A cleaning control method and system for a cable duct cleaning robot

Abstract

The present invention discloses a cleaning control method and system for a cable duct cleaning robot. The method comprises the following steps: obtaining the inner wall image of the duct to be cleaned in real time, extracting features from the inner wall image, and obtaining the inner wall condition type of the duct; obtaining the pipe wall distance data of the duct to be cleaned in real time, obtaining the pipe center line based on the pipe wall distance data by using a curve fitting method, and obtaining the pipe bending section length and curvature parameters according to the pipe center line; determining the angle adjustment amount of each joint of the robot according to the inner wall condition type, pipe curvature and curvature parameters of the pipe; obtaining contact pressure data, fusing the contact pressure data, gyroscope data and accelerometer data to obtain fused data, and judging whether the robot fits the inner wall of the duct according to the fused data. If not, adjusting the angle adjustment amount of each joint of the robot. The above method effectively improves the cleaning accuracy of the cable duct cleaning robot.

Description

Technical Field

[0001] The present invention relates to the technical field of cleaning robots, and in particular to a cleaning control method and system for a cable duct cleaning robot. Background Art

[0002] In the cable duct cleaning, the core technical challenges faced by the cable duct cleaning robot when performing tasks stem from the complexity and variability of the pipeline environment. The diversity of the inner wall conditions of the pipeline is the primary problem, including changes in roughness, uncertainty in the type of attachments and their distribution. These factors directly affect the cleaning effect and increase the difficulty of designing the cleaning head. Secondly, the curvature and curvature changes of the pipeline put higher requirements on the robot's travel ability. In the curved section, the robot needs to maintain a stable cleaning pressure while avoiding damage to the pipe wall. This requires the robot to have precise posture control and adaptive capabilities. However, achieving this adaptive capability faces the dilemma of sensor integration. In the narrow pipeline space, it is difficult to obtain comprehensive environmental information, which can easily affect the flexibility and cleaning effect of the robot. At the same time, the collection device also needs to consider the characteristics of different types of dirt, such as viscosity, hardness, etc., to prevent blockage or affect the cleaning efficiency. Therefore, it is difficult to achieve multifunctional integration under limited space and energy constraints. Summary of the invention

[0003] In order to solve the above technical problems, an embodiment of the present invention provides a cleaning control method and system for a cable duct cleaning robot, which effectively improves the cleaning accuracy of the cable duct cleaning robot.

[0004] A first aspect of an embodiment of the present invention provides a cleaning control method of a cable duct cleaning robot, the method comprising:

[0005] The inner wall image of the pipeline to be cleaned is acquired in real time, and features are extracted from the inner wall image to obtain the roughness features and the concave-convexity features of the pipeline wall. Based on the roughness features and the concave-convexity features of the pipeline wall, the type of the inner wall condition of the pipeline is obtained.

[0006] Obtain the pipe wall distance data of the pipe to be cleaned in real time, obtain the pipe center line based on the pipe wall distance data using the curve fitting method, and obtain the pipe curvature and curvature parameters according to the pipe center line;

[0007] According to the type of inner wall condition of the pipeline, the curvature of the pipeline and the curvature parameter, the angle adjustment amount of each joint of the robot is determined; if the curvature of the pipeline is greater than a first preset value or the thickness of the attachment does not meet the preset condition, the angle adjustment amount of one or more joints is increased to the first preset angle; if the curvature of the pipeline is less than the preset value or the thickness of the attachment meets the preset condition, the angle adjustment amount of one or more joints is reduced to the second preset angle, so that the robot cleans the cleaning pipeline according to the first preset angle or the second preset angle;

[0008] The contact pressure data between the robot and the inner wall of the pipe to be cleaned is obtained, and the contact pressure data, gyroscope data and accelerometer data are fused to obtain fused data. Based on the fused data, it is determined whether the robot fits the inner wall of the pipe. If not, the angle adjustment amount of each joint of the robot is adjusted.

[0009] In a possible implementation of the first aspect, acquiring an inner wall image of the pipeline to be cleaned in real time includes:

[0010] Using a high-resolution camera to move and shoot along the axial direction of the pipeline to be cleaned to obtain an image sequence, wherein the image sequence includes a plurality of inner wall images;

[0011] The image sequence is stitched using an image stitching algorithm based on feature point matching to obtain the unfolded image of the pipeline inner wall.

[0012] In a possible implementation of the first aspect, obtaining the pipeline inner wall condition type according to the pipeline wall roughness characteristics and the pipeline wall concave-convex degree characteristics includes:

[0013] Performing median filtering on the pipeline inner wall development image to obtain a first image with noise removed;

[0014] A histogram equalization method is used to perform contrast enhancement on the first image to obtain an enhanced panoramic image of the inner wall of the pipeline;

[0015] Perform image segmentation and edge detection on the enhanced panoramic image of the inner wall of the pipeline to extract the roughness characteristics and the concave-convexity characteristics of the pipeline wall;

[0016] The support vector machine classifier based on radial basis function kernel is used to classify the roughness characteristics and the concave-convex degree characteristics of the pipe wall to obtain the types of attachments.

[0017] An evaluation index system is established based on the roughness characteristics of the pipe wall, the concave-convexity characteristics of the pipe wall, the thickness distribution data of the attachments, the types of attachments and the area of ​​the attachments.

[0018] The weight coefficient of each indicator in the evaluation index system is determined, and a fuzzy relationship matrix is ​​constructed. Fuzzy synthesis operation is performed according to the fuzzy relationship matrix and the type of attachments to obtain the type of pipeline inner wall condition.

[0019] In a possible implementation of the first aspect, based on the pipeline wall distance data, a pipeline centerline is obtained by using a curve fitting method, and the pipeline curvature and curvature parameters are obtained according to the pipeline centerline, including:

[0020] The quartile method is used to eliminate outliers in the pipe wall distance data to obtain effective wall distance data;

[0021] The effective wall distance data is fitted by cubic spline interpolation method to obtain the pipeline cross-sectional profile, the geometric center point coordinates of the pipeline cross-sectional profile are calculated, and the pipeline center line is obtained by fitting the geometric center point coordinates;

[0022] Based on the centerline of the pipeline, the least square objective function is established and the equation group is solved to calculate the pipeline bending radius. The length of the bending section is calculated by integration. According to the length of the bending section, the curvature parameter is calculated using the Frenet formula.

[0023] If the length of the curved section is greater than the first threshold, or the curvature parameter is greater than the second threshold, the pipeline section to be cleaned is determined to be an abnormal pipeline section, and the pipeline curvature and curvature parameters are output.

[0024] In a possible implementation of the first aspect, determining the angle adjustment amount of each joint of the robot according to the type of inner wall condition of the pipeline, the curvature of the pipeline, and the curvature parameter includes:

[0025] The fuzzy comprehensive evaluation method is used to quantitatively score the pipeline inner wall condition type and obtain the scoring result;

[0026] A corresponding table of scores and joint angle adjustments is established according to the scoring results, and the scoring results and joint angle adjustments are normalized and feature selected to obtain processed scoring results and joint angle adjustments;

[0027] The trained neural network model is used for prediction to obtain the angle adjustment of each joint of the robot. The neural network model is trained by a support vector regression algorithm using a radial basis function kernel.

