A micro visual detection vehicle and method for detecting cavitation defects of a water turbine top cover

Abstract

This invention discloses a miniature visual inspection vehicle and method for detecting cavitation defects in turbine roofs. The vehicle includes a drive module, a sampling device, a guide mechanism, a vehicle shell, and a diffuse reflection light source plate. The drive module provides stable movement for the miniature visual inspection vehicle and includes a load-bearing chassis located at the bottom of the vehicle shell. The load-bearing chassis has a drive wheel and a driven wheel, with the input end of the drive wheel connected to the output end of a combined motor. The sampling device projects structured light and acquires images. It includes an integrated sampling fixture mounted on the vehicle shell, on which a miniature structured light source and an endoscope are installed. The guide mechanism ensures stable movement of the inspection vehicle along a predetermined path within a narrow annular cavity. It includes guide wheels mounted on the vehicle shell. The diffuse reflection light source plate provides a high signal-to-noise ratio illumination environment for visual inspection. This invention enables in-situ detection of cavitation defects in turbine roofs.

Description

Technical Field

[0001] This invention relates to the field of machine vision measurement technology, and in particular to a miniature vision inspection vehicle and method for detecting cavitation defects in the top cover of a water turbine. Background Technology

[0002] The top cover and runner are the core components of the turbine. The lower surface of the top cover and the upper crown of the runner form a narrow annular cavity. During long-term operation, this annular cavity is continuously subjected to multiple complex loads such as water flow impact, mechanical vibration, and temperature fluctuations, which leads to honeycomb-like cavitation defects on the inner surface of the top cover. This significantly reduces the unit's operating efficiency and accelerates the wear process of related components.

[0003] The height of the annular cavity is only 20. mm Existing penetration devices, such as pipeline robots, are all larger than 20 cm in size. mm The current detection technology still relies on disassembly during annual overhauls, which is costly and inefficient, and makes it difficult to grasp the dynamic evolution of defects in the annular cavity during maintenance intervals. However, using in-situ micro-detection devices for in-situ detection has significant advantages. Wang Qian et al. from Shenyang University of Technology invented a "multi-pipeline defect detection trolley based on machine vision." This patent achieves automated detection of pipeline defects through an adaptive variable diameter structure, spherical connectors, and machine vision technology, compatible with pipes of different diameters. However, this patent uses a fisheye lens solution based on a monocular camera, and image distortion in the edge area leads to a decrease in defect detection accuracy. Chen Fafa et al. from Three Gorges University invented a "visual detection trolley and method for detecting defects on the inner wall of a scraper conveyor." This patent uses permanent magnet wheels and electromagnets to adhere to the inner wall of the scraper conveyor, with multiple cameras rotating collaboratively and IMU sensors for positioning, achieving remote intelligent detection of defects on the inner wall of the scraper conveyor. However, this visual detection trolley lacks a guiding mechanism, causing path deviation during its movement. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a miniature vision inspection vehicle and method for detecting cavitation defects in the top cover of a water turbine, so as to achieve in-situ detection of cavitation defects in the top cover of a water turbine.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a miniature visual inspection vehicle for detecting cavitation defects in the top cover of a water turbine, comprising a drive module, a sampling device, a guiding mechanism, a vehicle shell, and a diffuse reflection light source plate, characterized in that: The drive module provides stable walking capability for the miniature vision inspection vehicle; it includes a load-bearing chassis located at the bottom of the vehicle body, on which a drive wheel and a driven wheel are provided, and the input end of the drive wheel is connected to the output end of the combined motor; The sampling device projects structured light and acquires images; it includes an integrated sampling fixture mounted on the vehicle body, on which a miniature structured light source and an endoscope are installed. The guiding mechanism ensures that the testing vehicle moves stably along a predetermined path within a narrow annular cavity; it includes guide wheels mounted on the vehicle body. Diffuse light source panels provide a high signal-to-noise ratio lighting environment for visual inspection.

[0006] Preferably, the combined motor integrates two independent drive motors, and steering is achieved through the differential principle of the two motors.

[0007] Preferably, the load-bearing chassis is made of carbon fiber material; the drive wheel and driven wheel are made of neodymium iron boron permanent magnet material and the wheel surface has a knurled structure; the drive module adopts a symmetrical layout of two drive wheels and two driven wheels.

[0008] Preferably, there are two sets of the sampling device, symmetrically arranged on the vehicle body to form an overlapping measurement field of view.

[0009] Preferably, the integrated feeding and mining fixture is made of aviation aluminum alloy, connected to the vehicle body through a three-point positioning system, and the cable outlet has a built-in silicone cable protector.

[0010] Preferably, there are three guide wheels, evenly distributed on one side of the vehicle body, made of neodymium iron boron permanent magnet material, and the middle guide wheel has a preset radial offset relative to the two side guide wheels; the radial offset parameter of the middle guide wheel relative to the two side guide wheels is determined by the following formula: ; In the formula: This represents the radial offset of the middle guide wheel relative to the two side guide wheels. Let be the radius of curvature of the arc to be tracked. w The center distance between the mounting bases of the guide wheels on both sides.

[0011] Preferably, the diffuse reflection light source plate is a broadband array of ultra-thin surface-mount LEDs, and adopts a combination structure of ring LEDs and diffuse plate.

[0012] Preferably, the miniature vision inspection vehicle is adapted to the annular cavity of the turbine top cover with a height of 20mm.

