct real-time friction field; Get the ideal contact force vector, build the kinematics model to get the joint torque demand, calculate the joint angle adjustment amount, generate control instructions through feedforward compensation and PID feedback control, build wear estimation model, get the current estimated wear, calculate the core indicators, and generate rating report; The step of setting reference parameters comprises: Get the vector information of the air cooling island fin and the core parameters of the spray head, build the fin vector information table through the coordinate system algorithm; Based on the fin row spacing and the spray head coverage width, use spatial uniform discretization to divide the workspace grid and assign semantic labels, divide the to-be-cleaned area, welding point area and non-cleaning boundary, and generate the cleaning grid; Through the wear test platform, get the friction coefficient under the rated driving pressure and clean fin scene, calculate the initial friction coefficient; Combine the set spray head wear threshold and initial spraying parameters to generate the reference control set; The step of calculating the actual friction coefficient of the measured grid comprises: Collect multi-dimensional data and preprocess, timestamp alignment to generate dynamic data set, calculate the dominant friction coefficient and torque auxiliary friction coefficient; Use the dirt recognition model to identify the fin dirt type, get the friction correction coefficient by combining the coefficient correction table, and correct the torque auxiliary friction coefficient; Weight the dominant friction coefficient and the corrected torque auxiliary friction coefficient to calculate the actual friction coefficient and the friction deviation to determine whether the actual friction coefficient is valid; If the friction deviation exceeds the friction comparison threshold, data re-sampling is triggered, and continuous data acquisition continues. Next, if it exists If the friction deviation is greater than the friction comparison threshold, the actual friction coefficient is determined to be valid; otherwise, the sensor is determined to be abnormal. If the friction deviation is less than or equal to the friction comparison threshold, the actual friction coefficient is valid.

2. The low-temperature cooling water cooling air-cooled island cleaning robot path planning method according to claim 1, characterized in that, The step of assigning semantic labels comprises: For each workspace grid, compute the boundary distance , safe distance , and set a safe distance threshold , assign a semantic label to each workspace grid; If , mark the area to be cleaned, semantic label is ; If , mark as weld point region, semantic label as ; Otherwise mark as non-cleaning boundary, semantic label is .

3. The low-temperature cooling water cooling air-cooled island cleaning robot path planning method according to claim 2, characterized by, The step of calculating the actual friction coefficient of the measured grid further comprises: Assign the actual friction coefficient to the corresponding grid, select the high sensitivity area, calculate the high sensitivity friction coefficient through the damage weight factor; wherein the high sensitivity area is the area with a height difference greater than the damage threshold; Get the dirt coverage, select the grid with a dirt coverage greater than the standard coverage, and correct the actual friction coefficient using the friction correction coefficient; Set a two-level adaptive threshold, determine the environment category of the current air cooling island through the environmental humidity, and configure the diffusion coefficient; Smooth the actual friction coefficient, build the measured grid friction field, and calculate the friction fluctuation value; If the friction fluctuation value is greater than the fluctuation upper limit, reduce the diffusion coefficient through the diffusion adjustment coefficient, and update the measured grid friction field; If the friction fluctuation value is not greater than the fluctuation lower limit, increase the diffusion coefficient through the diffusion adjustment coefficient, and update the measured grid friction field.

4. The low-temperature cooling water cooling air-cooled island cleaning robot path planning method according to claim 3, characterized in that, The step of constructing real-time friction field comprises: Traverse the cleaning grid, divide the measured grid and the unmeasured grid, select the nearest measured grid of the unmeasured grid through the Euclidean distance, and get the nearest Euclidean distance; Calculate the friction coefficient gradient of the nearest measured grid, combine the coordinate difference to calculate the extrapolation coefficient; Combine the actual friction coefficient of the nearest measured grid to calculate the extrapolation friction coefficient of the unmeasured grid; Once the nearest Euclidean distance exceeds the reliable extrapolation range, combine the initial friction coefficient to correct the extrapolation friction coefficient to generate a real-time friction field; Calculate the mean and standard deviation of the friction coefficient of all grids in the real-time friction field, and compare it with the reasonable range of the friction coefficient; If the mean of the friction coefficient exceeds the preset reasonable range, trigger the multi-source fusion weight adjustment; if the standard deviation of the friction coefficient is greater than the upper limit of the friction fluctuation, supplement the missing high-sensitivity area; otherwise, the real-time friction field is globally consistent.

5. The low-temperature cooling water cooling air-cooled island cleaning robot path planning method according to claim 4, characterized by, The steps of constructing the real-time friction field further include: Once the cleaning robot moves to the extrapolated unmeasured grid, the actual friction coefficient is obtained, and the local measurement error is calculated for local verification; If the local measurement error is not greater than the extrapolation standard accuracy, it is determined that the real-time friction field is valid; If the local measurement error is greater than the extrapolation standard accuracy, the friction coefficient gradient is corrected by configuring the gradient weight, and the extrapolation is performed again on other unmeasured grids.

6. The low-temperature cooling water cooling air-cooled island cleaning robot path planning method according to claim 5, characterized by, The steps of generating control instructions include: Obtain the friction coefficient of the current contact grid and the surrounding adjacent grids, and calculate the local friction mean; Configure the constraint range of contact force, and calculate the ideal contact force vector; Construct the forward kinematics model, calculate the Jacobian matrix according to the current robot joint angle, generate the required joint torque, and calculate the joint angle adjustment amount; Get the robot's current cleaning path and the area in front. The friction coefficient sequence of each grid is used to calculate the forward sliding mean and standard deviation, calculate the velocity adjustment factor, obtain the initial velocity benchmark, and calculate the target velocity. Obtain the current predicted friction resistance, generate the feedforward compensation torque; Set the target and threshold of feedback control, set the feedback controller parameters based on the benchmark control set using the PID control algorithm, and generate the control instructions.

7. The low-temperature cooling water cooling air-cooled island cleaning robot path planning method according to claim 6, characterized by, The steps of generating the rating report include: Build a wear estimation model to obtain the estimated wear at the current time, and if the estimated wear is greater than the nozzle wear threshold, trigger the shutdown maintenance; Calculate the cleanliness change rate, and if the cleanliness change rate is less than the minimum cleaning effect threshold, determine that the cavitation effect is insufficient; Real-time calculation of unit area cleaning energy consumption, prediction of remaining cleanable area for charging planning; Calculate cleaning quality indicators, equipment protection indicators, and energy efficiency indicators, and set rating standards for each indicator to generate a rating report.

8. A low-temperature cooling water-cooled air-cooled island cleaning robot path planning system for implementing the low-temperature cooling water-cooled air-cooled island cleaning robot path planning method according to any one of claims 1 to 7, characterized by, It includes: Initialization module, friction evaluation module, control module and evaluation module; The initialization module is used to construct a cleaning network with semantic labels, set a benchmark control set, calibrate a benchmark fin column to determine the initial pose of the robot, activate all sensors and set the synchronization acquisition frequency; The friction evaluation module is used to calculate the actual friction coefficient and perform smoothing processing, calculate the extrapolation friction coefficient of the unmeasured grid based on the adjacent gradient, and construct a real-time friction field; The control module is used to extract the local friction mean, obtain the required joint torque and joint angle adjustment amount, and generate control instructions by fusing feedforward compensation and PID feedback; The evaluation module is configured to parse and execute the control instructions, build a wear-out estimation model, calculate a cleanliness change rate to determine a cavitation effect, predict a remaining cleanable area to perform a charging plan, and generate a rating report based on a cleaning quality index, an equipment protection index, and an energy efficiency index.

Images