A method for measuring the unevenness of a steel wire rope based on a robot

Abstract

The present application relates to the technical field of nondestructive testing, and particularly relates to a steel wire unevenness measurement method based on a robot, which comprises the following steps: robot positioning and initial calibration, steel wire surface pretreatment, unevenness preliminary scanning, high-precision three-dimensional profile measurement, real-time preprocessing of collected original data, automatic generation of a detailed measurement report, autonomous resetting after the measurement is completed, and prediction of the remaining life of the steel wire according to the measurement result of this time. According to the method, multi-modal sensors and an adaptive robot platform are deeply fused, and through a unique algorithm process of global positioning, local precision measurement, data fusion and dynamic compensation, unmanned, full-coverage and high-precision measurement of the three-dimensional appearance of the steel wire surface is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nondestructive testing, and particularly relates to a steel wire unevenness measurement method based on a robot. BACKGROUND

[0002] As a key load-bearing component, the surface unevenness, wear and broken wire of the steel wire are directly related to the safety of the overall structure. Traditional detection methods mainly rely on manual visual inspection or handheld detectors. Manual detection is slow, and the detection results depend on the experience of personnel. It is difficult to achieve continuous and full-coverage measurement with handheld devices, which may miss local defects and cannot accurately assess the depth / height and volume of unevenness such as depressions and protrusions.

[0003] At present, some steel wire detection devices based on machine vision are disclosed in the prior art, but they are mostly for static or indoor short ropes, and lack deep integration with robot motion control. There is an urgent need to overcome the shortcomings of the prior art and provide a full-automatic, high-precision, high-reliability and complex-environment-adaptable steel wire unevenness measurement method and system. SUMMARY

[0004] The present application relates to the technical field of nondestructive testing, and particularly relates to a steel wire unevenness measurement method based on a robot.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical scheme: a steel wire unevenness measurement method based on a robot, comprising the following steps:

[0006] Step S1, robot positioning and initial calibration, the robot automatically identifies the starting point of the steel wire through an integrated vision system and performs precise positioning, and the vision system captures the image of the steel wire;

[0007] Step S2, steel wire surface pretreatment, a non-contact cleaning device is mounted on the robot to clean the surface of the steel wire and remove impurities;

[0008] Step S3, preliminary scanning of unevenness, the robot moves quickly along the steel wire using a multi-sensor array to perform preliminary scanning and identify obvious uneven areas;

[0009] Step S4, high-precision three-dimensional profile measurement, the robot carrying a scanning module moves slowly along the steel wire to obtain detailed three-dimensional surface profile data;

[0010] Step S5, real-time data preprocessing, the robot performs real-time preprocessing on the collected raw data during movement;

[0011] Step S6, automatic generation of detailed measurement report, including unevenness distribution map, defect position coordinates, safety assessment and recommended measures;

[0012] Step S7, the robot is reset automatically after the measurement is completed, and the remaining life of the steel wire rope is predicted according to the measurement result.

[0013] The technical solution provided by the application has at least the following beneficial effects:

[0014] The application realizes unmanned, full coverage and high-precision measurement of the three-dimensional surface topography of the steel wire rope through a unique algorithm process of global positioning, local precision measurement, data fusion and dynamic compensation.

[0015] The robot system can automatically and accurately find the starting end of the steel wire rope to be measured, and establish a robot measurement coordinate system with the steel wire rope axis as the reference, and realize seamless connection from macro navigation to micro precision measurement by combining the wide angle of global vision and the micro distance of cross laser.

[0016] The application combines ultrasonic waves, eddy currents and static electricity, which are three physical principles, and forms a complete cleaning chain for adhesion, blowing and suspended residues, which is much more efficient than a single method.

[0017] The application combines visible light surface texture imaging and near-infrared transmission imaging, which can detect surface geometric irregularities and subsurface material defects at the same time, greatly improving the comprehensiveness and reliability of preliminary scanning.

[0018] The application combines robot vibration, thermal deformation, kinematic error and other error sources to establish a comprehensive compensation model, which improves the absolute accuracy of the data from the system level, which surpasses the simple sensor denoising, improves the non-local mean filter algorithm, and makes it able to distinguish noise and real surface features, avoiding the smoothing effect of traditional filters on small defects. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, below will briefly introduce the drawings required in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The method step diagram provided by the embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to further elaborate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects of the robot-based steel wire unevenness measurement method according to the present application will be described in detail below in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0023] The following embodiments are for illustrative purposes only and are not intended to limit the scope of the present application.

[0024] The specific scheme of the robot-based steel wire unevenness measurement method provided by the present application will be specifically described below in combination with the drawings.