[0013] Furthermore, this invention also discloses a method for detecting cavitation defects in turbine roofs using the aforementioned miniature vision inspection vehicle, characterized in that it includes the following steps: S1. The inspection vehicle starts, the magnetic adsorption wheel adheres to the metal wall, and the three-point magnetic guide wheel completes the initial positioning and attitude calibration. S2. The inspection vehicle is driven by dual motors at differential speed and moves at a constant speed along the annular cavity trajectory, fully covering the cavitation detection area. S3. The dual sampling devices operate synchronously, projecting four-step phase-shifting sinusoidal stripes, and the endoscope collects deformed stripes. S4. The system introduces a correction factor k(x,y) and a compensation coefficient ξ to complete the anti-high-reflection phase solution and eliminate metal reflection interference in the image; S5. Calculate the phase difference Δφ(x,y) between the surface to be measured and the reference surface, and obtain the true cavitation erosion height through depth-of-field compensation phase-height mapping; S6. Based on the camera's intrinsic and extrinsic parameters, the pixel coordinates are converted into global 3D coordinates to complete the single-module point cloud calculation; S7. Based on the pre-calibrated pose matrix, the point clouds of the two modules are unified to the global coordinate system, and the overlapping field of view is weighted and fused to eliminate stitching errors. S8. The inspection vehicle continuously executes the imaging-reconstruction-stitching process, and finally outputs a three-dimensional point cloud and morphological parameter report of the full cavity cavitation defect.

[0014] Furthermore, in S3, the expression for the fringe intensity corresponding to the projected four-step phase-shift sinusoidal fringes is: ; In the formula: It represents the stripe intensity distribution acquired by the endoscope at the nth phase shift, after being modulated by cavitation morphology and affected by metal reflection; n is 0, 1, 2, and 3 respectively; This is an adaptive correction factor for high reflectivity of metal, with a value of 0~1, which automatically suppresses the intensity weight of the high-reflectivity region of the metal mirror. Background light intensity in the cavity environment. For stripe contrast, The measured true phase is modulated by the cavitation morphology.

[0015] Furthermore, in step S4, the system introduces a correction factor k(x,y) and a compensation coefficient ξ, and the specific calculation expression for the anti-high anti-packaging phase solution is as follows: ; In the formula: This is the phase offset compensation coefficient.

[0016] Furthermore, in step S5, the specific process of calculating the phase difference Δφ(x,y) between the surface to be measured and the reference surface, and obtaining the true height of cavitation erosion through depth-of-field compensation phase-height mapping is as follows: First, calculate the phase difference between the cavitation erosion surface to be tested and the standard reference plane: ; Construct a quadratic mapping model adapted to the nonlinear concave-convex morphology of cavitation erosion: ; In the formula: C is the 20mm narrow cavity depth of field compensation term, which is adapted to the short depth of field characteristics of miniature optical paths; Z 0 represents the fixed reference distance from the reference plane to the optical center of the camera; The measured cavitation erosion surface is the true height relative to the reference surface, and A and B are the system-specific phase-height calibration coefficients. In the calibration experiment, the depth of the camera coordinate system at the point to be measured along the Z-axis is: ; By combining pre-calibrated camera intrinsic parameters, distortion coefficients, and camera extrinsic pose matrices, and through the geometric relationships of pinhole perspective imaging, the entire process of transforming two-dimensional pixel coordinates to local three-dimensional coordinates of the camera and then to global three-dimensional coordinates of the field is completed step by step, accurately solving the true spatial three-dimensional coordinates of each defect pixel. X w , Y w , Z w ).

[0017] Furthermore, in S7, based on the pre-calibrated pose matrix, the point clouds of the two modules are unified to the global coordinate system, and the overlapping field of view is weighted and fused to eliminate stitching errors. The specific process is as follows: S7.1 Place the high-precision chessboard calibration board within the common field of view of the dual modules and acquire multiple sets of pose images; S7.2 Simultaneously calculate the camera intrinsic and extrinsic parameters of the left and right acquisition modules to obtain the rotation matrix of the left module to the world coordinate system. R L Translation vector T L Rotation matrix from the right module to the world coordinate system R R Translation vector T R ; S7.3 Construct the relative pose transformation matrix between the two modules: ; In the formula: R LR This is the rotation matrix from the left module to the right module. T LR This is the translation vector from the left module to the right module. This matrix is ​​a fixed value and is permanently saved after assembly and calibration, so there is no need to repeat the calibration during the testing process. S7.4. Convert the local 3D point clouds calculated by the left and right projection modules to the global world coordinate system: ; S7.5 For the overlapping field of view of the two modules, a weighted mean fusion algorithm is used to complete the precise registration of the point cloud, eliminating stitching gaps and errors; distance-weighted fusion is used for the point cloud in the overlapping area, and the original point cloud is directly retained in the non-overlapping area to eliminate stitching deviations caused by assembly errors and vibrations. ; In the formula: W L , W R Assign point cloud weights to the left and right modules.

[0018] Beneficial effects of this invention: The miniature visual inspection vehicle for cavitation defects in turbine roofs proposed in this invention features a compact and miniaturized overall structure, allowing it to smoothly enter the annular cavity of turbine roofs with a height of only 20mm for in-situ inspection of cavitation defects without disassembly. The drive module employs a neodymium iron boron permanent magnet magnetic adsorption wheel assembly with a diamond-knurled wheel surface design, which can effectively adhere to smooth metal surfaces and pierce water stains and oil films, effectively suppressing the risk of slippage and deviation. Combined with dual-motor differential steering and a single-sided three-point magnetic guide wheel structure, it can accurately match the geometric trajectory of the annular cavity, ensuring a stable and deviation-free inspection path. Furthermore, it utilizes miniaturized four-step phase-shifting fringe projection three-dimensional measurement technology and specifically incorporates metal high reflectivity correction. The system features optimized algorithms for adaptive phase unfolding and depth-of-field compensation in narrow cavities, significantly improving the accuracy and anti-interference capability of 3D reconstruction of cavitation defects without the need for complex optical path parameter calibration. Two sets of symmetrical acquisition devices form an overlapping measurement field of view and achieve full-domain blind-spot-free detection through precise point cloud stitching, completely restoring the full-dimensional morphological features of defects. In addition, it is equipped with an ultra-thin diffuse reflection light source plate to optimize the illumination effect and suppress the interference of metal mirror reflection to obtain high signal-to-noise ratio imaging data. The vehicle has high integration and strong adaptability to working conditions, and can adapt to the harsh working environment of vibration, humidity and impurities in the cavity of water turbines. It has multiple advantages such as high detection accuracy, strong practicality and stability and reliability. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a miniature vision inspection vehicle used for detecting cavitation defects in the top cover of a water turbine.