[0028] In a possible implementation of the first aspect, the contact pressure data, the gyroscope data, and the accelerometer data are fused to obtain fused data, and based on the fused data, it is determined whether the robot fits the inner wall of the pipeline, and if not, the angle adjustment amount of each joint of the robot is adjusted, including:

[0029] The Kalman filter algorithm is used to fuse the gyroscope data, accelerometer data and contact pressure data to obtain the fused data and the robot's posture information;

[0030] The fused data and posture information are input into the constructed robot and pipe wall fit model for calculation to obtain the fit score;

[0031] If the fit score is lower than the fit score threshold, it is determined that the robot does not fit the inner wall of the pipeline, and the posture information is processed using the adaptability model of the robot posture and the pipe curvature and pipe diameter changes established by multivariate linear regression analysis to obtain the posture adaptability score;

[0032] If the posture adaptability score is lower than the posture adaptability score threshold, the robot posture is judged to be incompatible with the pipeline characteristics, and a posture adjustment instruction is obtained to adjust the angle adjustment amount of each joint of the robot, wherein the posture adjustment instruction includes the adjustment value and execution order of each joint angle.

[0033] In a possible implementation manner of the first aspect, after adjusting the angle adjustment amount of each joint of the robot, the method further includes:

[0034] Record the adjusted robot posture parameter data, pipeline inner wall condition type and cleaning working parameters;

[0035] The robot posture parameter data, pipeline inner wall condition and cleaning working parameters are input into the cleaning effect evaluation model for calculation to obtain the cleaning effect score. If the cleaning effect score is lower than the effect threshold, the cleaning parameter optimization algorithm is triggered so that the robot uses the particle swarm optimization algorithm to adjust the speed and pressure of the cleaning tool.

[0036] A second aspect of an embodiment of the present invention provides a cleaning control system for a cable duct cleaning robot, the system comprising an extraction module, a fitting module, a first adjustment module and a second adjustment module.

[0037] The extraction module is used to obtain the inner wall image of the pipeline to be cleaned in real time, perform feature extraction on the inner wall image, obtain the pipe wall roughness feature and the pipe wall concave-convex degree feature, and obtain the pipe inner wall condition type according to the pipe wall roughness feature and the pipe wall concave-convex degree feature;

[0038] A fitting module is used to obtain the pipe wall distance data of the pipe to be cleaned in real time, obtain the pipe center line based on the pipe wall distance data by using a curve fitting method, and obtain the pipe curvature and curvature parameters according to the pipe center line;

[0039] The first adjustment module is used to determine the angle adjustment amount of each joint of the robot according to the type of inner wall condition of the pipeline, the curvature of the pipeline and the curvature parameter. If the curvature of the pipeline is greater than a first preset value or the thickness of the attachment does not meet the preset condition, the angle adjustment amount of one or more joints is increased to the first preset angle. If the curvature of the pipeline is less than the preset value or the thickness of the attachment meets the preset condition, the angle adjustment amount of one or more joints is reduced to a second preset angle, so that the robot cleans the cleaning pipeline according to the first preset angle or the second preset angle.

[0040] The second adjustment module is used to obtain the contact pressure data between the robot and the inner wall of the pipe to be cleaned, fuse the contact pressure data, gyroscope data and accelerometer data to obtain fused data, and judge whether the robot fits the inner wall of the pipe according to the fused data. If not, the angle adjustment amount of each joint of the robot is adjusted.

[0041] A third aspect of an embodiment of the present invention provides a computer device, including:

[0042] Memory for storing computer programs;

[0043] The processor is used to implement the cleaning control method of the cable duct cleaning robot as described in the first aspect when executing the computer program.

[0044] A fourth aspect of an embodiment of the present invention provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the cleaning control method of the cable duct cleaning robot of the first aspect are implemented.

[0045] The cleaning control method of the cable pipe cleaning robot provided by the embodiment of the present invention obtains the inner wall image of the pipe to be cleaned in real time, extracts features of the inner wall image, obtains the pipe wall roughness features and the pipe wall concave-convex degree features, and obtains the pipe inner wall condition type according to the pipe wall roughness features and the pipe wall concave-convex degree features; obtains the pipe wall distance data of the pipe to be cleaned in real time, obtains the pipe center line by using a curve fitting method based on the pipe wall distance data, and obtains the pipe bending section length and curvature parameters according to the pipe center line; determines the angle adjustment amount of each joint of the robot according to the pipe inner wall condition type, the pipe curvature and the curvature parameters, and if the pipe curvature is greater than a first preset value, the robot adjusts the angle of each joint of the robot; If the value or the thickness of the attachment does not meet the preset conditions, the angle adjustment amount of one or more joints is increased to the first preset angle. If the pipe curvature is less than the preset value or the thickness of the attachment meets the preset conditions, the angle adjustment amount of one or more joints is reduced to the second preset angle, so that the robot cleans the cleaning pipe according to the first preset angle or the second preset angle; obtain the contact pressure data between the robot and the inner wall of the pipe to be cleaned, fuse the contact pressure data, gyroscope data and accelerometer data to obtain the fused data, and judge whether the robot fits the inner wall of the pipe according to the fused data. If not, adjust the angle adjustment amount of each joint of the robot. Through the above method, the cleaning accuracy of the cable pipe cleaning robot is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 : A flow chart of an analysis control method for an embodiment of a cleaning control method for a cable duct cleaning robot provided by the present invention;

[0047] Figure 2 : A flow chart of abnormal pipeline marking processing according to an embodiment of the cleaning control method of the cable pipeline cleaning robot provided by the present invention;

[0048] Figure 3 : A system block diagram of an embodiment of the cable duct cleaning robot cleaning control method provided by the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Please refer to Figure 1 , which is a flow chart of an embodiment of a cleaning control method of a cable duct cleaning robot provided by an embodiment of the present invention, including steps S101 to S104, each of which is specifically as follows:

[0051] S101. Acquire an inner wall image of the pipeline to be cleaned in real time, extract features of the inner wall image, obtain pipe wall roughness features and pipe wall concavity features, and obtain the pipe inner wall condition type according to the pipe wall roughness features and pipe wall concavity features.

[0052] In this embodiment, a high-resolution camera is moved along the axial direction of the pipeline to capture an inner wall image, and an image stitching algorithm based on feature point matching is used according to the image sequence to obtain a complete pipeline inner wall expansion image. After the pipeline inner wall expansion image is subjected to median filtering to remove noise, feature extraction is performed on the inner wall image to obtain pipe wall roughness features and pipe wall concavity features, and the pipe wall inner wall condition type is obtained based on the pipe wall roughness features and pipe wall concavity features.

[0053] In some embodiments, the step S101 of "real-time acquisition of the inner wall image of the pipeline to be cleaned" includes but is not limited to the following steps:

[0054] Using a high-resolution camera to move and shoot along the axial direction of the pipeline to be cleaned to obtain an image sequence, wherein the image sequence includes a plurality of inner wall images;

[0055] The image sequence is stitched using an image stitching algorithm based on feature point matching to obtain the unfolded image of the pipeline inner wall.

[0056] In this embodiment, a high-resolution camera is used to move and shoot along the axial direction of the pipeline to be cleaned to obtain an image sequence, wherein the image sequence includes multiple inner wall images, and the continuous images are synthesized into a complete expanded image of the inner wall of the pipeline through an image stitching algorithm based on feature point matching.

[0057] In some embodiments, the step S101 of "obtaining the type of the inner wall condition of the pipeline according to the pipe wall roughness characteristics and the pipe wall concave-convex degree characteristics" includes but is not limited to the following steps:

[0058] Performing median filtering on the pipeline inner wall development image to obtain a first image with noise removed;

[0059] A histogram equalization method is used to perform contrast enhancement on the first image to obtain an enhanced panoramic image of the inner wall of the pipeline;

[0060] Perform image segmentation and edge detection on the enhanced panoramic image of the inner wall of the pipeline to extract the roughness characteristics and the concave-convexity characteristics of the pipeline wall;

[0061] The support vector machine classifier based on radial basis function kernel is used to classify the roughness characteristics and the concave-convex degree characteristics of the pipe wall to obtain the types of attachments.

