Insulator mold cleaning method with cooperation of multiple mechanical arms

Abstract

The invention discloses a multi-mechanical-arm collaborative insulator mold cleaning method which is specifically implemented according to the following steps that firstly, an insulator mold cleaning assembly is arranged, the assembly comprises an insulator mold to be cleaned and a blowing mechanical arm installed right in front of the insulator mold, and a high-pressure nozzle is arranged at the tail end of the blowing mechanical arm; the high-pressure nozzle is driven by an air pressure system, the air pressure system is connected with the blowing mechanical arm through a pipeline, provides high-pressure gas for the blowing mechanical arm and executes the blowing action, and a pressure sensor is further arranged on the blowing mechanical arm; the device further comprises a projection detection mechanical arm located above the side of the insulator mold, and the projection detection mechanical arm integrates a stripe projector and a camera and is used for detecting the surface cleanliness. The device further comprises a dust-sticking mechanical arm and the like. Then self-adaptive dust blowing is carried out, the mold is cleaned through a dust-sticking mechanical arm, and finally the cleaning effect is evaluated. The problems that in the prior art, the detection precision of the high-reflection three-dimensional complex curved surface is insufficient, and regulation and control in the cleaning process are extensive are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of insulator mold cleaning, and in particular relates to an insulator mold cleaning method coordinated by multiple robotic arms. Background Art

[0002] Three-dimensional vision measurement technology can obtain three-dimensional topographic information of an object's surface with high precision, and is widely used in the fields of surface reconstruction, defect detection, reverse engineering, etc. Among them, various metal parts in major national equipment projects such as aviation, aerospace, shipbuilding, energy and power, such as turbine blades, aircraft, ship and automobile shells and other highly reflective workpiece surfaces, pose challenges to accurate three-dimensional vision measurement. Currently, there are two main technologies for three-dimensional topography detection of highly reflective surfaces: contact detection and non-contact detection. As a commonly used contact detection method, the three-dimensional coordinate measuring machine has the disadvantages of complex operation, low efficiency, and easy damage to the surface of the measured part. The non-contact detection method mainly realizes visual measurement by spraying diffuse reflective materials on the highly reflective surface, but the thickness and uniformity of the spray powder will introduce errors to the measurement.

[0003] With the rapid development of image processing and computer technology, three-dimensional visual measurement of highly reflective surfaces has achieved widespread application in the field of three-dimensional reconstruction. In their invention patent number 201710689618.7, "A Three-Dimensional Scanning Measurement Device and Method for Highly Reflective Surfaces," Feng Wei et al. employed a DMD camera-based three-dimensional scanning measurement device. This device sequentially projects sinusoidal grating patterns of varying phases using a four-step phase shifting method. Next, the DMD camera is used to adaptively image the incident light with a high dynamic range, obtaining a clear image to be measured. Finally, a phase method based on grating projection is used to achieve three-dimensional measurement. In their invention patent number 201810739037.4, "Method and System for Structured Light Three-Dimensional Imaging of Highly Reflective Object Surfaces," Liu Kai et al. proposed a structured light three-dimensional imaging method for highly reflective surfaces. This method projects a binary phase-shifted coded pattern onto the surface of the object being measured. Based on the reflected image, multiple modulated image groups at different illumination intensities are generated. The image groups are arranged in descending order of illumination intensity, and the saturated regions are determined. Saturated and replacement pixels are obtained and their phases are calculated. The phases of the replacement pixels are then replaced with the phases of the saturated pixels to obtain a three-dimensional image. In their invention patent number 2020101822244, "Method for High-Precision Three-Dimensional Visual Measurement of Workpieces on Highly Reflective Surfaces," Ge Ji et al. proposed a method that includes setting a structured light projection brightness range and projecting a phase-shifted grating pattern. Using image fusion, an accurate three-dimensional point cloud image of the object being measured is reconstructed. While this method can obtain three-dimensional features of highly reflective surfaces, it suffers from complex system structures and experimental procedures.

[0004] The challenge for 3D measurement systems for highly reflective surfaces based on adaptive fringe projection lies in determining the relationship between the image grayscale and projection intensity in highly reflective areas, and accurately determining the adjustment areas and values ​​for the projected image. In their paper "3D Surface Measurement of Highly Reflective Objects Based on Adaptive Fringe Projection," Wang Liu et al. proposed a 3D measurement method based on adaptive fringe projection. This method projects a high-grayscale fringe pattern onto the surface of the object to be measured, identifying and marking overexposed points. After reducing the projection intensity, a nonlinear least-squares fitting method is used to determine the optimal maximum input grayscale for each saturated pixel. The regenerated adaptive fringe pattern is then used to acquire images for phase calculation and 3D topography recovery. However, this method's low grayscale values ​​are empirically derived and thus subject to certain errors. In their paper "Research on 3D Measurement Technology Using High Dynamic Range Grating Projection," Li Shaoxu proposed an adaptive grating projection method based on a bisection method. This method replaces the process of determining the camera-projector grayscale response curve with a bisection method to find the optimal grayscale value. However, this method is computationally intensive, slow, and complex, making it unsuitable for practical measurement needs.

[0005] It can be seen that the existing high-reflective surface measurement method based on adaptive fringe projection is inaccurate or complicated in obtaining low grayscale levels, inaccurate in adjusting projection pixels, and ignores the quantization error caused by different resolutions. The experimental process and the accuracy and completeness of three-dimensional reconstruction cannot meet the needs of engineering practice. Summary of the Invention

[0006] The purpose of the present invention is to provide an insulator mold cleaning method coordinated by multiple robotic arms, which solves the problems of insufficient detection accuracy of highly reflective three-dimensional complex curved surfaces and rough control of the cleaning process in the prior art.

[0007] The technical solution adopted by the present invention is a multi-manipulator collaborative insulator mold cleaning method, which is specifically implemented according to the following steps:

[0008] Step 1: Mold dust detection;

[0009] Step 2: Adaptive dust blowing;

[0010] Step 3: The dust-adhesive robotic arm cleans the mold;

[0011] Step 4: Evaluate the cleaning effect.

[0012] The present invention is also characterized in that:

[0013] In step 1, the insulator mold cleaning assembly is first arranged. The assembly includes the insulator mold to be cleaned, a purge robot arm installed directly in front of the insulator mold, and a high-pressure nozzle at the end of the purge robot arm. The high-pressure nozzle is driven by a pneumatic system. The pneumatic system is connected to the purge robot arm through a pipeline to provide high-pressure gas to the purge robot arm to perform the purge action. The purge robot arm is also equipped with a pressure sensor A; the pressure sensor A is connected to the control unit.

[0014] The system also includes a projection detection robot arm located above the side of the insulator mold, the projection detection robot arm integrating a fringe projector and a camera for detecting surface cleanliness; the projection detection robot arm is connected to the control unit;

[0015] The device also includes a dust-binding robot arm, which is connected to the hydraulic system and the control unit at the same time. The dust-binding robot arm is deployed on the other side of the insulator mold and is used to clean fine particles remaining after purging. The dust-binding robot arm is also equipped with a replaceable dust-binding paper tape. The head of the dust-binding robot arm is also provided with a silicone ejector rod, which presses down the dust-binding paper tape to remove residual dust on the mold surface. The dust-binding robot arm is also provided with a pressure sensor B; the pressure sensor B is in turn connected to the control unit;

[0016] The control unit is integrated into the equipment base or independent cabinet, with a built-in PLC or industrial computer to coordinate the robot arm movement and sensor feedback;

[0017] The hydraulic system is distributed at the joints of the robotic arm, providing high-precision motion driving force for the robotic arm.

[0018] Step 1 is implemented as follows:

[0019] Step 1.1. Obtain the three-dimensional coordinates of the geometric center of the center hole of the insulator mold through the camera integrated in the end effector of the projection inspection robot arm, and establish a mold coordinate system. In the mold coordinate system, plan the collision-free motion trajectories of the projection inspection robot arm, the purge robot arm, and the dust collection robot arm from their respective initial positions to the front of the mold. The end points of the trajectories are perpendicular to the mold surface and are at a preset distance of 100-300 mm from the mold surface. The planned path parameters are stored in the control unit. The projection inspection robot arm detects and calls the stored path parameters, drives the projection inspection robot arm to move and position itself in front of the insulator mold according to the preset trajectory, adjusts the projector and camera at the front end of the projection inspection robot arm to a distance of 100-300 mm from the mold surface, and completes the system calibration to ensure that the projection fringes are aligned with the coordinate system of the image capture.