[0020] Figure 2 This is a schematic diagram of the driver module.

[0021] Figure 3 A schematic diagram of the inspection process for a miniature visual inspection vehicle for cavitation defects on the turbine roof. Explanation of reference numerals in the attached drawings: 1. Driven wheel; 2. Load-bearing chassis; 3. Car body; 4. Integrated mining fixture; 5. Miniature structured light source; 6. Endoscope; 7. Combined motor; 8. Guide wheel; 9. Diffuse reflection light source plate; 10. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0023] Example 1: A miniature visual inspection vehicle for detecting cavitation defects in the top cover of a water turbine, comprising a drive module, a sampling device, a guiding mechanism, a vehicle body, and a diffuse reflection light source plate. The drive module consists of a drive wheel 1, a driven wheel 2, a load-bearing chassis 3, and a combined motor 8, wherein the combined motor 8 integrates two independent drive motors. The sampling device consists of an integrated sampling fixture 5, a miniature structured light source 6, and an endoscope 7.

[0024] The drive module provides stable mobility for the miniature vision inspection vehicle; the projection and acquisition device projects structured light and acquires images; the guide mechanism ensures that the inspection vehicle moves stably along a predetermined path in a narrow annular cavity; the vehicle shell provides a reliable operating platform for the inspection vehicle in harsh cavity environments; and the diffuse reflection light source plate provides a high signal-to-noise ratio lighting environment for vision inspection.

[0025] The specific technical solution is as follows: (1) Driver module design The core function of the drive module is to provide stable and controllable mobility for the miniature vision inspection vehicle, enabling it to operate at a height of only 20 meters. mm It moves autonomously within a narrow annular cavity, overcoming challenges from complex environments such as smooth metal surfaces and water residue. The drive module consists of a drive wheel 1, a driven wheel 2, a load-bearing chassis 3, and a combined motor 8.

[0026] The combined motor 8 is fixed to the load-bearing chassis 3 by bolts; the driving wheel 1 is installed on the output shaft end of the combined motor by interference fit, while the driven wheel 2 is assembled to the corresponding mounting position of the load-bearing chassis 3 by transition fit.

[0027] The miniature vision inspection vehicle adopts a monocoque chassis design. Considering the limited overall size of the inspection vehicle, the chassis cannot be too thick while meeting load-bearing requirements. Therefore, lightweight and high-strength carbon fiber material is selected to construct the monocoque chassis. This material not only ensures adequate mechanical strength but also allows the chassis to adapt to confined spaces, fully meeting the dual requirements of a compact structure and reliable performance. To meet the requirement of the miniature vision inspection vehicle traveling along an arc-shaped trajectory while also considering the design requirements of equipment miniaturization, this solution abandons the traditional steering mechanism and adopts a dual-motor drive scheme. By integrating two independent drive motors (8 motors in total), the rotational speeds of the inner and outer drive wheels are controlled, and the arc-shaped trajectory is achieved using the differential steering principle. The rotational speeds of the two wheels are determined according to the following formula: ; In the formula: The speed of the outer driving wheel, The speed of the inner driving wheel, The radius of motion of the center of the two wheels, The distance between the two drive wheels. ω is the angular velocity at the center of the two wheels.

[0028] This design effectively reduces the space occupied by the mechanical structure while ensuring motion accuracy.

[0029] To overcome the insufficient weight of the miniature vision inspection vehicle and ensure its stable movement on smooth, inclined, and vibrating surfaces, avoiding slippage or deviation, and improving the adhesion between the wheels and the ground, the walking system adopts a symmetrical layout of two driving wheels and two driven wheels. All wheels are made of neodymium iron boron permanent magnet material, achieving a strong bond with the metal surface through magnetic adsorption. The magnetic force not only makes the wheel surface fit tightly against the walking surface, but also generates additional normal pressure, increasing the friction between the driving wheel (1), driven wheel (2) and the walking surface. ; In the formula: The adhesion between the wheels and the ground. The coefficient of friction between the drive wheel and the ground. For the weight of the miniature vision inspection vehicle, It refers to the magnetic attraction between the driving wheel, the driven wheel, and the traveling plane.

[0030] Meanwhile, the drive wheel 1 and driven wheel 2 adopt a knurled design on their surfaces, which has the following three core functions: First, the sharp protrusions can pierce the continuous liquid film on the surface of oil or water stains, allowing the wheel surface to directly contact the base metal and eliminating slippage caused by fluid lubrication; Second, the discrete protrusion design improves friction performance by increasing the effective contact area between the wheel surface and the traveling plane; In addition, the concave and convex texture can adaptively compensate for local unevenness on the surface, with the protrusions prioritizing contact with the high points to maintain traction, while the grooves accommodate debris, thereby enhancing adaptability to complex working conditions.