[0025] EMBODIMENT

[0026] Please refer to Figure 1 which shows the method flowchart of the robot-based steel wire unevenness measurement method provided by one embodiment of the present application, and the method comprises the following steps:

[0027] Step S1, robot positioning and initial calibration, the robot automatically identifies the starting point of the steel wire through the integrated vision system and performs accurate positioning, and the vision system captures the steel wire image;

[0028] In step S1, the following sub-steps are further included:

[0029] S1-1, select a six-degree-of-freedom high-precision industrial robot, fix a high-resolution global CCD camera above the measurement scene, the field of view covers the entire steel wire to be measured, use the fixed global camera to obtain the two-dimensional image of the scene, and identify the approximate position and direction of the steel wire through image processing technology, to provide navigation information for the robot to move to the approximate starting point;

[0030] The global camera takes a photo of a scene containing a steel wire rope, and a semantic segmentation model based on deep learning is applied, which is specially trained to quickly identify the steel wire rope category and effectively distinguish the steel wire rope from a complex background.

[0031] The segmented steel wire rope region is skeletonized to obtain the pixel-level center line of the steel wire rope, and the center line is fitted as a straight line using the least squares method to obtain its equation in the image coordinate system. The global camera hand-eye calibration parameters are used to convert the image center line to a rough three-dimensional space straight line L in the robot base coordinate system. g This straight line defines the approximate direction and position of the steel wire rope, and the control system guides the robot to move to a safe preliminary position above the starting end of the steel wire rope. g

[0032] S1-2, after the robot is in place, the cross-line laser integrated at the end is projected onto the surface of the steel wire rope, forming a deformed laser line. By analyzing the shape of the deformed laser line, the three-dimensional profile of the steel wire rope surface can be inversely calculated, and the center axis can be accurately calculated.

[0033] The cross-line laser at the end of the robot is turned on, projecting a horizontal and vertical laser line onto the steel wire rope. A macro CMOS camera synchronously captures the image of the projection area. Since the steel wire rope is a cylinder, the horizontal laser line will become an arc, and the vertical laser line will become a curve following the surface of the steel wire rope. The horizontal laser arc in the image is extracted, and this arc is regarded as the outer contour of the steel wire rope at that cross section. An ellipse fitting algorithm is used to fit the contour, and the center of the fitted ellipse is the approximate center point of the cross section.

[0034] Extract the vertical laser line, and move the robot end along the rough direction of the steel wire rope by a small distance, and continuously capture multiple images during this period. Extract the center line of the vertical laser line in each frame, and fuse the multiple cross-section center points obtained by the horizontal laser line with the point cloud extracted by the vertical laser line. Use the random sample consensus algorithm to fit the fused three-dimensional point set to a spatial straight line, and finally obtain a high-precision steel wire axis L p The spatial straight line fitting is usually defined by a point P0(x0, y0, z0) and a direction vector V(a, b, c), and the goal of the RANSAC algorithm is to find P0 and V that minimize the sum of the distances of all inliers to the straight line.

[0035] The distance formula from a point to a straight line is:

[0036] d=|(P i -P0)×V| / |V|

[0037] ​Wherein, x represents the vector cross product, by minimizing the sum of all the inner points d, get the optimal L p .

[0038] S1-3, dynamic establishment of robot measurement coordinate system and verification, the steel wire axis L p As a reference, dynamic definition of the measurement coordinate system of the robot, and verify the accuracy of the coordinate system:

[0039] Coordinate system Z axis, that is, the measurement direction, coincides with the direction vector V of L p , the positive direction is the measurement direction;

[0040] Origin O, set as the nearest point of L p from the origin of the robot base coordinate system, usually close to the starting end of the steel wire;

[0041] X axis, defined as the direction perpendicular to Z axis and pointing directly above the surface of the steel wire;

[0042] Y axis, determined by the right hand rule, complete the construction of the coordinate system {M};

[0043] Accurately calculate the homogeneous transformation matrix of the robot base coordinate system {B} to the newly defined measurement coordinate system {M}, all sensor data will be converted to {M} coordinate system for processing, establish the coordinate conversion relationship, the robot moves a small distance along the newly defined Z axis, repeat the process of the above step, calculate the axis L v of a small piece of steel wire again, compare the direction and position deviation of L v and L p , if the deviation is within the threshold, it means the calibration is successful, otherwise, fine tuning iteration.

[0044] In unstructured environment, the robot system can automatically and accurately find the starting end of the steel wire to be measured, and establish a robot measurement coordinate system with the steel wire axis as the reference, combined with the wide angle of global vision and the micro distance of cross laser, it realizes seamless connection from macro navigation to micro measurement, has strong anti-environmental interference ability, uses one laser line for section center positioning and another for axial guidance, through multi-frame data fusion and RANSAC fitting, it can more accurately restore the spatial axis, the coordinate system is dynamically generated before each measurement according to the actual spatial pose of the steel wire, perfectly adapts to the bending, suspension and other uncertainties of the steel wire, provides a unique and accurate reference for the subsequent steps, and increases the verification closed loop, ensures the reliability of the initial calibration, guarantees the measurement accuracy of the whole system from the first step, avoids error transmission and accumulation.