[0062] An evaluation index system is established based on the roughness characteristics of the pipe wall, the concave-convexity characteristics of the pipe wall, the thickness distribution data of the attachments, the types of attachments and the area of ​​the attachments.

[0063] The weight coefficient of each indicator in the evaluation index system is determined, and a fuzzy relationship matrix is ​​constructed. Fuzzy synthesis operation is performed according to the fuzzy relationship matrix and the type of attachments to obtain the type of pipeline inner wall condition.

[0064] In this embodiment, the synthesized image is subjected to median filtering to remove noise, and then histogram equalization is used to enhance the image contrast to obtain a clear panoramic view of the inner wall of the pipeline. Structured light scanning is used to obtain the depth information of the inner wall of the pipeline, and three-dimensional point cloud data is constructed. Image segmentation and edge detection are performed on the processed panoramic view of the inner wall of the pipeline to extract the roughness and concave-convexity characteristics of the pipe wall. The energy, contrast, entropy and other statistical features of the gray-level co-occurrence matrix are calculated, and the uniform pattern histogram of the local binary pattern is calculated. Combined with the depth information, a feature vector characterizing the condition of the pipe wall is constructed.

[0065] According to the pre-established database of pipeline inner wall conditions, the support vector machine classifier based on the radial basis function kernel is used to classify the extracted feature vectors and determine the type of pipeline inner wall attachments. The coverage area of ​​the attachments is calculated by the threshold-based regional growing algorithm. The seed point selection is based on the local grayscale extreme value, and the growth condition is that the grayscale difference of adjacent pixels is less than the preset threshold. The thickness distribution of the attachments is obtained using three-dimensional point cloud data. An evaluation index system is established based on the characteristics of pipe wall roughness, concave-convexity, attachment type, coverage area and thickness distribution. The weight coefficient of each indicator is determined and a fuzzy relationship matrix is ​​constructed. Fuzzy synthesis operation is performed to obtain the type of pipeline inner wall condition. According to the classification results, a pipeline inner wall condition assessment report is established, which includes the type of attachments, coverage area percentage and thickness distribution map.

[0066] As an example of this embodiment, a high-resolution industrial camera is used to collect images of the inner wall of the pipeline, with a resolution of 3840x2160 pixels and a frame rate of 60fps. The camera moves along the axis of the pipeline at a speed of 0.1m / s, and takes a frame of image every 0.5m. The SIFT algorithm is used to extract image feature points, and the RANSAC algorithm is used for feature matching to achieve image stitching. A median filter with a kernel size of 5x5 is applied to the spliced ​​image to remove noise, and then the contrast is enhanced by histogram equalization. Structured light scanning uses a line laser projector and a high-speed camera to obtain depth information through the principle of triangulation, and the scanning accuracy reaches 0.1mm. The Otsu threshold method is used for image segmentation, and the Canny algorithm is used for edge detection. When extracting the roughness features of the pipe wall, the root mean square value of the surface height change is calculated. The concave-convex degree feature is represented by calculating the peak-to-valley ratio of the surface profile.

[0067] When calculating the gray-level co-occurrence matrix, the distance parameter d=1, the direction θ=0°, 45°, 90°, 135°, and the energy, contrast, entropy and other features are statistically analyzed. The local binary pattern uses a circular neighborhood with a radius of 3 to calculate the histogram of 58 uniform patterns. The support vector machine classifier uses the RBF kernel function, the kernel parameter γ=0.1, and the penalty factor C=10. The parameters are optimized by grid search and cross-validation. In the regional growing algorithm, the seed point selects the extreme point whose gray value is located in the local 10x10 window, and the growth condition is that the gray difference between adjacent pixels is less than 20. The coverage area is obtained by counting pixels and then divided by the total area to obtain the percentage. The thickness distribution uses point cloud data to calculate the distance from the surface of the attachment to the reference plane of the pipe wall. The evaluation index system includes five indicators: roughness, concave-convexity, attachment type, coverage area and thickness. The hierarchical analysis method is used to determine the weight coefficient and construct a 5x5 fuzzy relationship matrix. The maximum membership principle is used for fuzzy synthesis to obtain the type of pipeline inner wall condition. The assessment report includes the types of deposits such as scale and rust, the percentage of coverage area accurate to 0.1%, and the thickness distribution diagram represented by a heat map. The colors range from blue to red to represent the change in thickness from 0mm to 10mm.

[0068] S102, obtaining the pipe wall distance data of the pipe to be cleaned in real time, obtaining the pipe center line by using a curve fitting method based on the pipe wall distance data, and obtaining the pipe curvature and curvature parameters according to the pipe center line.

[0069] In this embodiment, multiple groups of ultrasonic distance measuring sensors are arranged evenly distributed along the circumference of the pipeline, and the ultrasonic distance measuring sensors collect pipeline wall distance data and transmit the pipeline wall distance data to the data processing unit through a data acquisition card. Then, the pipeline centerline is obtained by using a curve fitting method, and the pipeline bending radius, bending section length and bending section curvature parameters are calculated to obtain the pipeline curvature and curvature change.

[0070] In some embodiments, in step S102, "based on the pipeline wall distance data, using a curve fitting method to obtain the pipeline centerline, and obtaining the pipeline curvature and curvature parameters according to the pipeline centerline" includes but is not limited to the following steps:

[0071] The quartile method is used to eliminate outliers in the pipe wall distance data to obtain effective wall distance data;

[0072] The effective wall distance data is fitted by cubic spline interpolation method to obtain the pipeline cross-sectional profile, the geometric center point coordinates of the pipeline cross-sectional profile are calculated, and the pipeline center line is obtained by fitting the geometric center point coordinates;

[0073] Based on the centerline of the pipeline, the least square objective function is established and the equation group is solved to calculate the pipeline bending radius. The length of the bending section is calculated by integration. According to the length of the bending section, the curvature parameter is calculated using the Frenet formula.

[0074] If the length of the curved section is greater than the first threshold, or the curvature parameter is greater than the second threshold, the pipeline section to be cleaned is determined to be an abnormal pipeline section, and the length of the curved section and the curvature parameter of the pipeline are output.

[0075] In this embodiment, if Figure 2 As shown in the figure, the median filter is used to remove the impulse noise and the sliding average method is used to smooth the data to obtain the preprocessed wall distance data. The quartile method is used to remove the outliers in the preprocessed wall distance data to obtain the effective wall distance data.

[0076] Then, the effective wall distance data is fitted by cubic spline interpolation method to obtain the cross-sectional profile of the pipeline and calculate the coordinates of the geometric center point of the cross-sectional profile of the pipeline. The centerline of the pipeline is obtained by fitting the coordinates of the geometric center point, and the least squares objective function is established and the equation group is solved to calculate the bending radius of the pipeline. The length of the bending section is calculated by integration, and the curvature parameter is calculated using the Frenet formula. The pipeline bending radius, bending section length and curvature parameters are compared with the preset threshold. If the curvature or curvature change exceeds the preset threshold, it is marked as an abnormal bending section. Generate a pipeline curvature analysis report, which includes the pipeline bending radius, bending section length, curvature mean, standard deviation and coefficient of variation parameters, as well as the location and degree of abnormal bending sections.