[0020] Step 1.2: The projector at the front end of the projection detection robot projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method, and the absolute phase is obtained by decoding. Statistical filtering is used to remove outliers with twice the standard deviation as the threshold. The direction perpendicular to the geometric center surface of the mold center hole is the main direction. Based on the main direction, the phase map in the pixel coordinate system is mapped to the world coordinate system, and then point cloud data collection and matching are performed: First, the projection grayscale is uniform and Saturated image: Capture an image at a high exposure time, detect the saturated pixels in the image captured by the camera integrated in the end effector of the robot arm by marking the projection, and obtain the imaging range of the high-reflective area on the object surface; project an image with uniform and saturated grayscale as the above, reduce the exposure time, so that the imaging of the high-reflective area on the object surface is not saturated, re-capture the image under this exposure, combine the two exposure data to generate adaptive low-grayscale sinusoidal stripes, overexpose the high-reflective area, compare the height value point by point, set the height value greater than microns as the suspicious residue value, merge adjacent suspicious points into the defect area, generate a three-dimensional distribution map of the difference points before and after cleaning, and obtain a three-dimensional distribution map of the difference points;

[0021] Step 1.3: Complete difference point extraction and defect detection: First, construct the discrete difference point cloud into a triangular mesh to form a continuous surface model. For each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle. Set the height threshold h = 5μm and mark points with a height difference Δh>h as suspicious residue points. Integrate the S basis i×Δh by microelement accumulation addition. Assuming the contaminant density ρ = 1.2g / cm3, calculate the stain mass: m = V×ρ. Establish a local coordinate system with the geometric center of the mold center hole as the origin. Align the main direction of the defect area with the coordinate axis and collect the length, width, and depth data of the defect: record the maximum span along the main direction as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D. Map the three-dimensional coordinates and height values ​​of the defect area to a color gradient to generate an intuitive defect distribution map.

[0022] Step 1.4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified and proceed to the next step.

[0023] The parameters in step 1 include joint angle sequence, end pose coordinates and motion speed.

[0024] Step 2 is implemented as follows:

[0025] The purge robot arm is positioned at the mold position according to the precise mold surface model determined in step 1. A pressure sensor is selected and integrated into the front end of the nozzle to detect the gas pressure ejected from the nozzle in real time and provide high-pressure air during the purge process on the mold surface. The control unit is connected to the pressure sensor A through an electrical interface, and the corresponding sensor range, zero point calibration, filter parameters, sensor working mode, and accuracy setting parameters are configured on the control unit end to facilitate subsequent accurate conversion and analysis of the collected data. The pressure fluctuation rate δP / δt is calculated in real time based on the above data settings. If the pressure fluctuation rate δP / δt exceeds the set value, it is determined to be a pipeline blockage or leakage; and the air pressure system output is dynamically adjusted by the control unit to maintain the purge pressure within the set range of 0.3±0.02MPa; at the same time, the pressure-time curve is recorded, and the long-term stability is analyzed based on statistical process control (SPC), and a calibration reminder is automatically generated to achieve accurate monitoring and control of the mold purge process.

[0026] In step 2, the range is set according to the sensor specification input range, and a linear mapping formula between the electrical signal and the physical quantity is established: P = k·V+b, where P represents the physical pressure value, i.e., the actual purge pressure, in MPa; V represents the voltage signal output by the pressure sensor; k represents the calibration coefficient, which indicates the conversion sensitivity between voltage and pressure, in MPa / V; b represents the zero offset, which reflects the output error of the sensor at zero input, in MPa. The range setting is determined through data analysis. The sensor output value V0 is collected in the no-load state, and the zero offset compensation is calculated: P 实际 =k·(V 采集 -V0), P 实际 Indicates the actual pressure value after calibration, in MPa; V 采集It represents the voltage signal collected by the sensor in the no-load state, in V; V0 represents the zero offset voltage, that is, the output voltage of the sensor when there is no pressure, in V. Calculate the filter parameter α to calculate the cutoff frequency fc, P 滤波 =α·P 当前 +(1-α)·P 前次 , α=T+RCT, α represents the filter coefficient, 0<α<1; T represents the sampling time interval, the unit is s; R represents the filter resistance value, the unit is Ω; C represents the filter capacitance value, the unit is F; P 当前 Indicates the pressure value collected at the current moment, in MPa; P 前次 Indicates the pressure value after filtering at the last moment, in MPa; P 滤波 Indicates the pressure output value after filtering, in MPa; α=T / (T+τ), fc=1 / (2πτ). According to the above data, set the sensor working mode: set the sampling frequency fs based on fs>2fc; based on the range of P, the actual physical quantity P after zero point calibration 实际 Set the alarm threshold.

[0027] Step 3 is implemented as follows:

[0028] The mold holder is rotated to face the dust-sticking robot arm. The control unit marks the area with an Ra value greater than 0.4 in step 1 as the area to be dust-sticked. The dust-sticking robot arm activates the dual-roller system wrapped with sticky tape to rotate and release a section of sticky tape. The silicone ejector rod is controlled to approach the mold. Pressure sensor B controls the force with which the silicone ejector rod contacts the mold, thereby removing stubborn dust and stains through the sticky tape. After cleaning is complete, the mold holder is rotated to move the mold directly in front of the projection inspection robot arm.

[0029] Step 4 is implemented as follows:

[0030] The projection detection robot arm moves back to the geometric center of the mold center hole, and synchronously projects the coded stripes and collects the reflected image according to the process of step 1 to generate the three-dimensional point cloud and surface model after cleaning. The projection detection robot arm is equipped with a projector to project two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot arm is triggered to collect the deformed image, and the phase map is generated by the four-step phase shift method. The absolute phase is obtained by decoding. Statistical filtering is used to eliminate outliers with twice the standard deviation as the threshold. The phase map is aligned to the world coordinate system based on the main direction of the mold, and then point cloud data is collected and matched. The high-reflective area is always overexposed, and the height value is compared point by point. The height value is set to be greater than 5 microns as a suspicious residual value. The adjacent suspicious points are merged into the defect area to generate the difference point distribution before and after cleaning. 3D graph, after obtaining the 3D graph of difference point distribution, the difference point extraction and defect detection are completed: first, the difference point cloud area is triangulated, and then the defect area is calculated by multiplying the base area and the dust height. Finally, the calculated defect area is used to estimate the residual stain amount and output the defect position distribution map and the length, width and depth data of the defect. Finally, the quantitative standard calculation and evaluation are performed: the defect area is triangulated, and the microelement volume is calculated by the length, width and depth data and then summed. The density value is set to 1.2 g / cm3 based on the obtained volume. The residual mass of the dirt is calculated by the volume and area score. A continuous contour line is intercepted along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is taken out; when the Ra value is greater than 0.4, it is judged to be unqualified;

[0031] The test results are fed back to the control system in real time. If the standards are met, the mold cleaning work is completed. If the standards are not met, the secondary cleaning process is triggered and the cleaning path is optimized. At the same time, the projection brightness or camera exposure parameters are dynamically adjusted according to the changes in surface reflectivity to adapt to the detection requirements.

[0032] The beneficial effect of the present invention is that the insulator mold cleaning method coordinated by multiple robotic arms significantly improves the performance and reliability of the insulator mold cleaning system by integrating high-precision three-dimensional visual detection, multi-robotic arm intelligent collaboration and full-process closed-loop feedback technology. Specifically, the adaptive fringe projection and double exposure fusion algorithm are used to effectively suppress the overexposure phenomenon of highly reflective areas, reduce the three-dimensional reconstruction error of complex surfaces with a curvature radius of less than 50mm from 0.1mm to within 0.05mm, and reduce the missed detection rate of tiny particles (height>5μm) from 15% to below 3%, thereby stably controlling the surface roughness (Ra value) of the mold after cleaning to below 0.4μm. On this basis, the air pressure system based on pressure sensor and PID control realizes dynamic adjustment of the purge pressure (fluctuation rate <±0.02MPa), combines the force application technology of the dust-sticking robotic arm with force feedback control (contact force 0-5N adjustable), and uses statistical process control (SPC) to calibrate the cleaning parameters in real time, so that the fluctuation amplitude of the residual stain quality is reduced from 0.1g / m 2 Reduced to 0.02g / m 2 , achieving full batch consistency in mold surface cleanliness (qualified rate > 99.5%). Furthermore, a phase method is used to generate a three-dimensional difference point cloud map, quantifying the volume and distribution characteristics of stains in real time. Combined with an adaptive surface reflectivity adjustment mechanism (brightness dynamic range 100-64 gray levels), a "measurement-evaluation-feedback" closed-loop system is constructed, reducing the risk of repeated contamination to less than 10%, shortening the defect detection response time to less than 5 seconds, and supporting a production line cycle of 60 seconds per piece, improving cleaning efficiency by 80% compared to traditional methods. Ultimately, relying on a multi-robot collaborative path planning algorithm to avoid motion conflicts and redundant operations, the system integration level is increased by 30%, the complexity of the experimental process is reduced by 50%, and the cloud database (which has included 12,000 sets of defect samples) drives the AI ​​model to dynamically optimize the cleaning path, reducing the mold change time from 15 minutes to 8 seconds, and increasing the comprehensive equipment utilization rate to 92%, fully meeting the industrial cleaning needs of efficient, stable, and easy-to-maintain scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a schematic diagram of the main structure of multi-robotic arm collaboration.