[0045] Step S2, steel wire surface pretreatment, non-contact cleaning device is carried on the robot, the surface of the steel wire is cleaned, and the impurities are removed;

[0046] wherein in step S2, further comprising the following sub-steps:

[0047] S2-1, a six-degree-of-freedom high-precision industrial robot is integrated with a pretreatment executor at the end of the robot, the core of which is a ring-shaped processing head capable of 360-degree wrapping around the steel wire rope without contact, an ultrasonic transducer array is installed on the inner wall of the ring-shaped head, which can generate low-frequency ultrasonic waves, a plurality of precisely designed tangential vortex nozzle is installed on the ring-shaped head, which is connected to a high-pressure gas source to generate a high-speed rotating air knife, a ring-shaped electrostatic adsorption ring with a high-voltage electrostatic generator can adsorb the peeled-off particles, and a miniature spectral confocal sensor is installed in the ring of the ring-shaped head to monitor the surface cleanliness in real time; the high-pressure gas source, the ultrasonic wave generator and the high-voltage electrostatic generator are all moved with the robot through pipelines;

[0048] Oil stains can cause special diffuse reflection and color change of laser lines, water film can produce mirror reflection spot, rust and dust can change the texture characteristics of the surface, in the image saved in the last step, the clarity of the laser line profile, color space features, texture features are extracted;

[0049] Clarity of laser line profile: use image gradient operator to calculate average gradient value G avg . Oil stains and water film can reduce the gradient value.

[0050] Color space features: in HSV color space, analyze the saturation S and lightness V of the laser line region, oil stains usually cause specific changes in S and V.

[0051] Texture features: use local binary pattern LBP to analyze the texture uniformity of the regions on both sides of the laser line.

[0052] Construct a pollution index C i , which is the weighted sum of the above-mentioned multiple features:

[0053] C i =w1*(1-G avg norm)+w2*ΔS+w3*ΔV+w4*LBP

[0054] Where w1-w4 represent weight coefficients, which are calibrated through experiments, G avg norm represents the normalized gradient value; ΔS represents the saturation difference between the current sample and the clean sample after normalization to 0-1 using the Min-Max formula, ΔV represents the difference in the lightness channel between the current sample and the clean sample after normalization to 0-1 using the Min-Max formula; LBP is the texture entropy, which is normalized to 0-1 according to the C i value and the dominant feature, and the pollution degree is divided into mild, moderate and severe according to the pollution index, and the pollution type is preliminarily judged;

[0055] S2-2, intelligent matching and execution of multi-modal pretreatment parameters, dynamically adjusting the intensity, action time and execution order of different cleaning modes for different pollution combinations, presetting an expert knowledge base, inputting pollution type and pollution degree, and finally outputting the optimal pretreatment parameter combination, including ultrasonic power, cyclone pressure, static voltage and processing time;

[0056] For example, for heavy oil pollution, high ultrasonic power + medium cyclone pressure + low static voltage is used, the oil film is first broken by ultrasonic waves, then large droplets are blown away by cyclone, and finally residual oil mist is adsorbed by static electricity; for light dust, low ultrasonic power or off + high cyclone pressure + medium static voltage is used, cyclone is mainly used for direct blowing, and static electricity is used to prevent dust raising.

[0057] The robot carries a ring-shaped treatment head and moves along the Z-axis of the measurement coordinate system defined in step one at a constant speed, while starting the corresponding pretreatment modules according to the above parameter combination. Low-frequency ultrasonic waves produce cavitation effects in the medium on the surface of the steel wire rope, forming tiny bubbles and breaking them, producing a powerful impact force that causes the contaminants to peel off the substrate. The action strength is related to the power and frequency. High-pressure gas is ejected from the tangential nozzle to form a high-speed rotating vortex field in the ring-shaped cavity. This air knife can produce extremely high local shear force with relatively low overall gas consumption, efficiently removing the peeled contaminants. The blowing efficiency is related to the cyclone pressure and the angular velocity of the air flow. The static adsorption ring applies high voltage to generate a strong electrostatic field around it. Micron and sub-micron charged or uncharged particles that are blown up by the cyclone but not completely removed will be attracted to the inner wall of the electrostatic ring due to induced charging or polarization, completely removing the measurement area.

[0058] S2-3, use a miniature spectral confocal sensor integrated in the pretreatment head to perform point measurement on the same position immediately after cleaning. The sensor determines the exact distance to the surface by analyzing the spectrum of the reflected light, and analyzes the intensity of the reflected light, which can indirectly evaluate the cleanliness of the surface.

[0059] If it does not meet the standard, the robot control system will record the coordinates of the position, and may automatically return to the point after completing a trip, dynamically adjust the pretreatment parameters according to the real-time measurement of the pollution characteristics for secondary processing until it meets the standard. For example, low reflectivity may indicate residual oil pollution, and the ultrasonic action time can be increased.