[0077] As an example of this embodiment, multiple groups of ultrasonic ranging sensors are arranged evenly distributed along the circumference of the pipeline, each group of sensors includes 4 ranging points, and two adjacent groups of sensors are spaced 0.5 meters apart along the axial direction of the pipeline. The measurement range is 0-5 meters and the accuracy is ±1 mm.

[0078] Collect the distance data of the pipeline wall, and transmit the distance data to the data processing unit through the data acquisition card. Preprocess the collected raw data, including using median filtering to remove impulse noise and using the sliding average method to smooth the data. Use the quartile method to remove outliers from the preprocessed wall distance data. Use the cubic spline interpolation method to fit the pipeline cross-sectional profile, select the nodes as the actual measurement points, and set the boundary conditions to natural boundary conditions. Calculate the coordinates of the geometric center point of each cross section, and fit the coordinates of the center points of all cross sections to obtain the center line of the pipeline.

[0079] According to the pipeline centerline obtained by fitting, the least squares objective function is established, and the equation group is solved to calculate the pipeline bending radius. The length of the bending section is calculated by integration, and the curvature parameters are calculated using the Frenet formula. A threshold is set based on the curvature change rate to determine the starting and ending points of the bending section to obtain the pipeline curvature. The mean, standard deviation and coefficient of variation of the curvature are calculated to quantify the curvature change. The calculated bending radius, bending section length and curvature parameters are compared with the preset threshold, which is determined based on the statistical results of historical data. If the curvature or curvature change exceeds the preset threshold, it is marked as an abnormal bending section. Generate a pipeline curvature analysis report, including parameters such as bending radius, bending section length, curvature mean, standard deviation and coefficient of variation, as well as the location and degree of abnormal bending sections.

[0080] During the pipeline detection process, an ultrasonic distance sensor model HC-SR04 is used, with a measurement range of 0-5 meters and an accuracy of ±1 mm. A sensor is installed every 90 degrees along the circumference of the pipeline, and a group is set every 0.5 meters in the axial direction, forming a grid layout. The data acquisition frequency is set to 10Hz and transmitted to the data processing unit through the RS485 interface. The raw data is filtered by the median filter of the 5x5 window to remove the impulse noise, and then the data is smoothed using the 7-point sliding average method. The quartile method is used to eliminate outliers, and the lower limit is set to Q1-1.5IQR and the upper limit is Q3+1.5IQR, where Q1 represents the first quartile, that is, the value at the 25% position after the data is sorted from small to large, Q3 represents the third quartile, that is, the value at the 75% position after the data is sorted from small to large, and IQR represents the interquartile range, which is calculated as IQR=Q3-Q1.

[0081] The processed data is interpolated using cubic spline, the nodes are selected as actual measurement points, and the boundary conditions are natural boundary conditions. When calculating the geometric center of the cross section, the area weighted method is used. The least squares method is used for fitting the pipeline centerline, and the fitting polynomial order is 3. The objective function for the bending radius calculation is:

[0082]

[0083] Solve the system of equations: =0, =0, =0.

[0084] The length of the curved section is calculated by numerical integration method with a step size of 0.01 m. The curvature parameter is calculated using the Fresnet formula, where the Fresnet formula is:

[0085]

[0086] Set the curvature change rate threshold to 0.01m -1 / m, determine the starting and ending points of the curved section. Calculate the curvature mean μ, standard deviation σ and coefficient of variation CV=σ / μ. The preset thresholds are based on historical data statistics, such as the bending radius threshold of 50 meters, the bending section length threshold of 10 meters, and the curvature coefficient of variation threshold of 0.2. The generated analysis report contains the bending radius, bending section length, curvature mean, standard deviation and coefficient of variation, as well as the starting and ending mileage and maximum curvature value of the abnormal bending section.

[0087] S103. Determine the angle adjustment amount of each joint of the robot according to the type of inner wall condition of the pipeline, the curvature of the pipeline and the curvature parameters. If the curvature of the pipeline is greater than a first preset value or the thickness of the attachment does not meet the preset conditions, increase the angle adjustment amount of one or more joints to the first preset angle. If the curvature of the pipeline is less than the preset value or the thickness of the attachment meets the preset conditions, reduce the angle adjustment amount of one or more joints to the second preset angle, so that the robot cleans the cleaning pipeline according to the first preset angle or the second preset angle.

[0088] In this embodiment, the pipeline inner wall condition type, tortuosity and curvature change data are obtained, and a fuzzy comprehensive evaluation method is used to quantitatively score the pipeline condition. The scoring includes three evaluation factors: inner wall roughness, tortuosity and curvature change.

[0089] According to the scoring results, a corresponding table of scores and joint angle adjustments is established, which includes angle adjustment ranges for three levels: excellent, general, and poor. The pipeline condition scores and joint angle adjustment data are normalized and feature selected, and the support vector regression algorithm with radial basis function kernel is used to train the model. The penalty factor and error parameter of the model are set according to the preset values. The coefficient of variation of the thickness of the attachment is calculated. If the coefficient of variation is greater than the preset threshold, the joint angle adjustment is increased; if the coefficient of variation is less than the preset threshold, the joint angle adjustment is reduced. The joint angle adjustment is increased according to the curvature of the pipeline. The weight optimization method based on gradient descent is used to fine-tune the joint angle adjustment. The fine-tuning is performed by setting the learning rate and the number of iterations. The optimization objective function includes two indicators: pipeline curvature adaptability and uniformity of attachment distribution. The weight coefficient of each joint is updated through the back propagation algorithm to determine the angle adjustment of each joint of the robot.

[0090] In some embodiments, the step S103 of “determining the angle adjustment amount of each joint of the robot according to the type of inner wall condition of the pipeline, the curvature of the pipeline and the curvature parameter” includes but is not limited to the following steps:

[0091] The fuzzy comprehensive evaluation method is used to quantitatively score the pipeline inner wall condition type and obtain the scoring result;

[0092] A corresponding table of scores and joint angle adjustments is established according to the scoring results, and the scoring results and joint angle adjustments are normalized and feature selected to obtain processed scoring results and joint angle adjustments;

[0093] The trained neural network model is used for prediction to obtain the angle adjustment of each joint of the robot. The neural network model is trained by a support vector regression algorithm using a radial basis function kernel.

[0094] In this embodiment, the pipeline inner wall condition type, curvature and curvature change data are obtained, and the pipeline condition is quantitatively scored using a fuzzy comprehensive evaluation method. The evaluation factors include the inner wall roughness weight of 0.3, the curvature weight of 0.4 and the curvature change weight of 0.3.

[0095] According to the scoring results, the pipeline conditions are divided into three levels: excellent 80-100 points, general 60-79 points, and poor 0-59 points. A corresponding table between the score and the joint angle adjustment amount is established, such as the excellent condition corresponds to the ±5° adjustment range, the general condition corresponds to the ±10° adjustment range, and the poor condition corresponds to the ±15° adjustment range.

[0096] The pipeline condition score and joint angle adjustment data were preprocessed, including normalization and feature selection. The support vector regression algorithm with radial basis function kernel was used to train the model, with the penalty factor C set to 1.0 and the error parameter set to 0.1. The pipeline condition score was input, the angle adjustment of each joint was output, and a preliminary joint angle adjustment plan was obtained.