[0034] In the figure, 1. Projection detection robot arm, 2. Purge robot arm, 3. Dust-sticking robot arm, 4. Pressure sensor A, 5. Nozzle, 6. Air pressure system, 7. Silicone ejector rod, 8. Dust-sticking paper tape, 9. Hydraulic system, 10. Control unit, 11. Mold, 12. Mold center hole, 13. Mold bracket, 14. Pressure sensor B. DETAILED DESCRIPTION

[0035] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] The insulator mold cleaning method using multiple robotic arms in the present invention is specifically implemented according to the following steps:

[0037] Step 1: Mold dust detection;

[0038] In step 1, the insulator mold cleaning assembly is first arranged, which includes the insulator mold to be cleaned, Figure 1 The center position may contain dust or particles on the surface. The purge robot arm 2 is installed in front of the insulator mold. The end of the purge robot arm 2 is equipped with a high-pressure nozzle 5. The high-pressure nozzle 5 is driven by the air pressure system 6. The air pressure system is connected to the purge robot arm 2 through a pipeline to provide high-pressure gas to the purge robot arm 2 to perform the purge action. The purge robot arm 2 is also equipped with a pressure sensor A4; the pressure sensor A4 is connected to the control unit.

[0039] The system also includes a projection detection robot arm 1 located above the side of the insulator mold. The projection detection robot arm 1 integrates a fringe projector and a camera for detecting surface cleanliness. The projection detection robot arm 1 is connected to the control unit.

[0040] The device further includes a dust-binding robot arm 3, which is connected to both the hydraulic system and the control unit. The dust-binding robot arm 3 is deployed on the other side of the insulator mold and is used to clean fine particles remaining after purging. The dust-binding robot arm 3 is also equipped with a replaceable dust-binding paper tape 8. The head of the dust-binding robot arm 3 is also provided with a silicone ejector 7, which presses down the dust-binding paper tape 8 to remove residual dust on the mold surface. The dust-binding robot arm 3 is also provided with a pressure sensor B14; the pressure sensor B14 is in turn connected to the control unit.

[0041] The control unit is integrated into the equipment base or independent cabinet, with a built-in PLC or industrial computer to coordinate the robot arm movement and sensor feedback;

[0042] The hydraulic system is distributed at the joints of the robotic arm, providing high-precision motion driving force (such as rotation and extension) for the robotic arm.

[0043] Step 1 is implemented as follows:

[0044] Step 1.1. Obtain the three-dimensional coordinates of the geometric center of the center hole 12 of the insulator mold through the camera integrated in the end effector of the projection detection robot arm 1, and establish a mold coordinate system. In the mold coordinate system, plan the collision-free motion trajectories of the projection detection robot arm 1, the purge robot arm 2, and the dust-binding robot arm 3 from their respective initial positions to the front of the mold. The end points of the trajectories are perpendicular to the mold surface and are at a preset distance of 100-300 mm from the mold surface. The planned path parameters are stored in the control unit. The projection detection robot arm 1 is detected and calls the stored path parameters. The projection detection robot arm 1 is driven to move and position itself in front of the insulator mold according to the preset trajectory. The projector and camera at the front end of the projection detection robot arm 1 are adjusted to a distance of 100-300 mm from the mold surface. The system calibration is completed to ensure that the projection fringes are aligned with the coordinate system of the image capture.

[0045] Step 1.2: The projector at the front end of the projection detection robot 1 projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot 1 is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method, and the absolute phase is obtained by decoding. Statistical filtering is used to remove outliers with twice the standard deviation as the threshold. The direction perpendicular to the geometric center surface of the mold center hole 12 is the main direction. Based on the main direction, the phase map in the pixel coordinate system is mapped to the world coordinate system, and then the point cloud data is collected and matched: First, an image with uniform and saturated grayscale is projected. At high exposure The image is collected under a light time (50ms), and the saturated pixels in the image collected by the camera integrated in the end effector of the detection robot 1 are marked and projected to obtain the imaging range of the high-reflective area on the object surface; an image with uniform and saturated grayscale as the above is projected, and the exposure time is reduced (10ms) so that the imaging of the high-reflective area on the object surface is not saturated, and the image under the exposure is re-collected. The adaptive low-grayscale sinusoidal stripes (maximum grayscale 64) are generated by combining the two exposure data, and the high-reflective area is always overexposed. The height value is compared point by point, and a height value greater than 5 microns is set as a suspicious residual value. Adjacent suspicious points are merged into a defect area, and a three-dimensional distribution map of the difference points before and after cleaning is generated to obtain a three-dimensional distribution map of the difference points;

[0046] Step 1.3, complete the difference point extraction and defect detection: first construct the discrete difference point cloud into a triangular mesh to form a continuous surface model; for each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle; set the height threshold h=5μm, and mark the points with height difference Δh>h as suspicious residual points; integrate the S basis i×Δh by micro-element accumulation addition, assuming the contaminant density ρ=1.2g / cm3, and calculate the contaminant density. Stain mass: m = V × ρ; establish a local coordinate system with the geometric center of the mold center hole 12 as the origin, align the main direction of the defect area (such as the longest axis) with the coordinate axis, and collect the length, width, and depth data of the defect: record the maximum span along the main direction (Euclidean distance between two points) as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D. Map the three-dimensional coordinates and height values ​​of the defect area to a color gradient (such as red indicates a large depth and blue indicates a shallow depth) to generate an intuitive defect distribution map;

[0047] Step 1.4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line (set the length to 5 mm) along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified, and proceed to the next step.

[0048] The parameters in step 1 include joint angle sequence, end pose coordinates and motion speed.

[0049] Step 2: Adaptive dust blowing;

[0050] Step 2 is implemented as follows:

[0051] The purge robot arm 2 is positioned at the mold position according to the precise mold surface model determined in step 1. A high-precision pressure sensor A4 is selected and integrated at the front end of the nozzle 5 to detect the gas pressure ejected from the nozzle 5 in real time and provide high-pressure air during the purge process on the mold surface. The control unit is connected to the pressure sensor A4 through an electrical interface, and the corresponding sensor range, zero point calibration, filter parameters, sensor working mode, and accuracy setting parameters are configured on the control unit end to facilitate subsequent accurate conversion and analysis of the collected data. The pressure fluctuation rate δP / δt is calculated in real time based on the above data settings. If the pressure fluctuation rate δP / δt exceeds the set value (0.5MPa), it is determined to be a pipeline blockage or leakage; and the air pressure system output is dynamically adjusted by the control unit to maintain the purge pressure within the set range of 0.3±0.02MPa. At the same time, the pressure-time curve is recorded, and the long-term stability is analyzed based on statistical process control (SPC). A calibration reminder is automatically generated to achieve accurate monitoring and control of the mold purge process.

[0052] In step 2, the range is set according to the sensor specification input range, and a linear mapping formula between the electrical signal and the physical quantity is established: P = k·V+b, where P represents the physical pressure value, i.e., the actual purge pressure, in MPa; V represents the voltage signal output by the pressure sensor; k represents the calibration coefficient, which indicates the conversion sensitivity between voltage and pressure, in MPa / V; b represents the zero offset, which reflects the output error of the sensor at zero input, in MPa. The range setting is determined through data analysis. The sensor output value V0 is collected in the no-load state, and the zero offset compensation is calculated: P 实际 =k·(V 采集 -V0), P 实际 Indicates the actual pressure value after calibration, in MPa; V 采集 It represents the voltage signal collected by the sensor in the no-load state, in V; V0 represents the zero offset voltage, that is, the output voltage of the sensor when there is no pressure, in V. Calculate the filter parameter α to calculate the cutoff frequency fc, P 滤波 =α·P 当前 +(1-α)·P 前次 , α=T+RCT, α represents the filter coefficient, 0<α<1; T represents the sampling time interval, the unit is s; R represents the filter resistance value, the unit is Ω; C represents the filter capacitance value, the unit is F; P 当前 Indicates the pressure value collected at the current moment, in MPa; P 前次 Indicates the pressure value after filtering at the last moment, in MPa; P 滤波Indicates the pressure output value after filtering, in MPa; α=T / (T+τ), fc=1 / (2πτ). According to the above data, set the sensor working mode: set the sampling frequency fs based on fs>2fc; based on the range of P, the actual physical quantity P after zero point calibration 实际 Set the alarm threshold.