[0060] Multimodal synergy combines three physical principles—ultrasound, vortex cyclone, and electrostatics—to target the three stages of adhesion, purging, and suspended residue, forming a complete cleaning chain with efficiency far exceeding that of a single method. High-precision spectral confocal sensors are introduced for in-situ quality inspection, forming a closed-loop control system to ensure reliable pretreatment results. This provides a high-quality surface guarantee for subsequent preliminary scanning of unevenness, avoiding misjudgments caused by incomplete cleaning. The entire process is completely non-contact, avoiding methods such as brushing that may scratch the wire rope or cause secondary pollution.

[0061] Step S3: Preliminary scan of unevenness. The robot uses a multi-sensor array to move quickly along the steel cable to perform a preliminary scan and identify obvious uneven areas.

[0062] Step S3 further includes the following sub-steps:

[0063] The S3-1 is a six-DOF high-precision industrial robot. Its end effector is equipped with a scanning module. The core sensors of the scanning module include: a high-frame-rate linear CMOS camera for high-speed surface texture acquisition; a near-infrared (NIR) linear camera for detecting potential damage below the surface; a lightweight, low-friction encoder-integrated wheel assembly that maintains contact with the steel cable to accurately measure the robot's displacement along the cable; dual-sided high-brightness LED linear light sources; a near-infrared backlight source for transmission imaging; and a high-performance FPGA board for real-time synchronization and preprocessing of sensor data.

[0064] S3-2 utilizes the differences in the interaction between light and steel wire rope in different wavelength bands to obtain complementary surface and subsurface information. Visible light imaging reflects macroscopic geometry and surface texture, while near-infrared light is more sensitive to density changes caused by internal corrosion of materials.

[0065] The robot's end effector moves at a relatively high, constant speed along the Z-axis of the measurement coordinate system defined in step one. The speed is maintained by real-time feedback from the encoder wheel assembly. Each minute movement of the encoder wheel assembly generates a trigger pulse, which synchronously triggers both the visible light array camera and the NIR line array camera to acquire one line of images. The processing unit precisely controls the timing to ensure strict spatial alignment of the two image streams. After continuous acquisition, two two-dimensional image data streams (I) are obtained. v (x,y) and I n (x,y), where the x-direction represents the axial direction of the wire rope and the y-direction represents the circumferential direction around the rope;

[0066] S3-3, Real-time anomaly detection based on image sequence, normal steel wire surface under uniform illumination should present a continuous and regular texture pattern, any unevenness will destroy this pattern, which is manifested as a mutation of image gray level, texture or edge feature, the average gray level gradient AGG, local binary pattern variance LBPV and transmittance anomaly index TAI based on NIR image of each image line or a small image block are calculated in parallel on FPGA.

[0067] Since the steel wire itself may have slow diameter changes or slight bending, fixed threshold is not applicable, therefore a statistical process control method based on sliding window is adopted to calculate the mean μ and standard deviation σ of each feature value in the current window, and set the control limit;

[0068] For example, the anomaly criterion, if the feature value f c satisfies |f c -μ|>k*σ, mark this point as "potential abnormal point", where k is a constant, which can be adjusted according to requirements.

[0069] When multiple features exceed the threshold at the same time, or a certain feature is seriously out of the limit, it is finally determined as a suspicious region segment, which greatly reduces the possibility of false positives.

[0070] S3-4, Aggregating continuous or closely spaced abnormal points into suspicious region segments, for each region segment, calculate its starting position and ending position, the average value of the feature deviation of all abnormal points in the region, and preliminarily judge whether it is a surface geometric defect or a material internal defect according to the feature that triggered the anomaly, and finally output a structured list.

[0071] Combining visible light surface texture imaging and near-infrared transmission imaging, surface geometric unevenness and subsurface material defects can be detected at the same time, greatly improving the comprehensiveness and reliability of preliminary scanning, using statistical process control method, the threshold can be dynamically adjusted according to the background texture of the steel wire itself, effectively overcoming the interference caused by uneven global lighting or natural bending of the rope body, and significantly reducing the false positive rate.

[0072] Step S4, high-precision three-dimensional profile measurement, the robot carrying the scanning module moves slowly along the steel wire to obtain detailed three-dimensional surface profile data;

[0073] In step S4, the following sub-steps are further included:

[0074] The S4-1 high-precision industrial robot end effector, equipped with a dedicated scanning module, also includes several ring-distributed high-precision spectral confocal probes, a miniaturized Mirau-type white light interferometer objective lens, and a high-speed galvanometer system. Based on the suspicious area segments generated in the previous step, the robot's motion path and measurement mode are dynamically planned to achieve optimal allocation of measurement resources. The robot path planning algorithm sorts all suspicious areas according to their positional order and calculates the shortest movement path to avoid unnecessary round trips. Areas with high anomaly intensity are marked as fine measurement areas, where the robot will move at low speed and simultaneously activate the spectral confocal array and the white light interferometer module. Areas with low anomaly intensity are marked as rapid survey areas, where the robot moves at higher speed and only activates the spectral confocal array for low-density data acquisition, used for contour verification and transition area measurement. The field of view and sampling density of the white light interferometer scan are preset according to the anomaly intensity, with high-resolution areas using a small field of view and high-density scanning to capture details.