[0097] The robot joint structure includes 6 degrees of freedom, including three translation joints and three rotation joints. The coefficient of variation of the thickness of the attachment is calculated, and the preset threshold is set to 0.5 based on historical data statistics. If the coefficient of variation is greater than 0.5, the joint angle adjustment is increased by 20% of the original adjustment on the basis of the preliminary plan. If the coefficient of variation is less than 0.5, the joint angle adjustment is reduced by 10% of the original adjustment. At the same time, the curvature of the pipeline is considered. If the curvature is greater than the preset threshold of 30° / m, the adjustment of the corresponding joint is increased by 15% of the original adjustment. The weight optimization method based on gradient descent is used to fine-tune the joint angle adjustment. The learning rate is set to 0.01 and the number of iterations is 100. The optimization objective function includes two indicators: adaptability of pipeline curvature and uniformity of attachment distribution. The weight coefficients of each joint are updated through the back propagation algorithm to achieve adaptive adjustment. Finally, the angle adjustment of each joint of the robot is determined, and it is verified and corrected according to the constraints of the robot kinematic model to ensure that the adjustment is within the working range of the robot.

[0098] Specifically, during the pipeline inspection process, a fuzzy comprehensive evaluation method is used to quantitatively score the pipeline condition. The inner wall roughness is measured by a laser scanner, with a range of 0-100μm and a weight of 0.3; the curvature is measured by a gyroscope, with a range of 0-60° / m and a weight of 0.4; the curvature change is calculated using the differential method, with a range of 0-0.5m -1 , weight 0.3. The scoring result is 80 points or above for excellent, corresponding to the adjustment range of ±5°; 60-79 points for fair, corresponding to the adjustment range of ±10°; 59 points or below for poor, corresponding to the adjustment range of ±15°. Min-Max normalization is used for data preprocessing, and principal component analysis is used for feature selection, retaining the principal components with a contribution rate of 95%. The support vector regression model uses the RBF kernel function, with kernel parameter γ=0.1, penalty factor C=1.0, and error=0.1. The robot joints include three translation joints of X, Y, and Z, each with a stroke of 500mm and three rotation joints of roll angle, pitch angle, and yaw angle, each with a range of ±180°. The coefficient of variation of the thickness of the attachment is calculated by dividing the standard deviation by the mean, and the threshold of 0.5 is determined based on the 90% quantile of the historical 1000 detection data. The curvature threshold of 30° / m is set with reference to the ASMEB31.3 pipeline standard. In the gradient descent optimization, the initial value of the learning rate is 0.01, and the adaptive adjustment strategy is adopted, with 100 iterations. The objective function weights are 0.6 for curvature adaptability and 0.4 for uniformity of attachment distribution. The back propagation algorithm uses the chain rule to calculate the gradient. The final joint angle adjustment is kinematically verified by the DH parameter method to ensure that within the workspace, the X, Y, and Z adjustments do not exceed ±100 mm, and the roll, pitch, and yaw angle adjustments do not exceed ±30°.

[0099] S104, obtaining contact pressure data between the robot and the inner wall of the pipe to be cleaned, fusing the contact pressure data, gyroscope data and accelerometer data to obtain fused data, and judging whether the robot fits the inner wall of the pipe based on the fused data. If not, adjusting the angle adjustment amount of each joint of the robot.

[0100] In this embodiment, a micro pressure sensor array is arranged on the robot body to obtain the contact pressure distribution data between the robot and the pipe wall, and the pressure data is integrated with the gyroscope and accelerometer data to determine whether the robot fits tightly against the inner wall of the pipe. The angle adjustment amount of each joint is combined to determine whether the robot posture is adapted to the curvature and diameter changes of the pipe.

[0101] In some embodiments, in step S104, "the contact pressure data, the gyroscope data and the accelerometer data are fused to obtain fused data, and based on the fused data, whether the robot fits the inner wall of the pipe is determined, and if not, the angle adjustment amount of each joint of the robot is adjusted" includes but is not limited to the following steps:

[0102] Using a Kalman filter algorithm to fuse the gyroscope data, the accelerometer data and the contact pressure data to obtain fused data, and acquire the posture information of the robot;

[0103] The fused data and posture information are input into the constructed robot and pipe wall fit model for calculation to obtain a fit score;

[0104] If the fit score is lower than the fit score threshold, it is determined that the robot does not fit the inner wall of the pipeline, and the posture information is processed using the adaptability model of the robot posture and the pipe curvature and pipe diameter changes established by multivariate linear regression analysis to obtain the posture adaptability score;

[0105] If the posture adaptability score is lower than the posture adaptability score threshold, the robot posture is judged to be incompatible with the pipeline characteristics, and a posture adjustment instruction is obtained to adjust the angle adjustment amount of each joint of the robot, wherein the posture adjustment instruction includes the adjustment value and execution order of each joint angle.

[0106] In this embodiment, the contact pressure distribution data between the robot and the pipe wall is obtained, and the original data is subjected to median filtering and normalization processing to obtain the contact pressure distribution characteristic data between the robot and the pipe wall; based on the gyroscope and accelerometer data, the Kalman filter algorithm is used for data fusion to obtain the robot's posture information, and the posture information includes the pitch angle, roll angle and yaw angle.

[0107] The support vector machine algorithm is used to establish the robot-wall fit model. The input features of the model include pressure mean, pressure variance, posture angle and angular velocity, and the output fit score is a fit score. If the fit score is lower than the preset threshold, it is judged that the robot does not fit tightly with the inner wall of the pipe. The adaptability model of the robot posture and pipe curvature and pipe diameter change is established through multivariate linear regression analysis. The independent variables of the model include joint angle, fit score, pipe curvature and pipe diameter change rate, and the dependent variable is the posture adaptability score. If the posture adaptability score is lower than the preset threshold, it is judged that the robot posture is not adapted to the pipe characteristics, and the posture adjustment instruction is triggered. The posture adjustment instruction contains the specific adjustment values ​​and execution order of each joint angle.

[0108] As an example of this embodiment, 100 micro pressure sensors are evenly arranged on the surface of the robot body to form a sensor array. The sensitivity of each sensor is 0.01N, the range is 0-100N, and the sampling frequency is 100Hz. The contact pressure distribution data between the robot and the pipe wall is obtained through the data acquisition module, and the raw data is filtered by a 5x5 window median filter to remove outliers, and then the Min-Max normalization process is used. At the same time, a laser ranging sensor is used to measure the cross-sectional dimensions of the pipeline in real time to obtain the pipe diameter change data. The Kalman filter algorithm is used to fuse the gyroscope and accelerometer data, and the process noise covariance is set to 0.001 and the observation noise covariance is set to 0.1.

[0109] The robot's posture information, including pitch angle, roll angle and yaw angle, is obtained, and the position and speed of the robot in the pipeline are calculated. The posture angle is used as a feature input for the subsequent fit model. Based on the pressure distribution data and posture information, the support vector machine algorithm is used to establish a robot-wall fit model. The RBF kernel function is selected, the kernel parameter γ is set to 0.1, and the penalty factor C is set to 10. The input features include pressure mean, pressure variance, posture angle and angular velocity, and the fit score is output. If the score is lower than the preset threshold of 0.8, it is judged that the robot is not tightly fitted to the inner wall of the pipeline.