[0053] Step 3: The dust-adhesive robotic arm cleans the mold;

[0054] Step 3 is implemented as follows:

[0055] Rotate the mold holder to face the insulator mold directly toward the dust sticking robot arm 3. The control unit marks the area with an Ra value greater than 0.4 in step 1 as the area to be dust-sticked. (Since the dust sticking robot arm 3 has a dual-roller system at the end, and a sticky paper tape 8 is wrapped around the dual-roller system, the sticky paper tape 8 is wound around the dual-roller system.) The sticky robot arm 3 activates the dual-roller system with the sticky paper tape 8 to rotate and release a section of the sticky paper tape. It also controls the silicone ejector rod 7 to slowly approach the mold. The pressure sensor B14 controls the force with which the silicone ejector rod 7 contacts the mold, thereby removing stubborn dust and stains through the sticky paper tape 8. After cleaning is complete, rotate the mold holder to move the mold directly in front of the projection inspection robot arm.

[0056] Step 4: Evaluate the cleaning effect.

[0057] Step 4 is implemented as follows:

[0058] The projection detection robot arm 1 moves back to the geometric center of the mold center hole, and synchronously projects the coded stripes and collects the reflected image according to the process of step 1 to generate the three-dimensional point cloud and surface model after cleaning. The projection detection robot arm 1 is equipped with a projector to project two groups of standard sinusoidal stripes onto the cleaned mold surface. The two groups of standard sinusoidal stripes include a group of horizontal standard sinusoidal stripes and a group of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot arm 1 is triggered to collect the deformed image, and a phase map is generated by a four-step phase shift method. The absolute phase is obtained by decoding, and statistical filtering is used to eliminate outliers with twice the standard deviation as the threshold. The phase map is aligned to the world coordinate system based on the main direction of the mold, and then point cloud data is collected and matched. The high-reflective area is always overexposed, and the height value is compared point by point. The height value is set to be greater than 5 microns as a suspicious residual value. The adjacent suspicious points are merged into a defective area to generate a three-dimensional distribution map of the difference points before and after cleaning. After obtaining the three-dimensional distribution map of the difference points, the difference point extraction and defect detection are completed: first, the difference point cloud area is triangulated, and then the defect area is calculated by multiplying the base area and the dust height. Finally, the calculated defect area is used to estimate the residual stain amount and output the defect position distribution map and the length, width and depth data of the defect. Finally, the quantitative standard calculation and evaluation are performed: the defect area is triangulated, and the microelement volume is calculated by the length, width and depth data and then summed. The density value is set to 1.2 g / cm3 based on the obtained volume. The residual mass of the dirt is calculated by the volume and area score. A continuous contour line is intercepted along the normal direction of the mold surface from the defect-free area (the length is set to 5 mm), thereby obtaining a discrete height value sequence. The average value of the sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is taken out; when the Ra value is greater than 0.4, it is judged to be unqualified;

[0059] The test results are fed back to the control system in real time. If the standards are met, the mold cleaning work is completed. If the standards are not met, the secondary cleaning process is triggered and the cleaning path is optimized. At the same time, the projection brightness or camera exposure parameters are dynamically adjusted according to the changes in surface reflectivity to adapt to the detection requirements.

[0060] Example 1

[0061] The insulator mold cleaning method using multiple robotic arms of the present invention is specifically implemented according to the following steps:

[0062] Step 1: Mold dust detection;

[0063] Step 2: Adaptive dust blowing;

[0064] Step 3: The dust-adhesive robotic arm cleans the mold;

[0065] Step 4: Evaluate the cleaning effect.

[0066] Example 2

[0067] The insulator mold cleaning method using multiple robotic arms in the present invention is specifically implemented according to the following steps:

[0068] Step 1: Mold dust detection;

[0069] In step 1, the insulator mold cleaning assembly is first arranged, which includes the insulator mold to be cleaned, Figure 1 The center position may contain dust or particles on the surface. The purge robot arm 2 is installed in front of the insulator mold. The end of the purge robot arm 2 is equipped with a high-pressure nozzle 5. The high-pressure nozzle 5 is driven by the air pressure system 6. The air pressure system is connected to the purge robot arm 2 through a pipeline to provide high-pressure gas to the purge robot arm 2 to perform the purge action. The purge robot arm 2 is also equipped with a pressure sensor A4; the pressure sensor A4 is connected to the control unit.

[0070] The system also includes a projection detection robot arm 1 located above the side of the insulator mold. The projection detection robot arm 1 integrates a fringe projector and a camera for detecting surface cleanliness. The projection detection robot arm 1 is connected to the control unit.

[0071] The device further includes a dust-binding robot arm 3, which is connected to both the hydraulic system and the control unit. The dust-binding robot arm 3 is deployed on the other side of the insulator mold and is used to clean fine particles remaining after purging. The dust-binding robot arm 3 is also equipped with a replaceable dust-binding paper tape 8. The head of the dust-binding robot arm 3 is also provided with a silicone ejector 7, which presses down the dust-binding paper tape 8 to remove residual dust on the mold surface. The dust-binding robot arm 3 is also provided with a pressure sensor B14; the pressure sensor B14 is in turn connected to the control unit.

[0072] The control unit is integrated into the equipment base or independent cabinet, with a built-in PLC or industrial computer to coordinate the robot arm movement and sensor feedback;

[0073] The hydraulic system is distributed at the joints of the robotic arm, providing high-precision motion driving force (such as rotation and extension) for the robotic arm.

[0074] Step 2: Adaptive dust blowing;

[0075] Step 3: The dust-adhesive robotic arm cleans the mold;

[0076] Step 4: Evaluate the cleaning effect.

[0077] Example 3

[0078] The insulator mold cleaning method using multiple robotic arms in the present invention is specifically implemented according to the following steps:

[0079] Step 1: Mold dust detection;

[0080] In step 1, the insulator mold cleaning assembly is first arranged. The assembly includes the insulator mold to be cleaned, Figure 1 The center position may contain dust or particles on the surface. The purge robot arm 2 is installed in front of the insulator mold. The end of the purge robot arm 2 is equipped with a high-pressure nozzle 5. The high-pressure nozzle 5 is driven by the air pressure system 6. The air pressure system is connected to the purge robot arm 2 through a pipeline to provide high-pressure gas to the purge robot arm 2 to perform the purge action. The purge robot arm 2 is also equipped with a pressure sensor A4; the pressure sensor A4 is connected to the control unit.

[0081] The system also includes a projection detection robot arm 1 located above the side of the insulator mold. The projection detection robot arm 1 integrates a fringe projector and a camera for detecting surface cleanliness. The projection detection robot arm 1 is connected to the control unit.

[0082] The device further includes a dust-binding robot arm 3, which is connected to both the hydraulic system and the control unit. The dust-binding robot arm 3 is deployed on the other side of the insulator mold and is used to clean fine particles remaining after purging. The dust-binding robot arm 3 is also equipped with a replaceable dust-binding paper tape 8. The head of the dust-binding robot arm 3 is also provided with a silicone ejector 7, which presses down the dust-binding paper tape 8 to remove residual dust on the mold surface. The dust-binding robot arm 3 is also provided with a pressure sensor B14; the pressure sensor B14 is in turn connected to the control unit.

[0083] The control unit is integrated into the equipment base or independent cabinet, with a built-in PLC or industrial computer to coordinate the robot arm movement and sensor feedback;

[0084] The hydraulic system is distributed at the joints of the robotic arm, providing high-precision motion driving force (such as rotation and extension) for the robotic arm.

[0085] Step 1 is implemented as follows:

[0086] Step 1.1. Obtain the three-dimensional coordinates of the geometric center of the center hole 12 of the insulator mold through the camera integrated in the end effector of the projection detection robot arm 1, and establish a mold coordinate system. In the mold coordinate system, plan the collision-free motion trajectories of the projection detection robot arm 1, the purge robot arm 2, and the dust-binding robot arm 3 from their respective initial positions to the front of the mold. The end points of the trajectories are perpendicular to the mold surface and are at a preset distance of 100-300 mm from the mold surface. The planned path parameters are stored in the control unit. The projection detection robot arm 1 is detected and calls the stored path parameters. The projection detection robot arm 1 is driven to move and position itself in front of the insulator mold according to the preset trajectory. The projector and camera at the front end of the projection detection robot arm 1 are adjusted to a distance of 100-300 mm from the mold surface. The system calibration is completed to ensure that the projection fringes are aligned with the coordinate system of the image capture.