[0031] (2) Design of mining equipment To address the need for precise reconstruction of the three-dimensional morphology of cavitation defects in turbine roofs, this vision inspection vehicle employs a binocular structured light three-dimensional measurement method: by projecting a structured light pattern onto the surface being measured, a binocular camera simultaneously captures the structured light distortion signal caused by surface deformation, thereby calculating a high-precision three-dimensional morphology. Due to the limited space of the miniature vision inspection vehicle, conventional industrial cameras and projectors are difficult to integrate directly. A customized projection and acquisition device is required to achieve a compact design, ensuring the coordinated operation of precise optical path projection and image acquisition within a limited space. The projection and acquisition device consists of an integrated projection and acquisition fixture (6), a miniature structured light source (7), and an endoscope (8). The two projection and acquisition devices are symmetrically arranged on the vehicle body and form an overlapping field of view.

[0032] The integrated light-emitting and imaging fixture is made of aerospace-grade aluminum alloy. A precision positioning structure ensures a rigid connection between the light source and the camera, guaranteeing a fixed relative position and effectively eliminating displacement deviations caused by vibration. This maintains a constant spatial relationship between the projected structured light and the imaging field of view. The fixture integrates a micro-structured light source mounting slot with a limiting surface perpendicular to the optical axis on the inner wall, effectively constraining the axial displacement of the micro-structured light source. The endoscope mounting hole uses a transition fit design to ensure coaxiality with the endoscope. The fixture achieves precise installation through a three-point positioning system: three triangularly distributed threaded fixing holes engage with high-precision positioning pins to form a precise positioning reference with the corresponding mounting surface on the vehicle body. The cable outlet has a built-in silicone cable protector, facilitating cable management and effectively preventing cable bending and wear.

[0033] (3) Design of guide wheels To ensure the stable movement of the miniature vision inspection vehicle along an arc-shaped trajectory within the annular cavity, a design scheme employing three evenly distributed magnetic guide wheels on one side is adopted. Precise trajectory control is achieved based on the three-point circular arc geometric positioning principle. The radial offset parameter of the middle guide wheel relative to the two side guide wheels is determined by the following formula: ; In the formula: This represents the radial offset of the middle guide wheel relative to the two side guide wheels. Let be the radius of curvature of the arc to be tracked. w The center distance between the mounting bases of the guide wheels on both sides.

[0034] Each guide wheel is made of neodymium iron boron permanent magnet, which adheres tightly to the cavity wall surface through magnetic attraction, forming a stable three-point contact constraint. This design uses magnetic attraction to replace the traditional mechanical pre-tightening device. While ensuring continuous contact between the wheel wall and the cavity wall, the geometric self-locking effect of the three-point contact limits the movement trajectory of the vehicle, allowing the inspection vehicle to travel along the side wall of the annular cavity.

[0035] (4) Lighting design For the inspection environment inside the turbine's top cover, which is confined, dark, and structurally complex, relying solely on structured light results in limited field of view. To meet the needs of operators for real-time monitoring and assisted navigation, a supplementary lighting solution was designed: ultra-thin surface-mount LEDs are integrated into the top of the trolley shell, achieving illumination while maintaining the device's compactness; a broadband LED array provides full-spectrum basic illumination, ensuring that the binocular camera can acquire visible light images of the measured surface even in completely dark environments; and to address the high reflectivity of metal surfaces, a combination of a ring-shaped LED array and a diffuser plate is configured to uniformize the light intensity distribution through diffuse reflection, effectively suppressing the interference of specular reflection on structured light imaging. This solution enhances the visibility of non-geometric features such as rust and water stains on the turbine surface, provides auxiliary visual information for subsequent structured light 3D measurements, and avoids increased size through compact design, ensuring imaging quality in complex environments.

[0036] (5) High-precision three-dimensional reconstruction method for cavitation defects Two sets of sampling devices are symmetrically arranged on the vehicle body shell, forming an effective overlapping measurement field of view. To achieve accurate reconstruction of the three-dimensional morphology of cavitation defects on the turbine roof, this miniature vision inspection vehicle integrates a miniature fringe projection contour measurement method to replace the traditional binocular stereo vision and single-point ranging method. The miniature structured light source 6 in the sampling device constitutes a miniaturized and controllable four-step phase-shift fringe projection unit, which can project a sinusoidal fringe pattern with standard four-step phase-shift characteristics onto the surface of the roof under test; the endoscope 7 acts as a miniature imaging camera, simultaneously acquiring deformed fringe images modulated by the depth of the cavitation defects. The system calculates the image based on the four-step phase-shift algorithm to obtain the wrapping phase distribution of the surface under test. To address the problems of strong specular reflection from the metal inner wall of the turbine cavity, saturation distortion of fringe light intensity, and large interference from background stray light, this invention introduces a metal high reflectivity correction factor k(x,y) into the classic four-step phase-shift light intensity model, constructing an anti-high reflectivity exclusive light intensity distribution formula: The system calculates the wrapping phase distribution of the surface under test according to the following four-step phase-shift algorithm: ; In the formula: It represents the stripe intensity distribution acquired by the endoscope at the nth phase shift, after being modulated by cavitation morphology and affected by metal reflection; n is 0, 1, 2, and 3 respectively; This is an adaptive correction factor for high reflectivity of metal, with a value of 0~1, which automatically suppresses the intensity weight of the high-reflectivity region of the metal mirror. Background light intensity in the cavity environment. For stripe contrast, The measured true phase is modulated by the cavitation morphology.