[0075] S4-2, the robot moves to the starting point of the first suspicious area, and multiple spectral confocal probes work simultaneously. As the robot moves along the Z-axis, it rapidly acquires the three-dimensional contour point cloud P of the measurement strip on the surface of the wire rope. ccd (x, y, z), a confocal probe is selected and designated as the master probe, whose measured value d m It is sent to the control system in real time, and the control system according to d m and the preset ideal working distance d i The focal deviation Δd is calculated as follows:

[0076] Δd=d m -d i ;

[0077] Based on the PID control algorithm, the PZT active focusing mechanism is driven in real time to adjust the Z-axis position of the white light interferometer objective, ensuring that its focus always tracks the surface undulations of the wire rope.

[0078] u(t) = K p *e(t)+K i *∫e(t)dt+K d *de(t) / dt

[0079] Where u(t) represents the control voltage output to PZT, e(t) represents the current focus deviation Δd, and K p K i K d These represent the proportional, integral, and differential coefficients, respectively. The mechanism ensures that white light interferometry is always in optimal working condition, avoiding signal attenuation or measurement failure caused by defocusing.

[0080] White light interferometry measures surface topography by analyzing the characteristics of interference fringes generated by a broadband light source as the optical path difference changes, the robot moves to the predetermined fine measurement point triggers white light interference scanning, the PZT drives the reference mirror to move N steps with equal step size Δz, at each step position, the hyperspectral camera captures an interference image I n (x,y), where n = 1, 2,..., N;

[0081] For each pixel point (x, y), its light intensity value changes with the position of the reference mirror to form an interference signal envelope, a five-step phase shift algorithm is used to calculate the phase, and due to the periodicity of the phase, the wrapped phase is obtained, which needs to be unwrapped to obtain the continuous phase distribution, and finally the three-dimensional topography data of a small area with nanometer resolution is obtained;

[0082] Precise registration and fusion of multi-source point cloud data, all data are converted to the measurement coordinate system established in step one, using the features shared by the two point clouds in the overlapping area, using the iterative closest point algorithm for fine registration, eliminating system errors caused by robot positioning errors and sensor installation deviations;

[0083] The three-dimensional topography data of the fine measurement area replaces the corresponding spectral confocal point cloud data; the spectral confocal point cloud data is retained in the rapid survey area, and finally a complete, seamless, three-dimensional contour model with different accuracies in different areas is generated.

[0084] Dynamic optimization of measurement path and parameters, so that high-cost high-precision measurement resources are concentrated only in the most needed places, greatly improving the overall efficiency, introducing a closed-loop focus control based on PID feedback, solving the defocusing problem when performing high-precision optical measurement on undulating surfaces, ensuring the stability of the white light interference measurement conditions, which is the key to obtaining reliable data, and through the precise registration algorithm, different sources and different accuracy point clouds are fused into a unified high-quality data set.

[0085] Step S5, real-time data preprocessing, the robot in the moving process, the raw data collected is preprocessed in real time;

[0086] In step S5, the following sub-steps are further included:

[0087] S5-1, streaming data reception and cache management, the preprocessing system receives and processes the continuously generated data in the previous step in a streaming manner, avoiding data accumulation, the system divides the continuous point cloud stream into data blocks according to fixed length or fixed time interval, when a data block is being processed, the next data block is stored in another cache area, realizing seamless connection of each data block, each data block is bound with its corresponding robot pose, timestamp, sensor ID metadata;

[0088] S5-2, the robot end is integrated with an inertial measurement unit (IMU), the acceleration and angular velocity measured by the IMU are double-integrated to estimate the end displacement caused by vibration, temperature sensors T_sensor are arranged at key positions of the robot and sensor, and the thermal deformation δ is calculated according to the thermal expansion coefficient α of the material:

[0089] δ = α * L * (T s -T c )

[0090] Wherein, L represents the characteristic length, T s represents the temperature sensor data, T c represents the temperature at the time of system calibration, the small deviation between the theoretical pose and the actual command pose is calculated by using the encoder feedback of the robot itself and the kinematics model, the previous error term is synthesized into a comprehensive error vector to compensate for each point P o (x, y, z) in the current data block.

[0091] The point cloud contains not only real surface details but also random noise, the traditional filter blurs the details while denoising, parallel search based on KD-tree is used to quickly estimate the normal vector of each point, and the curvature c i of each point is calculated, and the curvature threshold is set to divide the points into "high feature points" and "flat points":

[0092] c i ≈ λ min / (λ0+λ1+λ2),

[0093] Wherein λ represents the eigenvalue of the covariance matrix, for the "flat point" area, a larger search radius is used for strong smoothing filtering to effectively suppress noise, for the "high feature point" area, a minimal search radius is used for light filtering, or the filtering is directly skipped to maximize the preservation of sharp edges and details of defects, after the above filtering, the remaining obvious outliers are removed; the space is divided into a uniform voxel grid, only one point is retained in each voxel, a larger voxel is used in rapid survey area, and a smaller voxel is used in fine measurement area, to realize multi-resolution data simplification.