[0110] The fit score is used as an input variable of the posture adaptability model. Combined with the angle adjustment of each joint and the fit score, the adaptability model of the robot posture and the pipe curvature and diameter change is established through multivariate linear regression analysis. The independent variables include joint angle, fit score, pipe curvature and pipe diameter change rate, and the dependent variable is the posture adaptability score. The least squares method is used to estimate the regression coefficient. If the score is lower than the preset threshold of 0.7, the robot posture is judged to be incompatible with the pipeline characteristics, and the posture adjustment instruction is triggered. The adjustment instruction contains the specific adjustment value and execution order of each joint angle, and the adjustment operation is performed by the robot controller. On the pipeline inspection robot, a force-sensitive resistor model FSR402 is used as a pressure sensor, and 100 sensors are evenly arranged on the surface of the robot to form a 10x10 matrix array. The sensitivity of each sensor is 0.01N, the range is 0-100N, and the sampling frequency is 100Hz. Data acquisition uses a 16-bit ADC with a resolution of 0.0015N. The raw data is denoised by 5x5 window median filtering, and then the values ​​are mapped to the range of 0-1 by Min-Max normalization. SICKOD1000 is used as the laser ranging sensor, with a measurement range of 50-1000mm and an accuracy of ±2mm, which is used to measure the change of pipe diameter in real time. MPU-6050 is used as the attitude sensor, with a gyroscope range of ±250° / s and an accelerometer range of ±2g. Kalman filter parameter settings: process noise covariance Q=0.001, observation noise covariance R=0.1, and initial estimation covariance P=1. The attitude angle accuracy obtained after fusion reaches ±0.1°. The support vector machine model is implemented using the LIBSVM library, with the RBF kernel function parameter γ=0.1, the penalty factor C=10, and the optimal parameters are selected by cross-validation. The input features include pressure mean, variance, kurtosis, skewness, and three-axis attitude angle and angular velocity, a total of 10-dimensional feature vectors. The fit threshold is set to 0.8, and adjustment is triggered when it is lower than this value. The multivariate linear regression model uses the least squares method to estimate parameters. The independent variables include 6 joint angles, fit score, pipe bending curvature and pipe diameter change rate, and the dependent variable is the posture adaptability score. The coefficient of determination R² of the regression model is required to be greater than 0.85, and the F test p value is less than 0.01. If the adaptability score is lower than 0.7, the posture adjustment is triggered. The adjustment command is sent to each joint motor controller through the CAN bus and executed in the preset order. The adjustment amplitude does not exceed 5° each time. After adjustment, re-evaluation is performed until the requirements are met or the maximum number of attempts is 10.

[0111] In some embodiments, step S104 "after adjusting the angle adjustment amount of each joint of the robot" includes but is not limited to the following steps:

[0112] Record the adjusted robot posture parameter data, pipeline inner wall condition type and cleaning working parameters;

[0113] The robot posture parameter data, pipeline inner wall condition and cleaning working parameters are input into the cleaning effect evaluation model for calculation to obtain the cleaning effect score. If the cleaning effect score is lower than the effect threshold, the cleaning parameter optimization algorithm is triggered so that the robot uses the particle swarm optimization algorithm to adjust the speed and pressure of the cleaning tool.

[0114] In this embodiment, during the cleaning process, images of the inner wall of the pipeline and data on the curvature of the pipeline are continuously acquired. According to changes in the coverage area and thickness distribution of pipeline attachments and the degree of fit between the robot and the pipe wall, the robot joint angles and drive wheel torque are dynamically adjusted to keep the robot posture adapted to the pipeline condition.

[0115] A 4K resolution industrial camera is used to collect images of the inner wall of the pipeline in real time, with the image collected every 0.1 second. The Otsu threshold segmentation method is used to extract the attachment area, and then the coverage area and thickness distribution are calculated through morphological operations. At the same time, the six-axis inertial measurement unit is used to obtain the pipeline curvature data. This data is input into the pre-trained ResNet50 convolutional neural network, which adds two fully connected layers and outputs real-time assessment results of the pipeline condition, including indicators such as attachment type, severity, and cleaning difficulty.

[0116] The fit index between the robot and the pipe wall is calculated based on the data collected by the pressure sensor array. The fit index is calculated based on the variance and average value of the pressure data, reflecting the contact uniformity and pressure size. A fuzzy controller takes the fit index, pipe condition assessment results and current robot posture parameters as inputs, and outputs the joint angle adjustment and drive wheel torque adjustment. The fuzzy controller uses a triangular membership function, and the rule base contains 27 IF-THEN rules, covering different combinations of fit, pipe condition and posture.

[0117] Adaptive PID control algorithm is used to precisely control the motors of each joint of the robot. The Ziegler-Nichols method is used to initially set the PID parameters, and then the parameters are adjusted in real time according to the control error. The joint angle is dynamically adjusted according to the adjustment amount output by the fuzzy controller, and the adjustment frequency is 10Hz. At the same time, the output torque of the drive wheel motor is controlled by PWM modulation, the PWM frequency is 20kHz, and the duty cycle is dynamically adjusted according to the required torque. A relationship model between the posture parameters and the contact area of ​​the cleaning tool is established to predict the impact of posture adjustment on the cleaning effect.

[0118] A cleaning effect evaluation model was established, using the random forest algorithm, with inputs including the adjusted robot posture parameters, pipeline condition evaluation results, and cleaning tool working parameters, and the cleaning effect score was output. If the score is lower than the preset threshold of 0.8 (full score is 1), the cleaning parameter optimization algorithm is triggered. Then, the particle swarm optimization algorithm is used, with the cleaning effect score as the objective function, to adjust the speed range of the cleaning tool to 100-1000rpm and the pressure range to 10-100N.

[0119] The optimization results are sent to the cleaning tool actuator in real time through the controller to form a closed-loop feedback control. The whole process is executed every 0.5 seconds to ensure the real-time and adaptability of the cleaning process. In the actual application of the pipeline cleaning robot, a 4K industrial camera with a Sony IMX415 sensor is used, with a resolution of 3840x2160 and a frame rate of 30fps, and one frame of image is collected every 0.1 seconds. Image processing uses the OpenCV library to implement Otsu threshold segmentation, and the threshold range is dynamically adjusted between 0-255. The morphological operation uses the opening and closing operations of the 3x3 kernel to remove noise and smooth the edges.

[0120] The six-axis inertial measurement unit uses InvenSense MPU-6050, with a sampling rate of 1kHz, a gyroscope range of ±250° / s, and an accelerometer range of ±2g. The ResNet50 network is implemented using the PyTorch framework. The number of fully connected layer nodes added is 1024 and 512 respectively, and the output layer is 3 nodes, corresponding to the attachment type classification of 0-5, the severity of 0-10 points, and the cleaning difficulty of 0-10 points. The pressure sensor array consists of 100 FSR402 force-sensitive resistors with a sensitivity of 0.2N / bit and a sampling frequency of 100Hz. The fit index calculation formula is Index=(1-σ / μ)*(μ / Max_pressure), where σ is the pressure standard deviation, μ is the average pressure, and Max_pressure is the maximum range of the sensor 100N.

[0121] The fuzzy controller is designed using MATLAB FuzzyLogicToolbox. The input variables fit, pipeline status and posture are divided into 3 levels each, and the output variables angle adjustment and torque adjustment are divided into 5 levels each. The initial parameters of PID control are set by Ziegler-Nichols method, Kp=0.6Ku, Ki=2Kp / Tu, Kd=KpTu / 8, where Ku is the critical gain and Tu is the critical period. The adaptive adjustment adopts the MIT rule, and the learning rate is set to 0.01. PWM control is implemented using the timer of the STM32F4 series MCU, with a frequency of 20kHz and a resolution of 1000 levels. The random forest model is implemented using the scikit-learn library, with the number of trees set to 100 and the maximum depth of 10. The particle swarm optimization algorithm parameter settings: number of particles 50, maximum number of iterations 100, inertia weight w=0.7, acceleration constant c1=c2=1.4.