[0087] Step 1.2: The projector at the front end of the projection detection robot 1 projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot 1 is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method, and the absolute phase is obtained by decoding. Statistical filtering is used to remove outliers with twice the standard deviation as the threshold. The direction perpendicular to the geometric center surface of the mold center hole 12 is the main direction. Based on the main direction, the phase map in the pixel coordinate system is mapped to the world coordinate system, and then the point cloud data is collected and matched: First, an image with uniform and saturated grayscale is projected. At high exposure The image is collected under a light time (50ms), and the saturated pixels in the image collected by the camera integrated in the end effector of the detection robot 1 are marked and projected to obtain the imaging range of the high-reflective area on the object surface; an image with uniform and saturated grayscale as the above is projected, and the exposure time is reduced (10ms) so that the imaging of the high-reflective area on the object surface is not saturated, and the image under the exposure is re-collected. The adaptive low-grayscale sinusoidal stripes (maximum grayscale 64) are generated by combining the two exposure data, and the high-reflective area is always overexposed. The height value is compared point by point, and a height value greater than 5 microns is set as a suspicious residual value. Adjacent suspicious points are merged into a defect area, and a three-dimensional distribution map of the difference points before and after cleaning is generated to obtain a three-dimensional distribution map of the difference points;

[0088] Step 1.3, complete the difference point extraction and defect detection: first construct the discrete difference point cloud into a triangular mesh to form a continuous surface model; for each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle; set the height threshold h=5μm, and mark the points with height difference Δh>h as suspicious residual points; integrate the S basis i×Δh by micro-element accumulation addition, assuming the contaminant density ρ=1.2g / cm3, and calculate the contaminant density. Stain mass: m = V × ρ; establish a local coordinate system with the geometric center of the mold center hole 12 as the origin, align the main direction of the defect area (such as the longest axis) with the coordinate axis, and collect the length, width, and depth data of the defect: record the maximum span along the main direction (Euclidean distance between two points) as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D. Map the three-dimensional coordinates and height values ​​of the defect area to a color gradient (such as red indicates a large depth and blue indicates a shallow depth) to generate an intuitive defect distribution map;

[0089] Step 1.4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line (set the length to 5 mm) along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified, and proceed to the next step.

[0090] Step 2: Adaptive dust blowing;

[0091] Step 3: The dust-adhesive robotic arm cleans the mold;

[0092] Step 4: Evaluate the cleaning effect.

[0093] Example 4

[0094] The insulator mold cleaning method using multiple robotic arms in the present invention is specifically implemented according to the following steps:

[0095] Step 1: Mold dust detection;

[0096] Step 1 is implemented as follows:

[0097] Step 1.1. Obtain the three-dimensional coordinates of the geometric center of the center hole 12 of the insulator mold through the camera integrated in the end effector of the projection detection robot arm 1, and establish a mold coordinate system. In the mold coordinate system, plan the collision-free motion trajectories of the projection detection robot arm 1, the purge robot arm 2, and the dust-binding robot arm 3 from their respective initial positions to the front of the mold. The end points of the trajectories are perpendicular to the mold surface and are at a preset distance of 100-300 mm from the mold surface. The planned path parameters are stored in the control unit. The projection detection robot arm 1 is detected and calls the stored path parameters. The projection detection robot arm 1 is driven to move and position itself in front of the insulator mold according to the preset trajectory. The projector and camera at the front end of the projection detection robot arm 1 are adjusted to a distance of 100-300 mm from the mold surface. The system calibration is completed to ensure that the projection fringes are aligned with the coordinate system of the image capture.

[0098] Step 1.2: The projector at the front end of the projection detection robot 1 projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot 1 is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method, and the absolute phase is obtained by decoding. Statistical filtering is used to remove outliers with twice the standard deviation as the threshold. The direction perpendicular to the geometric center surface of the mold center hole 12 is the main direction. Based on the main direction, the phase map in the pixel coordinate system is mapped to the world coordinate system, and then the point cloud data is collected and matched: First, an image with uniform and saturated grayscale is projected. At high exposure The image is collected under a light time (50ms), and the saturated pixels in the image collected by the camera integrated in the end effector of the detection robot 1 are marked and projected to obtain the imaging range of the high-reflective area on the object surface; an image with uniform and saturated grayscale as the above is projected, and the exposure time is reduced (10ms) so that the imaging of the high-reflective area on the object surface is not saturated, and the image under the exposure is re-collected. The adaptive low-grayscale sinusoidal stripes (maximum grayscale 64) are generated by combining the two exposure data, and the high-reflective area is always overexposed. The height value is compared point by point, and a height value greater than 5 microns is set as a suspicious residual value. Adjacent suspicious points are merged into a defect area, and a three-dimensional distribution map of the difference points before and after cleaning is generated to obtain a three-dimensional distribution map of the difference points;

[0099] Step 1.3, complete the difference point extraction and defect detection: first construct the discrete difference point cloud into a triangular mesh to form a continuous surface model; for each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle; set the height threshold h=5μm, and mark the points with height difference Δh>h as suspicious residual points; integrate the S basis i×Δh by micro-element accumulation addition, assuming the contaminant density ρ=1.2g / cm3, and calculate the contaminant density. Stain mass: m = V × ρ; establish a local coordinate system with the geometric center of the mold center hole 12 as the origin, align the main direction of the defect area (such as the longest axis) with the coordinate axis, and collect the length, width, and depth data of the defect: record the maximum span along the main direction (Euclidean distance between two points) as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D. Map the three-dimensional coordinates and height values ​​of the defect area to a color gradient (such as red indicates a large depth and blue indicates a shallow depth) to generate an intuitive defect distribution map;

[0100] Step 1.4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line (set the length to 5 mm) along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified, and proceed to the next step.

[0101] Step 2: Adaptive dust blowing;

[0102] Step 3: The dust-adhesive robotic arm cleans the mold;

[0103] Step 4: Evaluate the cleaning effect.

[0104] Example 5

[0105] The insulator mold cleaning method using multiple robotic arms in the present invention is specifically implemented according to the following steps:

[0106] Step 1: Mold dust detection;

[0107] In step 1, the insulator mold cleaning assembly is first arranged, which includes the insulator mold to be cleaned, Figure 1 The center position may contain dust or particles on the surface. The purge robot arm 2 is installed in front of the insulator mold. The end of the purge robot arm 2 is equipped with a high-pressure nozzle 5. The high-pressure nozzle 5 is driven by the air pressure system 6. The air pressure system is connected to the purge robot arm 2 through a pipeline to provide high-pressure gas to the purge robot arm 2 to perform the purge action. The purge robot arm 2 is also equipped with a pressure sensor A4; the pressure sensor A4 is connected to the control unit.

[0108] The system also includes a projection detection robot arm 1 located above the side of the insulator mold. The projection detection robot arm 1 integrates a fringe projector and a camera for detecting surface cleanliness. The projection detection robot arm 1 is connected to the control unit.

[0109] The device further includes a dust-binding robot arm 3, which is connected to both the hydraulic system and the control unit. The dust-binding robot arm 3 is deployed on the other side of the insulator mold and is used to clean fine particles remaining after purging. The dust-binding robot arm 3 is also equipped with a replaceable dust-binding paper tape 8. The head of the dust-binding robot arm 3 is also provided with a silicone ejector 7, which presses down the dust-binding paper tape 8 to remove residual dust on the mold surface. The dust-binding robot arm 3 is also provided with a pressure sensor B14; the pressure sensor B14 is in turn connected to the control unit.

[0110] The control unit is integrated into the equipment base or independent cabinet, with a built-in PLC or industrial computer to coordinate the robot arm movement and sensor feedback;

[0111] The hydraulic system is distributed at the joints of the robotic arm, providing high-precision motion driving force (such as rotation and extension) for the robotic arm.

[0112] Step 1 is implemented as follows:

[0113] Step 1.1. Obtain the three-dimensional coordinates of the geometric center of the center hole 12 of the insulator mold through the camera integrated in the end effector of the projection detection robot arm 1, and establish a mold coordinate system. In the mold coordinate system, plan the collision-free motion trajectories of the projection detection robot arm 1, the purge robot arm 2, and the dust-binding robot arm 3 from their respective initial positions to the front of the mold. The end points of the trajectories are perpendicular to the mold surface and are at a preset distance of 100-300 mm from the mold surface. The planned path parameters are stored in the control unit. The projection detection robot arm 1 is detected and calls the stored path parameters. The projection detection robot arm 1 is driven to move and position itself in front of the insulator mold according to the preset trajectory. The projector and camera at the front end of the projection detection robot arm 1 are adjusted to a distance of 100-300 mm from the mold surface. The system calibration is completed to ensure that the projection fringes are aligned with the coordinate system of the image capture.