[0037] Based on the corrected light intensity, an anti-saturation phase calculation formula is constructed to replace the traditional pure arctangent calculation: ; In the formula: The phase offset compensation coefficient is used. Due to the properties of the arctangent function in equation (5), the demodulated phase value is wrapped between (-π, π). This invention uses a spatial phase expansion method to solve the continuous phase. By comparing the phase values ​​of adjacent pixels, a phase jump is judged when the phase difference between adjacent points is greater than π. The continuous phase is restored by adding or subtracting 2π from the phase value. In view of the problems of abrupt changes in cavitation defect depth, many honeycomb holes, easy errors in traditional row-by-row expansion, and noise sensitivity, this invention introduces the cavitation edge discrimination weight ω(x) and the noise smoothing factor σ to construct a phase expansion formula specific to cavitation morphology: ; In the formula, the size of the wrapped phase map is m×n. This algorithm performs phase unwrapping operation independently for each row of the image: φ(1) is the initial wrapped phase of the first column pixel of the current processing row, φ(x) is the wrapped phase of the x-th column pixel of the row, and Δφ(x) is the wrapped phase difference between adjacent columns of pixels; ω(x) is the cavitation edge discrimination weight, which is used to weight and strengthen the phase compensation of the abrupt edge region of the cavitation defect to ensure the phase continuity of the defect edge; σ is the noise smoothing factor, which is used to suppress the interference of metal reflection noise and electrical noise in the narrow cavity on the phase difference; W[ This is a phase jump compensation operator with cavitation erosion abrupt change fault tolerance. It can adaptively determine the phase period jump based on the phase difference between adjacent pixels and complete ±2π correction, effectively solving the phase unfolding disorder problem caused by abrupt changes in cavitation erosion defect depth and noise interference.

[0038] To address the challenge of accurately calibrating the relative spatial structure parameters of the camera and grating in a physical grating micro-projection optical path, this invention does not rely on the traditional triangulation ranging optical path model. Instead, it constructs a quadratic polynomial implicit phase-height mapping model that does not require optical path geometric parameters, directly establishing a one-to-one correspondence between continuous absolute phase difference and the height of the cavitation erosion surface.

[0039] First, calculate the phase difference between the cavitation erosion surface to be tested and the standard reference plane: ; Construct a quadratic mapping model adapted to the nonlinear concave-convex morphology of cavitation erosion: ; In the formula: C is the 20mm narrow cavity depth of field compensation term, which is adapted to the short depth of field characteristics of miniature optical paths; Z 0 represents the fixed reference distance from the reference plane to the optical center of the camera; A represents the true height of the cavitation erosion surface relative to the reference plane, and B represents the system-specific phase-height calibration coefficients.

[0040] In the calibration experiment, the depth of the camera coordinate system at the point to be measured along the Z-axis is: ; By combining pre-calibrated camera intrinsic parameters, distortion coefficients, and camera extrinsic pose matrices, and through the geometric relationships of pinhole perspective imaging, the entire process of transforming two-dimensional pixel coordinates to local three-dimensional coordinates of the camera and then to global three-dimensional coordinates of the field is completed step by step, accurately solving the true spatial three-dimensional coordinates of each defect pixel. X w , Y w , Z w ).

[0041] (6) Precise point cloud stitching of dual-projection modules After completing the three-dimensional point cloud calculation of cavitation defects in a single sampling module, this invention addresses the problem of limited field of view of a single module and measurement blind spots in narrow annular cavities. It uses two sets of sampling modules to form an overlapping measurement field of view. Based on the pre-calibrated relative physical pose relationship between the two modules, it achieves global point cloud stitching with no external markers, low computing power, and high precision, completely covering the cavitation area of ​​the turbine top cover and eliminating measurement blind spots.

[0042] During the system assembly phase, a high-precision checkerboard joint calibration method was used to obtain the relative pose relationship between the two sets of mining modules: A high-precision chessboard calibration board was placed within the common field of view of the two modules, and multiple sets of pose images were acquired. Simultaneously calculate the camera intrinsic and extrinsic parameters of the left and right acquisition modules to obtain the rotation matrix of the left module to the world coordinate system. R L Translation vector T L Rotation matrix from the right module to the world coordinate system R R Translation vector T R ; Relative pose transformation matrix between the two modules: ; In the formula: R LR This is the rotation matrix from the left module to the right module. T LR This is the translation vector from the left module to the right module. This matrix is ​​a fixed value and is permanently saved after assembly and calibration, so there is no need to repeat the calibration during the testing process.

[0043] The local 3D point clouds calculated by the left and right projection modules are uniformly transformed to the global world coordinate system: ; For the overlapping field of view of the two modules, a weighted mean fusion algorithm is used to complete the precise registration of point clouds, eliminating stitching gaps and errors. Distance-weighted fusion is applied to the point clouds in the overlapping area, while the original point clouds are directly retained in the non-overlapping areas, eliminating stitching deviations caused by assembly errors and vibrations. ; In the formula: W L , W R Assign point cloud weights to the left and right modules.

[0044] Example 2: In this example, the height of the annular cavity is selected as follows: 22 mm The diameter of the camera endoscope 7 is The field of view is 140° and the focal length is f. x =2487.62、f y =2468.38, the size of the miniature structured light source 6 is 4×3×2 mm The diffuse reflection light source panel 10 measures 6×3×0.2 mm. mm The radius of curvature of the annular cavity is 4000. mm .

[0045] Specific implementation steps: (1) Driver module The drive module consists of a drive wheel 1, a driven wheel 2, a load-bearing chassis 3, and a combined motor 8. The load-bearing chassis 3 has a thickness of 2 mm. mm It is made of carbon fiber composite material in one piece. The load-bearing chassis 3 has four M0.8 threaded holes for connecting the combined motor 8 and the body shell 4. The radius of motion of the two wheel centers. R For 4000 mm Distance between drive wheels H for 20 mm, angular velocity at the center of the two wheels for Substituting the data into equation (1) yields the solution. It is 6.99 mm / s, It is 6.96 mm / s.