[0094] For large-scale point cloud data, a pipeline processing flow is designed, combined with double buffering and parallel computing, real-time preprocessing is realized, multiple error sources such as robot vibration, thermal deformation and kinematics error are fused, a comprehensive compensation model is established, and the absolute accuracy of the data is improved from the system level, which surpasses the simple sensor denoising, improves the non-local mean filter algorithm, and makes it able to distinguish noise and real surface features, avoiding the smoothing effect of traditional filters on small defects.

[0095] Step S6, automatically generate a detailed measurement report, including unevenness distribution map, defect position coordinates, safety assessment and recommended measures;

[0096] In step S6, the preprocessed data of steps 3-5 are integrated by Python Pandas, heat maps and 3D models are generated using Matplotlib, risk levels are assessed based on ISO 4309 standard algorithms, and finally PDF reports are output by Jinja2 rendering LaTeX templates, transmitted to the cloud using MQTT protocol, and the built-in data conflict detection mechanism ensures the reliability of the report. When data is missing, the original scan path log is retained, and conflict data triggers a warning flag.

[0097] Step S7, after the robot completes the measurement, it automatically resets and predicts the remaining life of the steel wire rope based on the measurement results;

[0098] In step S5, after the robot completes the measurement, the remaining life is dynamically calculated by the built-in life prediction algorithm, which is trained based on historical wear data. The algorithm automatically triggers a three-level early warning system: when the remaining life exceeds 30 days, a green maintenance prompt is sent, 15-30 days triggers a yellow warning, and 15 days triggers a red emergency alarm. At the same time, the maintenance time, defect location and recommended measures are pushed to the cloud management platform through the NB-IoT module, realizing the closed-loop management from detection to decision.

[0099] In this way, a robot-based steel wire rope unevenness measurement method can be realized.

[0100] The above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for measuring irregularity of a steel wire rope based on a robot, characterized by, The method comprises: Step S1, robot positioning and initial calibration, the robot automatically identifies the starting point of the steel wire rope through the integrated vision system and performs accurate positioning, and the vision system captures the image of the steel wire rope; Step S2, steel wire rope surface pretreatment, a non-contact cleaning device is mounted on the robot to clean the surface of the steel wire rope and remove impurities; Step S3, preliminary scanning of unevenness, the robot moves along the steel wire rope at a high speed using a multi-sensor array to perform preliminary scanning and identify obvious uneven areas; Step S4, high-precision three-dimensional profile measurement, the robot carrying a scanning module moves slowly along the steel wire rope to obtain detailed three-dimensional surface profile data; Step S5, real-time data preprocessing, the robot performs real-time preprocessing on the collected original data during movement; Step S6, automatic generation of detailed measurement report, including unevenness distribution map, defect position coordinates, safety evaluation and recommended measures; Step S7, after the robot completes the measurement, it automatically resets and predicts the remaining life of the steel wire rope according to the measurement results; In step S3, the following sub-steps are further included: S3-1, a six-degree-of-freedom high-precision industrial robot, the end of which is also mounted with a scanning module, the core sensors of the scanning module include: a high-frame-rate linear array CMOS camera for high-speed surface texture acquisition; a near-infrared (NIR) linear array camera for detecting potential damage below the surface layer; a lightweight low-friction encoder integrated wheel set that maintains contact with the steel wire rope to accurately measure the displacement of the robot along the rope; a double-sided high-brightness LED linear light source; a near-infrared backlight light source for transmission imaging; a high-performance FPGA board for real-time synchronization and preprocessing of sensor data; S3-2, by utilizing the differences in the interaction of different wavebands with the steel wire rope, complementary surface and subsurface information is obtained, visible light imaging reflects the macro geometric appearance and surface texture, while near-infrared light is more sensitive to density changes caused by internal rust of the material; The robot end effector moves along the Z-axis of the measurement coordinate system at a relatively high constant speed, the motion speed is real-time feedback by the encoder wheel set to ensure stability, the encoder wheel set generates a trigger pulse for every small displacement, which synchronously triggers the visible light linear array camera and the NIR linear array camera to collect one row of image respectively, through the processing unit to accurately control the timing, ensure that the two image data streams are strictly aligned in space, after continuous acquisition, two two-dimensional image data streams are obtained respectively; S3-3, real-time anomaly detection based on image sequence, the normal steel wire rope surface under uniform illumination should present a continuous and regular texture pattern, any unevenness will destroy this pattern, which is manifested as a sudden change in image gray, texture or edge feature, the average gray gradient, local binary pattern variance and transmittance anomaly index based on NIR image of each image row or a small image block are calculated in parallel on the FPGA, when multiple features exceed the threshold value at the same time, or a certain feature is seriously out of standard, it is finally determined as a suspicious area segment, greatly reducing the possibility of false positives; S3-4, the continuous or closely spaced abnormal points are aggregated into suspicious area segments, for each area segment, the starting position and ending position, the average value of the feature deviation of all abnormal points in the area are calculated, and it is preliminarily judged whether it is a surface geometric defect or a material internal defect according to the triggered feature; In step S4, the following sub-steps are further included: S4-1, the high-precision industrial robot end carries a special scanning module, which includes a number of annularly distributed high-precision spectral confocal probes, a miniaturized Mirau type white light interference objective and a high-speed galvanometer system, according to the generated suspicious area segment, the robot path planning algorithm sorts all the suspicious areas in order according to the position and calculates the shortest moving path to avoid empty return; S4-2, the robot moves to the starting point of the first suspicious area, multiple spectral confocal probes work simultaneously, and a three-dimensional profile point cloud P of the measurement band of the steel wire rope surface is quickly acquired as the robot moves along the Z axis ccd (x, y, z), a confocal probe selected is designated as the master probe, and its measurement value d m is sent to the control system in real time, and the control system calculates the focal point deviation Δd according to d m and the preset ideal working distance d i . Δd = d m - d i ; Based on the PID control algorithm, the PZT active focusing mechanism is driven in real time to adjust the Z direction position of the white light interference objective, so that the focus point always tracks the ups and downs of the steel wire rope surface: u(t) = K p *e(t) + K i *∫e(t)dt + K d *de(t) / dt; where u(t) represents the control voltage output to the PZT, e(t) represents the current focus deviation Ad, K p , K i , K d respectively represent the proportional, integral, and differential coefficients. This algorithm ensures that the white light interferometer is always in the best working state, avoiding signal attenuation or measurement failure caused by defocusing.

