Regulating valve opening monitoring method and system based on image recognition

Abstract

The invention relates to the technical field of regulating valve monitoring, in particular to a regulating valve opening degree monitoring method and system based on image recognition, and the method comprises the following steps: S1, constructing a composite optical mark: machining annular grooves and longitudinal convex ridges which are distributed in a staggered manner on the surface of a valve rod to form a composite mark belt; s2, anti-interference stereo imaging: collecting visible light and short wave infrared images to generate a fusion point cloud set; s3, dynamic shielding compensation: generating a three-dimensional point cloud model of a complete composite marking band; s4, thermal expansion decoupling calculation: visual displacement errors caused by thermal expansion are eliminated; and S5, time-space domain closed-loop calibration: reversely reckoning the opening value by using the structured light three-dimensional coordinate to carry out cross validation. According to the invention, high-precision, high-robustness and verifiability monitoring of the opening degree of the regulating valve is realized, and the measurement stability and engineering applicability under complex working conditions are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of regulating valve monitoring, and in particular to a regulating valve opening monitoring method and system based on image recognition. Background Art

[0002] As a key actuator for regulating fluid flow, pressure, and temperature in industrial process control systems, the precise acquisition of the opening status of the control valve is of great significance for maintaining stable operating conditions and optimizing energy consumption distribution. Traditional opening monitoring methods mostly rely on mechanically connected potentiometers or angle encoders, which have problems such as complex wiring, limited installation space, susceptibility to vibration interference, and poor reliability. With the development of non-contact measurement technology, the opening monitoring method based on image recognition and three-dimensional reconstruction has gradually attracted attention. It has the advantages of flexible layout, non-contact, and large amount of information, and has become an important direction for realizing intelligent equipment status perception.

[0003] Existing methods for opening monitoring based on image recognition generally have technical bottlenecks such as limited accuracy, sensitivity to occlusion interference, and poor environmental adaptability. On the one hand, the valve stem surface lacks structured features, resulting in poor robustness of image tracking and feature recognition, and is prone to misjudgment due to lighting changes or image blur. On the other hand, occlusion (such as occlusion by handwheel spokes) will cause mark loss, seriously affecting the continuity of three-dimensional morphology reconstruction. In addition, the visual displacement error caused by thermal expansion during valve operation is not effectively decoupled in most schemes, resulting in deviations in measurement results. More importantly, current methods generally lack a closed-loop calibration mechanism for jump errors, and cannot autonomously verify and correct the recognition results under abnormal circumstances, affecting the overall stability and industrial availability of the system. Summary of the Invention

[0004] The present invention provides a method and system for monitoring the opening of a regulating valve based on image recognition, which realizes high-precision, strong robustness, and real-time monitoring of the opening of the regulating valve. It can operate stably in complex industrial environments and has extremely high engineering application value.

[0005] A method for monitoring the opening of a regulating valve based on image recognition comprises the following steps:

[0006] S1, composite optical marking construction: annular grooves and longitudinal ridges are machined on the surface of the valve stem in staggered distribution, the grooves are filled with infrared absorbing material, and the ridges are coated with fluorescent tracers to form a composite marking band;

[0007] S2, anti-interference stereo imaging: A line structured light projector arranged at a 45° angle works in conjunction with a high-speed CMOS camera to calculate the three-dimensional topography of the valve stem surface through the phase offset of the structured light fringes. Visible light and short-wave infrared images are simultaneously collected to generate a fused point cloud.

[0008] S3, dynamic occlusion compensation: Based on the fused point cloud set, the local curvature features of the composite marker belt are extracted. When the handwheel spoke occlusion is detected, the occlusion area curvature interpolation algorithm is used to reconstruct the topological structure of the occluded marker and generate a 3D point cloud model of the complete composite marker belt;

[0009] S4, thermal expansion decoupling calculation: Based on the 3D point cloud model of the complete composite marker tape, the deformation of the composite marker tape is decomposed into an axial displacement component and a radial thermal expansion component to eliminate the visual displacement error caused by thermal expansion;

[0010] S5, spatiotemporal closed-loop calibration: Wavelet denoising is performed on the axial displacement components of N consecutive frames. When the displacement difference between adjacent frames exceeds a threshold, the reverse projection verification mechanism is triggered, and the opening value is reversely calculated using the structured light 3D coordinates for cross-validation.

[0011] Optionally, the composite optical marker construction in S1 includes:

[0012] S11, precision turning: Use a CNC lathe to machine multiple annular grooves on the valve stem surface, with the groove bottom forming a 120° V-shaped structure, and longitudinal ridges between adjacent annular grooves;

[0013] S12, surface microstructure treatment: laser cladding treatment is performed on the inner wall of the groove to form a serrated surface texture, and ion beam sputtering technology is used to deposit a nanoscale yttrium oxide (Y2O3) coating on the surface of the ridge;

[0014] S13, filling with infrared absorbing material: Silicon carbide (SiC) powder and epoxy resin are mixed in a predetermined mass ratio (3:1) to form a slurry, which is injected into the groove and cured at room temperature, and then subjected to secondary curing in a vacuum environment at 80°C;

[0015] S14, fluorescent tracer coating: strontium aluminate (SrAl2O4) based Eu 3+ Doped phosphor and high-temperature resistant silica gel are mixed in a predetermined volume ratio (1:5), and a fluorescent layer is evenly coated on the ridge surface using micro-dispensing technology, which is then cured by UV light to form a peeling-resistant coating;

[0016] S15, spatial frequency calibration: Use white light interferometer to scan and measure the surface of the marker tape to confirm that the spacing change between adjacent grooves satisfies the linear gradient distribution, and the spacing difference range is Δs∈[0.1mm,0.3mm]. Perform Fourier transform spectrum analysis on the marker tape structure to verify that it has a spatial frequency range of f∈[0.5,2.0]mm -1 There are identifiable characteristic peaks.