[0122] After the cleaning is completed, a report containing cleaning data and effect evaluation is generated based on the pipeline inner wall condition parameters, pipeline curvature data, and robot posture adjustment data recorded during the cleaning process. Specifically, the Z-score standardization method is used to normalize the pipeline inner wall condition parameters, pipeline curvature data, and robot posture adjustment data recorded during the cleaning process. The interquartile range IQR method is used to eliminate outliers, and data points exceeding 1.5 times the IQR range are marked as abnormal. A standardized cleaning process data set is obtained by unifying the data format and aligning the timestamps. The autoregressive integrated moving average ARIMA model is used to perform time series analysis on the processed data set. The statistical characteristics of the pipeline inner wall condition parameters, curvature data, and robot posture adjustment data, including mean, variance, kurtosis, and skewness, are calculated.

[0123] The PELT algorithm is used for change point detection to identify the key time nodes of data changes. Based on the results of time series analysis, the most representative feature subset is selected using the recursive feature elimination (RFE) method. Based on the selected feature subset, the CART decision tree algorithm is used to build a cleaning effect evaluation model. The maximum depth of the decision tree is set to 10, and the minimum number of sample splits is set to 20. The pipeline inner wall condition parameters, curvature data, and robot posture adjustment data are used as input variables, and the cleaning effect score 0-100 points and the classification results are output as excellent, good, general, and poor. The model performance is evaluated by 10-fold cross validation, and the model with the highest accuracy is selected. Based on the output results of the evaluation model, a template-based natural language generation method is used to automatically generate a report. A report template is designed, which includes four parts: pipeline feature description, cleaning process analysis, robot performance evaluation, and cleaning effect summary. The numerical results output by the model are mapped to predefined text fragments, and complete sentences are generated through rule combination. Python's Matplotlib library is used to generate data visualization charts, including time series charts, scatter plots, and heat maps. Finally, the generated text and charts are integrated into a report in HTML format, and a dynamic interactive report is displayed through a web interface. After the pipeline cleaning robot completes its work, the data processing system first preprocesses the collected data. The Python SciPy library is used to implement Z-score standardization, converting all data into a distribution with a mean of 0 and a standard deviation of 1. The IQR method is used for outlier detection, and the thresholds are set to Q1-1.5IQR and Q3+1.5IQR. Data points that exceed the range are marked as abnormal and removed. The statsmodels library is used to implement the ARIMA model for time series analysis, and the (p, d, q) = (2, 1, 2) model is established for the pipeline inner wall condition, curvature, and robot posture data.

[0124] The PELT algorithm is implemented by the ruptures library, with the penalty parameter set to 0.5 to identify the key time points of data changes. The feature selection uses the RFE method of the sklearn library to select the top 10 most important features. The CART decision tree algorithm is also implemented by sklearn, with max_depth=10, min_samples_split=20, and hyperparameter tuning is performed through GridSearchCV. The model evaluation uses 10-fold cross validation to select the model with the highest accuracy. The report generation uses the Jinja2 template engine, pre-defines a text template with 4 parts, and maps the numerical values ​​output by the model to the corresponding descriptive text through conditional statements. Matplotlib is used for data visualization to generate a time series diagram with figsize=(10,6), dpi=300, a scatter plot alpha=0.6, cmap='viridis' and a heat map cmap='YlOrRd'. The final report is generated in HTML format, and the Flask framework is used to build a web interface to achieve dynamic interactive display of data.

[0125] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0126] In some embodiments, Figure 3 As shown, it shows a block diagram of a cable duct cleaning robot cleaning control system 300 provided in an embodiment of the present application, including: an extraction module 301, a fitting module 302, a first adjustment module 303 and a second adjustment module 304, wherein:

[0127] The extraction module 301 is used to obtain the inner wall image of the pipeline to be cleaned in real time, perform feature extraction on the inner wall image, obtain the pipe wall roughness feature and the pipe wall concave-convex degree feature, and obtain the pipe inner wall condition type according to the pipe wall roughness feature and the pipe wall concave-convex degree feature;

[0128] The fitting module 302 is used to obtain the pipe wall distance data of the pipe to be cleaned in real time, obtain the pipe center line based on the pipe wall distance data by using a curve fitting method, and obtain the pipe bending section length and curvature parameters according to the pipe center line;

[0129] The first adjustment module 303 is used to determine the angle adjustment amount of each joint of the robot according to the type of inner wall condition of the pipeline, the curvature of the pipeline and the curvature parameter. If the curvature of the pipeline is greater than the first preset value or the thickness of the attachment does not meet the preset condition, the angle adjustment amount of one or more joints is increased to the first preset angle. If the curvature of the pipeline is less than the preset value or the thickness of the attachment meets the preset condition, the angle adjustment amount of one or more joints is reduced to the second preset angle, so that the robot cleans the cleaning pipeline according to the first preset angle or the second preset angle.

[0130] The second adjustment module 304 is used to obtain the contact pressure data between the robot and the inner wall of the pipe to be cleaned, fuse the contact pressure data, gyroscope data and accelerometer data to obtain fused data, and judge whether the robot fits the inner wall of the pipe based on the fused data. If not, the angle adjustment amount of each joint of the robot is adjusted.

[0131] The specific implementation of the cleaning control system of the cable duct cleaning robot is basically the same as the specific implementation of the cleaning control method of the cable duct cleaning robot mentioned above, and will not be repeated here.

[0132] In one embodiment of the present application, a computer device is provided, which includes a memory and a processor, wherein a computer program is stored in the memory, and the above steps are implemented when the processor executes the computer program; the computer device provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiments, and will not be repeated here.

[0133] In one embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the above steps are implemented when the computer program is executed by a processor; the computer-readable storage medium provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiment, and will not be repeated here.

[0134] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0135] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0136] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cleaning control method for a cable duct cleaning robot, characterized in that: include: Acquire the inner wall image of the pipeline to be cleaned in real time, perform feature extraction on the inner wall image to obtain the pipe wall roughness feature and the pipe wall concave-convex degree feature, and obtain the pipe inner wall condition type according to the pipe wall roughness feature and the pipe wall concave-convex degree feature; Acquire the pipe wall distance data of the pipe to be cleaned in real time, obtain the pipe center line by using a curve fitting method based on the pipe wall distance data, and obtain the pipe curvature and curvature parameters according to the pipe center line; According to the pipeline inner wall condition type, the pipeline curvature and the curvature parameter, the angle adjustment amount of each joint of the robot is determined, the pipeline inner wall condition type is quantitatively scored using a fuzzy comprehensive evaluation method to obtain a scoring result, a corresponding table of scores and joint angle adjustments is established according to the scoring result, the scoring result and the joint angle adjustment amount are normalized and feature selected to obtain processed scoring results and joint angle adjustments, the angle adjustment amount of each joint of the robot is predicted using a trained neural network model, and a preliminary joint angle adjustment plan is obtained, wherein the neural network model is trained by a support vector regression algorithm using a radial basis function kernel; The coefficient of variation of the thickness of the attachment is calculated, and the preset threshold is set to 0.5 according to historical data statistics. If the coefficient of variation is greater than 0.5, the joint angle adjustment amount is increased on the basis of the preliminary joint angle adjustment plan, and the increase is 20% of the original adjustment amount. If the coefficient of variation is less than 0.5, the joint angle adjustment amount is reduced by 10% of the original adjustment amount. If the pipe curvature is greater than the preset threshold of 30° / m, the adjustment amount of the corresponding joint is increased by 15% of the original adjustment amount. The weight optimization method based on gradient descent is used to fine-tune the joint angle adjustment amount. The learning rate is set to 0.01, the number of iterations is set to 100, and the optimization objective function includes two indicators: pipe curvature adaptability and uniformity of attachment distribution. The weight coefficient of each joint is updated through the back propagation algorithm to achieve adaptive adjustment. The contact pressure data between the robot and the inner wall of the pipe to be cleaned is obtained, and the contact pressure data, the gyroscope data and the accelerometer data are fused to obtain fused data. According to the fused data, it is determined whether the robot fits the inner wall of the pipe. If not, the angle adjustment amount of each joint of the robot is adjusted.