2. The robot-based steel wire rope unevenness measurement method according to claim 1, characterized in that: In step S1, the following sub-steps are further included: S1-1, a six-degree-of-freedom high-precision industrial robot is selected, a high-resolution global CCD camera is fixed above the measurement scene, the field of view covers the entire steel wire rope to be measured, a two-dimensional image of the scene is obtained by using the fixed global camera, the approximate position and direction of the steel wire rope are recognized by image processing technology, and navigation information is provided for the robot to move to the approximate starting point; S1-2, after the robot is in place, the cross line laser integrated at the end is projected onto the surface of the steel wire rope, the laser line is deformed, the three-dimensional profile of the surface of the steel wire rope can be inversely calculated by analyzing the shape of the deformed laser line, and the central axis can be accurately calculated; S1-3, dynamically establish robot measurement coordinate system and verify, the steel wire rope axis L p As a reference, dynamically define the measurement coordinate system of the robot, and verify the accuracy of the coordinate system. Accurately calculate the homogeneous transformation matrix of the robot base coordinate system {B} to the newly defined measurement coordinate system {M}. All sensor data will be uniformly converted to the {M} coordinate system for processing. Establish the coordinate conversion relationship. Move the robot along the newly defined Z-axis by a small distance. Repeat the previous process. Calculate the axis L of a small section of steel wire rope again. v Compare the direction and position deviation of L v and L p . If the deviation is within the threshold, the calibration is successful.

3. The robot-based steel wire rope unevenness measurement method according to claim 1, characterized in that: In step S2, the following sub-steps are further included: S2-1, the six-degree-of-freedom high-precision industrial robot is integrated with a pretreatment executor at the end, the core of which is a ring-shaped processing head that can surround the steel wire rope at 360 degrees without touching, the inner wall of the ring-shaped head is installed with an ultrasonic transducer array that can generate low-frequency ultrasonic waves, the ring-shaped head is installed with a plurality of precisely designed tangential eddy current cyclone nozzles that can generate high-speed rotating air knives, the ring-shaped head is installed with a high-voltage electrostatic generator, and the ring-shaped electrostatic adsorption ring can adsorb the peeled particles, and the micro-spectral confocal sensor integrated in the ring of the ring-shaped head can monitor the surface cleanliness in real time; The high-pressure gas source, ultrasonic generator and high-voltage electrostatic generator are moved with the robot through pipelines; Oil stains can cause special diffuse reflection and color change of the laser line, water film can produce mirror reflection spot, rust and dust can change the texture characteristics of the surface, and the clarity, color space characteristics and texture characteristics of the laser line profile are extracted from the image saved by the CCD camera; A pollution index C is constructed i which is a weighted sum of the above features: C i = w1*(1 - G avg norm) + w2*ΔS + w3*ΔV + w4*LBP; wherein w1-w4 represent weight coefficients, calibrated by experiments, G avg norm represents the normalized gradient value; ΔS represents the saturation difference between the current sample and the clean sample after normalization to 0-1 by the Min-Max formula, ΔV represents the difference in the lightness channel between the current sample and the clean sample after normalization to 0-1 by the Min-Max formula; LBP is the texture entropy, and C i The value and the dominant feature are normalized to 0-1 by the Min-Max formula, the pollution degree is divided into mild, moderate and severe according to the pollution index, and the pollution type is preliminarily judged. S2-2, intelligent matching and execution of multi-modal pretreatment parameters, dynamically adjusting the intensity, action time and execution order of different cleaning modes for different pollution combinations, presetting an expert knowledge base, inputting pollution type and pollution degree, and finally outputting the optimal pretreatment parameter combination including ultrasonic power, cyclone pressure, static voltage and processing time; S2-3, using a miniature confocal spectral sensor integrated in the pretreatment head to perform point measurement on the same position immediately after cleaning, the sensor determines the accurate distance to the surface by analyzing the spectrum of reflected light, and analyzes the reflected light intensity to indirectly evaluate the surface cleanliness.