[0046] The diameters of driving wheel 1 and driven wheel 2 are 4 mm Width is 2 mm It is made of neodymium iron boron permanent magnet material and machined into one piece, and its magnetic attraction with the walking plane is strong. The weight of the miniature vision inspection vehicle is 10N. The coefficient of friction is 2N. The driving wheel 1 and driven wheel 2 feature a diamond knurled design, and their friction coefficient with the traveling surface is [missing information]. The value is 0.65. Finally, substituting the above parameters into equation (2), we obtain the adhesion force between the driving wheel 1, the driven wheel 2, and the ground. :

[0047] During assembly, the driving wheel 1 and the driven wheel 2 are first connected to the output shaft of the combined motor 8 through an interference fit. Then, the combined motor 8 is connected to the load-bearing chassis 3 through bolts. Finally, the driven wheel 2 is installed onto the load-bearing chassis through an transition fit.

[0048] (2) Mining equipment and lighting equipment The extraction device consists of an integrated extraction clamp 5, a miniature structured light source 6, and an endoscope 7. The integrated extraction clamp 5 is made of 6061-T6 aluminum alloy and is machined in one piece. The diameter of the endoscope 7 is... 6 mm The size of the miniature structured light source 6 is 4×3×2 mm The diameter of the mounting hole of the integrated mining fixture 5 is determined to be... 5.95 mm The mounting slot for the miniature structured light source 6 is 4×3×3. mm During installation, the endoscope 7 is first fitted into the mounting hole of the integrated delivery and acquisition fixture 5 using an interference fit. Then, the bottom surface of the micro-structured light source 6 is aligned with the limiting surface of the mounting groove of the micro-structured light source 6 and fixed with metal glue. In the final assembly stage, a precision assembly process is used to complete the installation of the two delivery and acquisition devices. First, the delivery and acquisition devices are rigidly connected to the vehicle body using anti-loosening bolts. The relative positions of the two devices are calibrated. While ensuring the coaxiality and parallelism tolerances of the mechanical structure, the spatial orientation of the endoscope 7 is adjusted to ensure that the two fields of view form an effective overlap area of ​​not less than 30%.

[0049] The lighting device consists of a diffuse reflection light source panel 10, which uses a light source with a wavelength of 700 nm. nm Its size is 6×3×0.2 mm The diffuse reflection light source plate 10 is attached to the car body 4 with metal glue.

[0050] (3) Guide wheel Guide wheel 9 is made of neodymium iron boron permanent magnet material and is integrally machined. The diameter of guide wheel 9 is [missing information]. 5 mm Thickness is 2 mm There are three guide wheels 9. These three guide wheels 9 are evenly distributed along the length of the miniature vision inspection vehicle. The radial offset of the middle guide wheel 9 relative to the two side guide wheels 9 is... The radius of curvature of the arc to be tracked is determined by equation (3). For 4000 mm The center distance between the mounting bases of the two guide wheels 9w For 20 mm Substituting the data into equation (3) yields It is 0.0125 mm Mounting holes for guide wheels 9 are machined on the body shell 4. The radial offset of the middle mounting hole relative to the other two mounting holes in the direction away from the center is 0.0125. mm Three guide wheels 9 are mounted on the body shell 4 with clearance fit.

[0051] (4) High-precision three-dimensional reconstruction of cavitation defects First, turn off the microstructured light source and turn on the diffuse reflection light source plate. Use an endoscope to acquire a pure background light intensity image of the 304 stainless steel standard plate. A total of 10 sets of data were collected, and the average value was taken to measure the light intensity range of the entire image. , Based on this, the metallic high reflectivity correction factor is calculated pixel by pixel. High reflectivity areas (Automatic suppression of light intensity weighting), ordinary reflection area (Preserve effective light intensity), dark areas in the background and will A pixel-level mapping table was generated; then a miniature structured light source was turned on, projecting four-step phase-shifted sinusoidal fringes, and images of the deformed fringes on the stainless steel plate were acquired, and four sets of light intensity data were measured. , , , Substitute the values ​​into the light intensity model to calculate the background light intensity. Stripe contrast Take the pre-calibrated phase offset compensation coefficient. Correction factor Take the ordinary reflection area High reflectivity area Under two typical operating conditions, the phase calculation formula yields a phase of approximately 0.089 rad for the ordinary reflection zone and approximately 0.053 rad for the high reflectivity zone. Subsequently, the phase calculation is performed using an adaptive phase expansion formula for abrupt erosion changes, with a pre-calibrated noise smoothing factor applied during the process. And adaptively assign weights based on the phase gradient difference at the cavitation edge. (The value is 1.5 for the abrupt change edge of cavitation, 1.0 for the transition region, and 0.5 for the smooth region.) Taking five consecutive pixels containing the edge of a cavitation defect as an example, their wrapping phases are 0.20 rad, 0.35 rad, -2.80 rad, -2.60 rad, and -2.45 rad, respectively. First, calculate the phase difference between adjacent pixels. The values ​​are 0.15 rad, -3.15 rad, 0.20 rad, and 0.15 rad, respectively, and then corrected by a phase jump operator. Phase differences exceeding the threshold were corrected by ±2π, resulting in corrected phase differences of 0.15 rad, approximately 3.13 rad, 0.20 rad, and 0.15 rad, respectively. These corrected phase differences were then weighted and summed, and a noise correction term was subtracted to obtain the continuous absolute phase at each point. The continuous phase of the 5th pixel was approximately 5.47 rad. Following these steps, absolute phase data were collected between the 304 stainless steel reference plane and the cavitation-eroded test surface to obtain the reference average phase. Substitute the values ​​into the phase difference calculation formula to obtain the phase difference of the region to be measured. The range is Offline calibration was performed using a precision displacement stage, determining fixed coefficients A=4.79, B=0.16, and depth-of-field compensation parameter C=0.07, combined with the reference distance. By substituting the phase difference into the depth-of-field compensation phase-height mapping formula, the true height of the cavitation defect can be accurately calculated. Combined with the pre-calibrated camera internal parameters , and pixel principal point , Select typical defect pixel coordinates u=700, v=400, corresponding to the defect height. Substituting the values ​​into the 3D coordinate transformation formula, the transformation from 2D pixel coordinates to camera coordinates and then to global world coordinates is completed sequentially, resulting in accurate single-module 3D point cloud data.