[0114] Step 1.2: The projector at the front end of the projection detection robot 1 projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot 1 is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method, and the absolute phase is obtained by decoding. Statistical filtering is used to remove outliers with twice the standard deviation as the threshold. The direction perpendicular to the geometric center surface of the mold center hole 12 is the main direction. Based on the main direction, the phase map in the pixel coordinate system is mapped to the world coordinate system, and then the point cloud data is collected and matched: First, an image with uniform and saturated grayscale is projected. At high exposure The image is collected under a light time (50ms), and the saturated pixels in the image collected by the camera integrated in the end effector of the detection robot 1 are marked and projected to obtain the imaging range of the high-reflective area on the object surface; an image with uniform and saturated grayscale as the above is projected, and the exposure time is reduced (10ms) so that the imaging of the high-reflective area on the object surface is not saturated, and the image under the exposure is re-collected. The adaptive low-grayscale sinusoidal stripes (maximum grayscale 64) are generated by combining the two exposure data, and the high-reflective area is always overexposed. The height value is compared point by point, and a height value greater than 5 microns is set as a suspicious residual value. Adjacent suspicious points are merged into a defect area, and a three-dimensional distribution map of the difference points before and after cleaning is generated to obtain a three-dimensional distribution map of the difference points;

[0115] Step 1.3, complete the difference point extraction and defect detection: first construct the discrete difference point cloud into a triangular mesh to form a continuous surface model; for each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle; set the height threshold h=5μm, and mark the points with height difference Δh>h as suspicious residual points; integrate the S basis i×Δh by micro-element accumulation addition, assuming the contaminant density ρ=1.2g / cm3, and calculate the contaminant density. Stain mass: m = V × ρ; establish a local coordinate system with the geometric center of the mold center hole 12 as the origin, align the main direction of the defect area (such as the longest axis) with the coordinate axis, and collect the length, width, and depth data of the defect: record the maximum span along the main direction (Euclidean distance between two points) as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D. Map the three-dimensional coordinates and height values ​​of the defect area to a color gradient (such as red indicates a large depth and blue indicates a shallow depth) to generate an intuitive defect distribution map;

[0116] Step 1.4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line (set the length to 5 mm) along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified, and proceed to the next step.

[0117] The parameters in step 1 include joint angle sequence, end pose coordinates and motion speed.

[0118] Step 2: Adaptive dust blowing;

[0119] Step 2 is implemented as follows:

[0120] The purge robot arm 2 is positioned at the mold position according to the precise mold surface model determined in step 1. A high-precision pressure sensor A4 is selected and integrated at the front end of the nozzle 5 to detect the gas pressure ejected from the nozzle 5 in real time and provide high-pressure air during the purge process on the mold surface. The control unit is connected to the pressure sensor 4 through an electrical interface, and the corresponding sensor range, zero point calibration, filter parameters, sensor working mode, and accuracy setting parameters are configured on the control unit end to facilitate subsequent accurate conversion and analysis of the collected data. The pressure fluctuation rate δP / δt is calculated in real time based on the above data settings. If the pressure fluctuation rate δP / δt exceeds the set value (0.5MPa), it is determined to be a pipeline blockage or leakage; and the air pressure system output is dynamically adjusted by the control unit to maintain the purge pressure within the set range of 0.3±0.02MPa; at the same time, the pressure-time curve is recorded, and the long-term stability is analyzed based on statistical process control (SPC), and a calibration reminder is automatically generated to achieve accurate monitoring and control of the mold purge process.

[0121] In step 2, the range is set according to the sensor specification input range, and a linear mapping formula between the electrical signal and the physical quantity is established: P = k·V+b, where P represents the physical pressure value, i.e., the actual purge pressure, in MPa; V represents the voltage signal output by the pressure sensor; k represents the calibration coefficient, which indicates the conversion sensitivity between voltage and pressure, in MPa / V; b represents the zero offset, which reflects the output error of the sensor at zero input, in MPa. The range setting is determined through data analysis. The sensor output value V0 is collected in the no-load state, and the zero offset compensation is calculated: P 实际 =k·(V 采集 -V0), P 实际 Indicates the actual pressure value after calibration, in MPa; V 采集 It represents the voltage signal collected by the sensor in the no-load state, in V; V0 represents the zero offset voltage, that is, the output voltage of the sensor when there is no pressure, in V. Calculate the filter parameter α to calculate the cutoff frequency fc, P 滤波 =α·P 当前 +(1-α)·P 前次 , α=T+RCT, α represents the filter coefficient, 0<α<1; T represents the sampling time interval, the unit is s; R represents the filter resistance value, the unit is Ω; C represents the filter capacitance value, the unit is F; P 当前 Indicates the pressure value collected at the current moment, in MPa; P 前次 Indicates the pressure value after filtering at the last moment, in MPa; P 滤波Indicates the pressure output value after filtering, in MPa; α=T / (T+τ), fc=1 / (2πτ). According to the above data, set the sensor working mode: set the sampling frequency fs based on fs>2fc; based on the range of P, the actual physical quantity P after zero point calibration 实际 Set the alarm threshold.

[0122] Step 3: The dust-adhesive robotic arm cleans the mold;

[0123] Step 4: Evaluate the cleaning effect.

[0124] Example 6

[0125] The insulator mold cleaning method using multiple robotic arms in the present invention is specifically implemented according to the following steps:

[0126] Step 1 is implemented as follows:

[0127] Step 1.1. Obtain the three-dimensional coordinates of the geometric center of the center hole 12 of the insulator mold through the camera integrated in the end effector of the projection detection robot arm 1, and establish a mold coordinate system. In the mold coordinate system, plan the collision-free motion trajectories of the projection detection robot arm 1, the purge robot arm 2, and the dust-binding robot arm 3 from their respective initial positions to the front of the mold. The end points of the trajectories are perpendicular to the mold surface and are at a preset distance of 100-300 mm from the mold surface. The planned path parameters are stored in the control unit. The projection detection robot arm 1 is detected and calls the stored path parameters. The projection detection robot arm 1 is driven to move and position itself in front of the insulator mold according to the preset trajectory. The projector and camera at the front end of the projection detection robot arm 1 are adjusted to a distance of 100-300 mm from the mold surface. The system calibration is completed to ensure that the projection fringes are aligned with the coordinate system of the image capture.

[0128] Step 1.2: The projector at the front end of the projection detection robot 1 projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot 1 is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method, and the absolute phase is obtained by decoding. Statistical filtering is used to remove outliers with twice the standard deviation as the threshold. The direction perpendicular to the geometric center surface of the mold center hole 12 is the main direction. Based on the main direction, the phase map in the pixel coordinate system is mapped to the world coordinate system, and then the point cloud data is collected and matched: First, an image with uniform and saturated grayscale is projected. At high exposure The image is collected under a light time (50ms), and the saturated pixels in the image collected by the camera integrated in the end effector of the detection robot 1 are marked and projected to obtain the imaging range of the high-reflective area on the object surface; an image with uniform and saturated grayscale as the above is projected, and the exposure time is reduced (10ms) so that the imaging of the high-reflective area on the object surface is not saturated, and the image under the exposure is re-collected. The adaptive low-grayscale sinusoidal stripes (maximum grayscale 64) are generated by combining the two exposure data, and the high-reflective area is always overexposed. The height value is compared point by point, and a height value greater than 5 microns is set as a suspicious residual value. Adjacent suspicious points are merged into a defect area, and a three-dimensional distribution map of the difference points before and after cleaning is generated to obtain a three-dimensional distribution map of the difference points;

[0129] Step 1.3, complete the difference point extraction and defect detection: first construct the discrete difference point cloud into a triangular mesh to form a continuous surface model; for each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle; set the height threshold h=5μm, and mark the points with height difference Δh>h as suspicious residual points; integrate the S basis i×Δh by micro-element accumulation addition, assuming the contaminant density ρ=1.2g / cm3, and calculate the contaminant density. Stain mass: m = V × ρ; establish a local coordinate system with the geometric center of the mold center hole 12 as the origin, align the main direction of the defect area (such as the longest axis) with the coordinate axis, and collect the length, width, and depth data of the defect: record the maximum span along the main direction (Euclidean distance between two points) as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D. Map the three-dimensional coordinates and height values ​​of the defect area to a color gradient (such as red indicates a large depth and blue indicates a shallow depth) to generate an intuitive defect distribution map;

[0130] Step 1.4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line (set the length to 5 mm) along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified, and proceed to the next step.