4. The method according to claim 3, wherein in step S4, the white light interferometry measures the surface topography by analyzing the characteristics of the interference fringes generated by the broadband light source with the change of the optical path difference, the robot moves to the predetermined fine measurement point to trigger the white light interference scanning, at each step position, the hyperspectral camera captures an interference graph, for each pixel point (x, y), the light intensity value changes with the reference mirror position to form an interference signal envelope, the phase is calculated using the phase shift algorithm, and since the phase is periodic, the wrapped phase is obtained, which needs to be unwrapped to obtain the continuous phase distribution, and finally the three-dimensional topography data of a small area with nanometer resolution is obtained. Precise registration and fusion of multi-source point cloud data, all data are converted to the measurement coordinate system established in step one, the common features of the two kinds of point clouds in the overlapping area are used, and the iterative closest point algorithm is used for fine registration to eliminate the system error caused by the robot positioning error and the installation deviation of the sensor.

5. The method according to claim 1, wherein in step S5, the following sub-steps are further included: S5-1, stream data receiving and cache management, the pretreatment system receives and processes the data continuously generated in the previous step in a streaming manner to avoid data accumulation, the system divides the continuous point cloud stream into data blocks according to fixed length or fixed time interval, when a data block is being processed, the next data block is stored in another cache area to realize seamless connection, and each data block is bound with its corresponding robot pose, time stamp and sensor ID metadata; S5-2, the robot end is integrated with an inertial measurement unit IMU, the acceleration and angular velocity measured by the IMU are double integrated to estimate the end displacement caused by vibration, temperature sensors T_sensor are arranged at the key positions of the robot and the sensor, and the thermal deformation δ is calculated according to the thermal expansion coefficient α of the material:

6. The method according to claim 5, wherein in step S6, the following sub-steps are further included:

7. The method according to claim 1, wherein in step S7, the following sub-steps are further included: δ = a * L * (T s - T c ); where L denotes the characteristic length, T s denotes the temperature sensor data, T c denotes the temperature at the time of system calibration, the small deviation between the theoretical pose and the actual command pose is calculated using the encoder feedback of the robot itself and the kinematics model, and the previous error terms are synthesized into a comprehensive error vector to compensate for each point P o (x, y, z) in the current data block. ​ Wherein in step S5, the point cloud contains both real surface details and random noise, traditional filters will blur the details while denoising, using parallel search based on KD-tree to quickly estimate the normal vector of each point, calculate the curvature c of each point i , set the curvature threshold to divide the points into "high feature points" and "flat points", for the "flat point" area, use a larger search radius for strong smoothing filtering to effectively suppress noise, for the "high feature point" area, use a minimal search radius for light filtering, or directly skip filtering to maximize the preservation of the sharp edges and details of the defects, after the above filtering, the remaining obvious outliers are removed; divide the space into a uniform voxel grid, only one point is retained in each voxel, use larger voxels in the rapid survey area and smaller voxels in the fine measurement area to achieve multi-resolution data simplification. ​ Wherein in step S6, the preprocessed data of steps 3-5 are integrated by Python Pandas, heat maps and 3D models are generated using Matplotlib, risk levels are evaluated based on ISO 4309 standard algorithms, and finally a PDF report is rendered by Jinja2 LaTeX template and transmitted to the cloud using MQTT protocol. Meanwhile, a data conflict detection mechanism is built to ensure the reliability of the report. In case of data missing, the original scanning path log is preserved. Conflict data triggers a warning marker.

8. The method according to claim 1, wherein the method further comprises: determining the remaining life of the steel wire rope based on the measured data. Wherein in step S7, after the robot completes the measurement, the remaining life is dynamically calculated by the built-in life prediction algorithm, which is trained based on historical wear data. The algorithm automatically triggers a three-level warning system: when the remaining life exceeds 30 days, a green maintenance prompt is sent; when it is between 15-30 days, a yellow warning is triggered; and when it is within 15 days, a red emergency alarm is started. At the same time, the maintenance time, defect location and recommended measures are pushed to the cloud management platform through the NB-IoT module, realizing the closed-loop management from detection to decision.

Images