[0052] (5) Precise point cloud stitching of dual-projection acquisition module First, offline joint calibration of the dual-projection and mining device was completed using a high-precision checkerboard calibration board to obtain the relative pose relationship between the left and right modules, including the relative rotation matrix. With relative translation vector Select typical cavitation defect points in the left module. Its global coordinates are The right module corresponds to the overlapping area point. The local coordinates are First, the local point cloud of the right module is processed through its own extrinsic parameters. , Transform to the global coordinate system, then perform precise registration using relative pose parameters, and substitute into the transformation formula: Calculated The global coordinates are This achieves initial alignment of the point clouds of the two modules in the global coordinate system. For the overlapping field of view of the two modules, a distance-weighted fusion algorithm is adopted, assigning weights based on the distance of the point cloud to the reference plane: nearby points ≤2mm from the reference plane are selected. , For points with a distance > 2mm, take... , Substitute into the weighted fusion formula ,right and The fusion calculation was performed, and the fused coordinates were obtained. .

[0053] (6) Inspection process of miniature visual inspection vehicle for cavitation defects in turbine roof After the inspection vehicle starts, the magnetic adsorption wheel adheres to the metal wall and completes initial positioning and attitude calibration through the three-point magnetic guide wheel. Then, driven by dual motors at differential speed, it moves at a constant speed along the annular cavity trajectory, covering the entire cavitation detection area. During movement, the dual projection devices simultaneously project four-step phase-shift sinusoidal fringes. An endoscope acquires the deformed fringe image modulated by the cavitation defect. The system sequentially introduces the metal high reflectivity correction factor k(x,y) and the phase shift compensation coefficient ξ to complete the anti-high reflectivity encapsulation phase calculation. Then, relying on the cavitation edge discrimination weight ω(x), noise smoothing factor σ, and phase jump compensation operator W[ To achieve adaptive phase expansion for abrupt cavitation changes, the phase difference Δ between the measured surface and the reference plane is calculated. The true height h(x,y) of the cavitation defect is calculated using a quadratic polynomial implicit mapping model with a narrow cavity depth compensation term. Then, based on the camera's intrinsic and extrinsic parameters, the two-dimensional pixel coordinates are converted into global three-dimensional coordinates to complete the single-module three-dimensional point cloud calculation. Subsequently, the left and right projected point clouds are unified to the global coordinate system according to the pre-calibrated dual-module relative pose matrix. The overlapping field-of-view point clouds are weighted and fused to eliminate stitching errors and generate a single-frame global point cloud. The inspection vehicle continues to move and cyclically executes the imaging, reconstruction and stitching process, and finally outputs a complete three-dimensional point cloud of the cavitation defect covering the entire annular cavity and a report on the defect morphology parameters.

[0054] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A miniature visual inspection vehicle for detecting cavitation defects in the top cover of a water turbine, comprising a drive module, a sampling device, a guiding mechanism, a vehicle shell (4), and a diffuse reflection light source plate (10), characterized in that: The drive module provides stable walking capability for the micro vision inspection vehicle; it includes a load-bearing chassis (3) located at the bottom of the vehicle body (4), and the load-bearing chassis (3) is provided with a drive wheel (1) and a driven wheel (2), the input end of the drive wheel (1) being connected to the output end of the combined motor (8); The data acquisition device projects structured light and acquires images; It includes an integrated mining fixture (5) mounted on the vehicle body (4), and a micro structured light source (6) and an endoscope (7) are installed on the integrated mining fixture (5). The guiding mechanism ensures that the testing vehicle moves stably along a predetermined path in a narrow annular cavity; it includes guide wheels (9) mounted on the vehicle body (4). The diffuse light source plate (10) provides a high signal-to-noise ratio lighting environment for visual inspection.

2. The miniature visual inspection vehicle for detecting cavitation defects in the top cover of a water turbine as described in claim 1, characterized in that: The combined motor (8) integrates two independent drive motors, and steering is achieved through the differential principle of the two motors.

3. A miniature vision inspection vehicle for detecting cavitation defects in the top cover of a water turbine, as described in claim 1, is characterized in that... The load-bearing chassis (3) is made of carbon fiber material; the drive wheel (1) and driven wheel (2) are made of neodymium iron boron permanent magnet material and the wheel surface is knurled. The drive module adopts a symmetrical layout of two drive wheels and two driven wheels.

4. A miniature vision inspection vehicle for detecting cavitation defects in the top cover of a water turbine, as described in claim 1, is characterized in that... The mining device consists of two sets, symmetrically arranged on the vehicle body (4) to form an overlapping measurement field of view.

5. A miniature vision inspection vehicle for detecting cavitation defects in the top cover of a water turbine, as described in claim 1, is characterized in that... The integrated mining fixture (5) is made of aviation aluminum alloy and is connected to the car body (4) through a three-point positioning system. The cable outlet has a built-in silicone cable protector.

6. A miniature vision inspection vehicle for detecting cavitation defects in the top cover of a water turbine, as described in claim 1, is characterized in that... The guide wheels (9) are three in number and evenly distributed on one side of the vehicle body (4). They are made of neodymium iron boron permanent magnet material. The middle guide wheel (9) has a preset radial offset relative to the two side guide wheels (9). The radial offset parameter of the middle guide wheel relative to the two side guide wheels is determined by the following formula: ； In the formula: This represents the radial offset of the middle guide wheel relative to the two side guide wheels. Let be the radius of curvature of the arc to be tracked. w The center distance between the mounting bases of the guide wheels on both sides.