[0131] The parameters in step 1 include joint angle sequence, end pose coordinates and motion speed.

[0132] Step 2: Adaptive dust blowing;

[0133] Step 2 is implemented as follows:

[0134] The purge robot 2 is positioned at the mold position according to the precise mold surface model determined in step 1. A high-precision pressure sensor 4 is selected and integrated at the front end of the nozzle 5 to detect the gas pressure ejected from the nozzle 5 in real time and provide high-pressure air during the purge process on the mold surface. The control unit is connected to the pressure sensor 4 through an electrical interface, and the corresponding sensor range, zero point calibration, filter parameters, sensor working mode, and accuracy setting parameters are configured on the control unit end to facilitate subsequent accurate conversion and analysis of the collected data. The pressure fluctuation rate δP / δt is calculated in real time based on the above data settings. If the pressure fluctuation rate δP / δt exceeds the set value (0.5MPa), it is determined to be a pipeline blockage or leakage; and the air pressure system output is dynamically adjusted by the control unit to maintain the purge pressure within the set range of 0.3±0.02MPa; at the same time, the pressure-time curve is recorded, and the long-term stability is analyzed based on statistical process control (SPC), and a calibration reminder is automatically generated to achieve accurate monitoring and control of the mold purge process.

[0135] In step 2, the range is set according to the sensor specification input range, and a linear mapping formula between the electrical signal and the physical quantity is established: P = k·V+b, where P represents the physical pressure value, i.e., the actual purge pressure, in MPa; V represents the voltage signal output by the pressure sensor; k represents the calibration coefficient, which indicates the conversion sensitivity between voltage and pressure, in MPa / V; b represents the zero offset, which reflects the output error of the sensor at zero input, in MPa. The range setting is determined through data analysis. The sensor output value V0 is collected in the no-load state, and the zero offset compensation is calculated: P 实际 =k·(V 采集 -V0), P 实际 Indicates the actual pressure value after calibration, in MPa; V采集 It represents the voltage signal collected by the sensor in the no-load state, in V; V0 represents the zero offset voltage, that is, the output voltage of the sensor when there is no pressure, in V. Calculate the filter parameter α to calculate the cutoff frequency fc, P 滤波 =α·P 当前 +(1-α)·P 前次 , α=T+RCT, α represents the filter coefficient, 0<α<1; T represents the sampling time interval, the unit is s; R represents the filter resistance value, the unit is Ω; C represents the filter capacitance value, the unit is F; P 当前 Indicates the pressure value collected at the current moment, in MPa; P 前次 Indicates the pressure value after filtering at the last moment, in MPa; P 滤波 Indicates the pressure output value after filtering, in MPa; α=T / (T+τ), fc=1 / (2πτ). According to the above data, set the sensor working mode: set the sampling frequency fs based on fs>2fc; based on the range of P, the actual physical quantity P after zero point calibration 实际 Set the alarm threshold.

[0136] Step 3: The dust-adhesive robotic arm cleans the mold;

[0137] Step 4: Evaluate the cleaning effect.

[0138] Step 4 is implemented as follows:

[0139] The projection detection robot arm 1 moves back to the geometric center of the mold center hole, and synchronously projects the coded stripes and collects the reflected image according to the process of step 1 to generate the three-dimensional point cloud and surface model after cleaning. The projection detection robot arm 1 is equipped with a projector to project two groups of standard sinusoidal stripes onto the cleaned mold surface. The two groups of standard sinusoidal stripes include a group of horizontal standard sinusoidal stripes and a group of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot arm 1 is triggered to collect the deformed image, and a phase map is generated by a four-step phase shift method. The absolute phase is obtained by decoding, and statistical filtering is used to eliminate outliers with twice the standard deviation as the threshold. The phase map is aligned to the world coordinate system based on the main direction of the mold, and then point cloud data is collected and matched. The high-reflective area is always overexposed, and the height value is compared point by point. The height value is set to be greater than 5 microns as a suspicious residual value. The adjacent suspicious points are merged into a defective area to generate a three-dimensional distribution map of the difference points before and after cleaning. After obtaining the three-dimensional distribution map of the difference points, the difference point extraction and defect detection are completed: first, the difference point cloud area is triangulated, and then the defect area is calculated by multiplying the base area and the dust height. Finally, the calculated defect area is used to estimate the residual stain amount and output the defect position distribution map and the length, width and depth data of the defect. Finally, the quantitative standard calculation and evaluation are performed: the defect area is triangulated, and the microelement volume is calculated by the length, width and depth data and then summed. The density value is set to 1.2 g / cm3 based on the obtained volume. The residual mass of the dirt is calculated by the volume and area score. A continuous contour line is intercepted along the normal direction of the mold surface from the defect-free area (the length is set to 5 mm), thereby obtaining a discrete height value sequence. The average value of the sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is taken out; when the Ra value is greater than 0.4, it is judged to be unqualified;

[0140] The test results are fed back to the control system in real time. If the standards are met, the mold cleaning work is completed. If the standards are not met, the secondary cleaning process is triggered and the cleaning path is optimized. At the same time, the projection brightness or camera exposure parameters are dynamically adjusted according to the changes in surface reflectivity to adapt to the detection requirements.

Claims

1. A multi-robot collaborative insulator mold cleaning method, characterized in that: Please follow the steps below to implement: Step 1: Mold dust detection; Step 2: Adaptive dust blowing; Step 3: The dust-adhesive robotic arm cleans the mold; Step 4: Evaluate the cleaning effect.

2. The insulator mold cleaning method using multiple robotic arms as claimed in claim 1, characterized in that: In the step 1, an insulator mold cleaning assembly is first arranged, the assembly including an insulator mold to be cleaned, a purge robot arm (2) installed in front of the insulator mold, a high-pressure nozzle (5) provided at the end of the purge robot arm (2), the high-pressure nozzle (5) being driven by a pneumatic system (6), the pneumatic system being connected to the purge robot arm (2) through a pipeline, providing high-pressure gas to the purge robot arm (2) to perform a purge action, and a pressure sensor (4) being further provided on the purge robot arm 2; the pressure sensor (4) being connected to the control unit; The device also includes a projection detection mechanical arm (1) located above the side of the insulator mold, wherein the projection detection mechanical arm (1) is integrated with a fringe projector and a camera for detecting surface cleanliness; the projection detection mechanical arm (1) is connected to a control unit; The invention also includes a dust sticking mechanical arm (3), which is connected to the hydraulic system and the control unit at the same time. The dust sticking mechanical arm (3) is deployed on the other side of the insulator mold and is used to clean fine particles remaining after purging. The dust sticking mechanical arm (3) is also equipped with a replaceable dust sticking paper tape (8). The head of the dust sticking mechanical arm (3) is also provided with a silicone push rod (7). The silicone push rod (7) presses down the dust sticking paper tape (8) to remove dust remaining on the mold surface. The dust sticking mechanical arm (3) is also provided with a pressure sensor (14); the pressure sensor (14) is connected to the control unit.