2. The cleaning control method of the cable duct cleaning robot according to claim 1, characterized in that: The real-time acquisition of the inner wall image of the pipeline to be cleaned includes: Using a high-resolution camera to move and shoot along the axial direction of the pipeline to be cleaned to obtain an image sequence, wherein the image sequence includes a plurality of inner wall images; The image sequence is stitched using an image stitching algorithm based on feature point matching to obtain an unfolded image of the inner wall of the pipeline.

3. The cleaning control method of the cable duct cleaning robot according to claim 2, characterized in that: The method of obtaining the type of inner wall condition of the pipeline according to the pipe wall roughness characteristics and the pipe wall concave-convex degree characteristics includes: Performing median filtering on the pipeline inner wall development image to obtain a first image with noise removed; Using a histogram equalization method to perform contrast enhancement on the first image to obtain an enhanced panoramic image of the inner wall of the pipeline; Performing image segmentation and edge detection on the enhanced panoramic image of the inner wall of the pipeline to extract the roughness characteristics of the pipeline wall and the concave-convex degree characteristics of the pipeline wall; A support vector machine classifier based on a radial basis function kernel is used to classify the pipe wall roughness characteristics and the pipe wall concave-convex degree characteristics to obtain the type of attachment; Establishing an evaluation index system according to the roughness characteristics of the pipe wall, the concave-convex degree characteristics of the pipe wall, the thickness distribution data of the attachments, the types of attachments and the areas of the types of attachments; The weight coefficient of each indicator in the evaluation index system is determined, a fuzzy relationship matrix is ​​constructed, and a fuzzy synthesis operation is performed according to the fuzzy relationship matrix and the type of attachment to obtain the type of pipeline inner wall condition.

4. The cleaning control method of the cable duct cleaning robot according to claim 1, characterized in that: The method of obtaining the pipeline centerline based on the pipeline wall distance data by using a curve fitting method, and obtaining the pipeline curvature and curvature parameters according to the pipeline centerline includes: The quartile method is used to eliminate outliers in the pipeline wall distance data to obtain effective wall distance data; The effective wall distance data is fitted using a cubic spline interpolation method to obtain a pipeline cross-sectional profile, the geometric center point coordinates of the pipeline cross-sectional profile are calculated, and the pipeline centerline is obtained by fitting according to the geometric center point coordinates; Based on the center line of the pipeline, a least square objective function is established and a set of equations is solved to calculate the bending radius of the pipeline, the length of the bending section is calculated by integration, and the curvature parameter is calculated according to the length of the bending section using the Frenet formula; If the length of the curved section is greater than a first threshold, or the curvature parameter is greater than a second threshold, the pipeline section to be cleaned is determined to be an abnormal pipeline section, and the pipeline curvature and curvature parameter are output.

5. The cleaning control method of the cable duct cleaning robot according to claim 1, characterized in that: The contact pressure data, the gyroscope data and the accelerometer data are fused to obtain fused data, and according to the fused data, it is determined whether the robot fits the inner wall of the pipeline, and if not, the angle adjustment amount of each joint of the robot is adjusted, including: Using a Kalman filter algorithm to fuse the gyroscope data, the accelerometer data and the contact pressure data to obtain fused data, and acquire the posture information of the robot; The fused data and posture information are input into the constructed robot and pipe wall fit model for calculation to obtain a fit score; If the fit score is lower than the fit score threshold, it is determined that the robot is not in fit with the inner wall of the pipeline, and the posture information is processed using an adaptability model of the robot posture and the pipe curvature and pipe diameter change established by multivariate linear regression analysis to obtain a posture adaptability score; If the posture adaptability score is lower than the posture adaptability score threshold, the robot posture is judged to be incompatible with the pipeline characteristics, a posture adjustment instruction is obtained, and the angle adjustment amount of each joint of the robot is adjusted, wherein the posture adjustment instruction includes the adjustment value and execution order of each joint angle.

6. The cleaning control method of the cable duct cleaning robot according to claim 1, characterized in that: After adjusting the angle adjustment amount of each joint of the robot, the method further includes: Record the adjusted robot posture parameter data, pipeline inner wall condition type and cleaning working parameters; The robot posture parameter data, the inner wall condition of the pipeline and the cleaning working parameters are input into the cleaning effect evaluation model for calculation to obtain a cleaning effect score. If the cleaning effect score is lower than the effect threshold, the cleaning parameter optimization algorithm is triggered so that the robot uses a particle swarm optimization algorithm to adjust the rotation speed and pressure of the cleaning tool.

7. A cleaning control system for a cable duct cleaning robot, characterized in that: include: An extraction module is used to obtain an inner wall image of the pipeline to be cleaned in real time, perform feature extraction on the inner wall image, obtain pipe wall roughness features and pipe wall concavity features, and obtain the type of pipe inner wall condition based on the pipe wall roughness features and pipe wall concavity features; A fitting module is used to obtain the pipe wall distance data of the pipe to be cleaned in real time, obtain the pipe center line based on the pipe wall distance data by using a curve fitting method, and obtain the pipe curvature and curvature parameters according to the pipe center line; The first adjustment module is used to determine the angle adjustment amount of each joint of the robot according to the pipeline inner wall condition type, the pipeline curvature and the curvature parameter, quantitatively score the pipeline inner wall condition type using a fuzzy comprehensive evaluation method to obtain a scoring result, establish a corresponding table of scores and joint angle adjustment amounts according to the scoring results, normalize and feature select the scoring results and the joint angle adjustment amounts to obtain processed scoring results and joint angle adjustment amounts, use a trained neural network model to predict the angle adjustment amounts of each joint of the robot, and obtain a preliminary joint angle adjustment plan, wherein the neural network model is trained by a support vector regression algorithm using a radial basis function kernel; The coefficient of variation of the thickness of the attachment is calculated, and the preset threshold is set to 0.5 according to historical data statistics. If the coefficient of variation is greater than 0.5, the joint angle adjustment amount is increased on the basis of the preliminary joint angle adjustment plan, and the increase is 20% of the original adjustment amount. If the coefficient of variation is less than 0.5, the joint angle adjustment amount is reduced by 10% of the original adjustment amount. If the pipe curvature is greater than the preset threshold of 30° / m, the adjustment amount of the corresponding joint is increased by 15% of the original adjustment amount. The weight optimization method based on gradient descent is used to fine-tune the joint angle adjustment amount. The learning rate is set to 0.01, the number of iterations is set to 100, and the optimization objective function includes two indicators: pipe curvature adaptability and uniformity of attachment distribution. The weight coefficient of each joint is updated through the back propagation algorithm to achieve adaptive adjustment. The second adjustment module is used to obtain the contact pressure data between the robot and the inner wall of the pipe to be cleaned, fuse the contact pressure data, gyroscope data and accelerometer data to obtain fused data, and judge whether the robot fits the inner wall of the pipe according to the fused data. If not, adjust the angle adjustment amount of each joint of the robot.

8. A computer device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the cleaning control method of the cable duct cleaning robot as described in any one of claims 1 to 6 when executing the computer program.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the cable duct cleaning robot cleaning control method according to any one of claims 1 to 6 are implemented.

Images