7. A miniature vision inspection vehicle for detecting cavitation defects in the top cover of a water turbine, as described in claim 1, is characterized in that... The diffuse reflection light source plate (10) is a wide-spectrum array integrated with ultra-thin patch LEDs, and adopts a combination structure of ring LEDs and diffuse plate.

8. A miniature vision inspection vehicle for detecting cavitation defects in the top cover of a water turbine, as described in claim 1, is characterized in that... The miniature vision inspection vehicle is adapted to the annular cavity of the turbine top cover with a height of 20mm.

9. A method for detecting cavitation defects in the roof of a turbine using a miniature vision inspection vehicle as described in any one of claims 1 to 8, characterized in that: It includes the following steps: S1. The inspection vehicle starts, the magnetic adsorption wheel adheres to the metal wall, and the three-point magnetic guide wheel completes the initial positioning and attitude calibration. S2. The inspection vehicle is driven by dual motors at differential speed and moves at a constant speed along the annular cavity trajectory, fully covering the cavitation detection area. S3. The dual sampling devices operate synchronously, projecting four-step phase-shifting sinusoidal stripes, and the endoscope collects deformed stripes. S4. The system introduces a correction factor k(x,y) and a compensation coefficient ξ to complete the anti-high-reflection phase solution and eliminate metal reflection interference in the image; S5. Calculate the phase difference Δφ(x,y) between the surface to be measured and the reference surface, and obtain the true cavitation erosion height through depth-of-field compensation phase-height mapping; S6. Based on the camera's intrinsic and extrinsic parameters, the pixel coordinates are converted into global 3D coordinates to complete the single-module point cloud calculation; S7. Based on the pre-calibrated pose matrix, the point clouds of the two modules are unified to the global coordinate system, and the overlapping field of view is weighted and fused to eliminate stitching errors. S8. The inspection vehicle continuously executes the imaging-reconstruction-stitching process, and finally outputs a three-dimensional point cloud and morphological parameter report of the full cavity cavitation defect.

10. The method for detecting cavitation defects in turbine roofs using a miniature vision inspection vehicle according to claim 9, characterized in that: In S3, the expression for the fringe intensity corresponding to the projected four-step phase-shift sinusoidal fringes is: ； In the formula: It represents the stripe intensity distribution acquired by the endoscope at the nth phase shift, after being modulated by cavitation morphology and affected by metal reflection; n is 0, 1, 2, and 3 respectively; This is an adaptive correction factor for high reflectivity of metal, with a value of 0~1, which automatically suppresses the intensity weight of the high-reflectivity region of the metal mirror. Background light intensity in the cavity environment. For stripe contrast, The measured true phase is modulated by the cavitation morphology.

11. The method for detecting cavitation defects in turbine roofs using a miniature vision inspection vehicle according to claim 9, characterized in that: In step S4, the system introduces a correction factor k(x,y) and a compensation coefficient ξ. The specific calculation expression for the anti-high anti-packaging phase solution is as follows: ； In the formula: This is the phase offset compensation coefficient.

12. The method for detecting cavitation defects in turbine roofs using a miniature vision inspection vehicle according to claim 9, characterized in that: In step S5, the specific process of calculating the phase difference Δφ(x,y) between the surface to be measured and the reference surface, and obtaining the true height of cavitation erosion through depth-of-field compensation phase-height mapping is as follows: First, calculate the phase difference between the cavitation erosion surface to be tested and the standard reference plane: ； Construct a quadratic mapping model adapted to the nonlinear concave-convex morphology of cavitation erosion: ； In the formula: C is the 20mm narrow cavity depth of field compensation term, which is adapted to the short depth of field characteristics of miniature optical paths; Z 0 represents the fixed reference distance from the reference plane to the optical center of the camera; The measured cavitation erosion surface is the true height relative to the reference surface, and A and B are the system-specific phase-height calibration coefficients. In the calibration experiment, the depth of the camera coordinate system at the point to be measured along the Z-axis is: ； By combining pre-calibrated camera intrinsic parameters, distortion coefficients, and camera extrinsic pose matrices, and through the geometric relationships of pinhole perspective imaging, the entire process of transforming two-dimensional pixel coordinates to local three-dimensional coordinates of the camera and then to global three-dimensional coordinates of the field is completed step by step, accurately solving the true spatial three-dimensional coordinates of each defect pixel. X w , Y w , Z w ).

13. The method for detecting cavitation defects in turbine roofs using a miniature vision inspection vehicle according to claim 9, characterized in that: In step S7, based on the pre-calibrated pose matrix, the point clouds of the two modules are unified to the global coordinate system, and the overlapping field of view is weighted and fused to eliminate stitching errors. The specific process is as follows: S7.1 Place the high-precision chessboard calibration board within the common field of view of the dual modules and acquire multiple sets of pose images; S7.2 Simultaneously calculate the camera intrinsic and extrinsic parameters of the left and right acquisition modules to obtain the rotation matrix of the left module to the world coordinate system. R L Translation vector T L Rotation matrix from the right module to the world coordinate system R R Translation vector T R ; S7.3 Construct the relative pose transformation matrix between the two modules: ； In the formula: R LR This is the rotation matrix from the left module to the right module. T LR This is the translation vector from the left module to the right module. This matrix is ​​a fixed value and is permanently saved after assembly and calibration, so there is no need to repeat the calibration during the testing process. S7.

4. Convert the local 3D point clouds calculated by the left and right projection modules to the global world coordinate system: ； S7.5 For the overlapping field of view of the two modules, a weighted mean fusion algorithm is used to complete the precise registration of the point cloud, eliminating stitching gaps and errors; distance-weighted fusion is used for the point cloud in the overlapping area, and the original point cloud is directly retained in the non-overlapping area to eliminate stitching deviations caused by assembly errors and vibrations. ； In the formula: W L , W R Assign point cloud weights to the left and right modules.

Images