3. The insulator mold cleaning method using multiple robotic arms as claimed in claim 2, characterized in that: The step 1 is specifically implemented according to the following steps: Step 1.1, obtain the three-dimensional coordinates of the geometric center of the center hole (12) of the insulator mold through the camera integrated in the end effector of the projection detection robot arm (1), and establish a mold coordinate system, and sequentially plan the collision-free motion trajectory of the projection detection robot arm (1), the purge robot arm (2) and the dust-sticking robot arm (3) from their respective initial positions to the front of the mold in the mold coordinate system, and the end point of the trajectory is perpendicular to the mold surface and is 100-300 mm away from the mold surface; store the planned path parameters in the control unit, detect the projection detection robot arm (1) to call the stored path parameters, drive the projection detection robot arm (1) to move and position to the front of the insulator mold according to the preset trajectory, adjust the projector and camera at the front end of the projection detection robot arm (1) to a distance of 100-300 mm from the mold surface, and complete the system calibration to ensure that the projection stripes are aligned with the coordinate system of the image capture; Step 1.2: The projector at the front end of the projection detection robot (1) projects two sets of standard sinusoidal stripes onto the cleaned mold surface. The two sets of standard sinusoidal stripes include a set of horizontal standard sinusoidal stripes and a set of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot (1) is triggered to collect the deformed image. The phase map is generated by the four-step phase shift method. The absolute phase is obtained by decoding. Statistical filtering is used to eliminate outliers with a threshold of twice the standard deviation. The direction perpendicular to the geometric center surface of the mold center hole (12) is the main direction. The phase map in the pixel coordinate system is mapped to the world coordinate system based on the main direction. Then, point cloud data collection and matching are performed: First, the grayscale average is projected. A uniform and saturated image is collected at a high exposure time, and the saturated pixels in the image collected by the camera integrated in the end effector of the detection robot (1) are marked and projected to obtain the imaging range of the high-reflective area on the surface of the object; an image with the same uniform and saturated grayscale as the above is projected, and the exposure time is reduced so that the imaging of the high-reflective area on the surface of the object is not saturated, and the image under the exposure is re-collected, and adaptive low-grayscale sinusoidal stripes are generated by combining the two exposure data, and the high-reflective area is always overexposed, and the height value is compared point by point, and a height value greater than 5 microns is set as a suspicious residual value, and adjacent suspicious points are merged into a defect area, and a three-dimensional distribution map of the difference points before and after cleaning is generated to obtain a three-dimensional distribution map of the difference points; Step 1.3, complete the difference point extraction and defect detection: first, construct the discrete difference point cloud into a triangular mesh to form a continuous surface model; for each triangular unit, calculate its projected area: S basis = 21×|(P2-P1)×(P3-P1)|, where P1, P2, and P3 are the coordinates of the three vertices of the triangle; set the height threshold h=5μm, and mark the point with height difference Δh>h as a suspicious residual point; integrate the S basis i×Δh by micro-element accumulation addition, assume that the contaminant density ρ=1.2g / cm3, and calculate the stain mass: m=V×ρ; take the geometric center of the mold center hole (12) as the origin, establish a local coordinate system, align the main direction of the defect area with the coordinate axis, and collect the length, width, and depth data of the defect: record the maximum span along the main direction as length L, the maximum span perpendicular to the main direction as width W, and the maximum height difference of the defect area relative to the reference plane as depth D, map the three-dimensional coordinates and height values ​​of the defect area to the color gradient, and generate an intuitive defect distribution map; Step 1.

4. Finally, perform quantitative standard calculation and evaluation: triangulate the defective area, calculate the volume of the microelement using the length, width and depth data, and then sum them up. Based on the obtained volume, set the density value to 1.2 g / cm3, calculate the residual mass of the dirt using the volume-area result, and intercept a continuous contour line along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the discrete height value sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is removed; when the Ra value is greater than 0.4, it is judged to be unqualified and proceed to the next step.

4. The insulator mold cleaning method using multiple robotic arms as claimed in claim 3, characterized in that: The parameters in step 1 include joint angle sequence, end position coordinates and movement speed.

5. The insulator mold cleaning method using multiple robotic arms as claimed in claim 3, characterized in that: The step 2 is specifically implemented according to the following steps: The purge robot 2 is positioned at the mold position according to the precise mold surface model determined in step 1, and a pressure sensor 4 is selected and integrated at the front end of the nozzle 5 to detect the gas pressure ejected from the nozzle 5 in real time, providing high-pressure air during the purge process on the mold surface. The control unit is connected to the pressure sensor 4 through an electrical interface, and the corresponding sensor range, zero point calibration, filtering parameters, sensor working mode, and accuracy setting parameters are configured on the control unit end to facilitate subsequent accurate conversion and analysis of the collected data. The pressure fluctuation rate δP / δt is calculated in real time based on the above data settings. If the pressure fluctuation rate δP / δt exceeds the set value, it is determined to be a pipeline blockage or leakage; and the air pressure system output is dynamically adjusted by the control unit to maintain the purge pressure within the set range of 0.3±0.02MPa; at the same time, the pressure-time curve is recorded, and the long-term stability is analyzed based on statistical process control (SPC), and a calibration reminder is automatically generated to achieve accurate monitoring and control of the mold purge process.

6. The insulator mold cleaning method using multiple robotic arms as claimed in claim 5, characterized in that: In step 2, the range is set according to the sensor specification input range, and a linear mapping formula between the electrical signal and the physical quantity is established: P = k·V+b, where P represents the physical pressure value, i.e., the actual purge pressure, in MPa; V represents the voltage signal output by the pressure sensor; k represents the calibration coefficient, which represents the conversion sensitivity between voltage and pressure, in MPa / V; b represents the zero offset, which reflects the output error of the sensor at zero input, in MPa. The range setting is determined by data analysis, and the sensor output value V0 is collected under no-load state to calculate the zero offset compensation: P 实际 =k·(V 采集 -V0), P 实际 Indicates the actual pressure value after calibration, in MPa; V 采集 It represents the voltage signal collected by the sensor in the no-load state, in V; V0 represents the zero offset voltage, that is, the output voltage of the sensor when there is no pressure, in V. Calculate the filter parameter α to calculate the cutoff frequency fc, P 滤波 =α·P 当前 +(1-α)·P 前次 , α=T+RCT, α represents the filter coefficient, 0<α<1; T represents the sampling time interval, the unit is s; R represents the filter resistance value, in Ω; C represents the filter capacitance value, in F; P 当前 Indicates the pressure value collected at the current moment, in MPa; P 前次 Indicates the pressure value after filtering at the last moment, in MPa; P 滤波 Indicates the pressure output value after filtering, in MPa; α=T / (T+τ), fc=1 / (2πτ). According to the above data, set the sensor working mode: set the sampling frequency fs based on fs>2fc; based on the range of P, the actual physical quantity P after zero point calibration 实际 Set the alarm threshold.

7. The insulator mold cleaning method using multiple robotic arms as claimed in claim 4, characterized in that: The step 3 is specifically implemented according to the following steps: Rotate the mold bracket to face the insulator mold to the dust sticking robot arm 3. The control unit marks the area with an Ra value greater than 0.4 in step 1 as the area to be dust-sticked. The dust sticking robot arm 3 starts the double-roller system wrapped with the sticky paper tape 8 to rotate and release a section of the sticky paper tape, and controls the silicone ejector rod 7 to approach the mold. The pressure sensor 14 controls the force with which the silicone ejector rod 7 contacts the mold, thereby sticking stubborn dust and stains through the sticky paper tape 8. After cleaning is completed, rotate the mold bracket to move the mold to the front of the projection detection robot arm.

8. The insulator mold cleaning method using multiple robotic arms as claimed in claim 7, characterized in that: The step 4 is specifically implemented according to the following steps: The projection detection robot arm 1 moves back to the geometric center of the mold center hole, and synchronously projects the coded stripes and collects the reflected image according to the process of step 1 to generate the three-dimensional point cloud and surface model after cleaning. The projection detection robot arm 1 is equipped with a projector to project two groups of standard sinusoidal stripes onto the cleaned mold surface. The two groups of standard sinusoidal stripes include a group of horizontal standard sinusoidal stripes and a group of vertical standard sinusoidal stripes. At the same time, the camera integrated in the end effector of the projection detection robot arm 1 is triggered to collect the deformed image, and the phase map is generated by the four-step phase shift method. The absolute phase is obtained by decoding, and statistical filtering is used to eliminate outliers with twice the standard deviation as the threshold. The phase map is aligned to the world coordinate system based on the main direction of the mold, and then point cloud data is collected and matched. The high-reflective area is always overexposed, and the height value is compared point by point. The height value is set to be greater than 5 microns as a suspicious residual value. The adjacent suspicious points are merged into the defect area to generate the difference points before and after cleaning. Distribution 3D map, after obtaining the 3D map of difference point distribution, complete the difference point extraction and defect detection: first, triangulate the difference point cloud area, then use the product of the base area and the dust height to calculate the defect area, and finally use the calculated defect area to estimate the residual stain amount to output the defect position distribution map and the length, width and depth data of the defect, and finally perform quantitative standard calculation and evaluation: triangulate the defect area, calculate the microelement volume through the length, width and depth data and then sum them up, and set the density value to 1.2 g / cm3 based on the obtained volume. Calculate the residual mass of dirt based on the volume and area results, and intercept a continuous contour line along the normal direction of the mold surface from the defect-free area to obtain a discrete height value sequence. The average value of the sequence is the surface Ra value. When the Ra value is less than 0.4, it is judged to be qualified, the process ends, and the cleaned mold is taken out; when the Ra value is greater than 0.4, it is judged to be unqualified; The test results are fed back to the control system in real time. If the standards are met, the mold cleaning work is completed. If the standards are not met, the secondary cleaning process is triggered and the cleaning path is optimized. At the same time, the projection brightness or camera exposure parameters are dynamically adjusted according to the changes in surface reflectivity to adapt to the detection requirements.

Images