[0017] Optionally, the anti-interference stereoscopic imaging in S2 includes:

[0018] S21, multi-band structured light calibration: Line structured light projectors with wavelengths of λ1 = 808nm and λ2 = 1550nm are symmetrically arranged on both sides of the valve stem at an oblique angle of 45°. The optical axis of the high-speed CMOS camera forms an elevation angle of 30° with the axis of the valve stem. Calibration is performed using a checkerboard calibration plate to determine the geometric relationship between the projector and the camera, establish the projection matrix P, and the camera intrinsic parameter matrix K;

[0019] S22, dual-frequency phase offset calculation: project a sinusoidal fringe pattern onto the valve stem surface, collect a deformed fringe image, and perform a windowed Fourier transform on the collected deformed fringe image to calculate the phase distribution;

[0020] S23, 3D topography reconstruction: Dual-frequency phase unwrapping algorithm is used to solve the absolute phase Φ m , and calculate the three-dimensional coordinates based on the phase-depth mapping model;

[0021] S24, Multispectral Image Registration: For Visible Light Image I v With shortwave infrared image I ir Perform affine transformation;

[0022] S25, point cloud fusion generation: Map the visible light texture and infrared radiation intensity to the three-dimensional point cloud to obtain the fused point cloud set P fused .

[0023] Optionally, the dynamic occlusion compensation in S3 includes:

[0024] S31, curvature field calculation: perform local surface fitting on the point cloud data of the composite marker band in the fused point cloud set and calculate the principal curvature k l (p i );

[0025] S32, occlusion area detection: setting the curvature mutation threshold Δk th , when |κ1(p i )-k1(p i+1 )|>Δk th When the area is judged as an occlusion area, k1(p i ) is point p i The principal curvature value at k1(p i+1 ) is point p i+1 The principal curvature value at ;

[0026] S33, curvature field interpolation: extract the occlusion area boundary point set B = {b j}, where b j Take the jth 3D point in the boundary point set and construct a cubic Bezier surface for interpolation fitting to generate a 3D Bezier surface S(u,v);

[0027] S34, topology reconstruction: discretize the 3D Bezier surface S(u,v) into a point cloud set P rec And taking the point cloud set as the optimization variable, the energy minimization objective function including the Laplace operator is introduced;

[0028] S35, data fusion output: merge the reconstructed point cloud with the original unoccluded area to obtain a three-dimensional point cloud model of the complete composite marker band.

[0029] Optionally, the thermal expansion decoupling calculation in S4 includes:

[0030] S41, total strain tensor calculation: Based on the complete composite marker band 3D point cloud model after occlusion compensation and its initial calibration state, the local displacement field is constructed by comparing the displacement difference between the current coordinates and the initial coordinates of each point, and the total strain tensor of each point is calculated;

[0031] S42, frequency domain filtering separation: The total strain tensor is mapped to the frequency domain, and a band-stop filter is applied to isolate the thermal expansion component concentrated in the low-frequency band, thereby achieving decoupling of the thermal response from the actual structural deformation;

[0032] S43, axial displacement correction: Based on the results of frequency domain filtering separation, the influence of thermal expansion is removed from the total strain tensor, the mechanical strain is extracted, and integrated along the center line of the composite marker band to finally obtain the true axial displacement value.

[0033] Optionally, the total strain tensor calculation in S41 includes:

[0034] S411, point cloud registration and displacement vector construction: select the complete composite marker band 3D point cloud model after occlusion compensation and pair it with its corresponding initial calibration point cloud model to construct a local displacement vector field;

[0035] S412, local neighborhood gradient estimation: For each point r k , construct its k-nearest neighbor point set Estimate the displacement field gradient tensor in the neighborhood by least squares fitting

[0036] S413, total strain tensor construction: For each point r k , calculate the total strain tensor based on the displacement gradient

[0037] Optionally, the frequency domain filtering separation in S42 includes:

[0038] S421, Time series construction of total strain tensor: For each point r in the complete composite marker 3D point cloud model k , construct the total strain tensor sequence on continuous time frames;

[0039] S422, Fourier transform mapping to frequency domain: performing discrete Fourier transform (DFT) component by component on the strain tensor sequence of each frame point to map the strain signal from the time domain to the frequency domain;

[0040] S423, Band-stop filter construction and application: Construct a band-stop filter transfer function and apply it to the frequency domain strain signal to generate the thermal expansion frequency domain component;

[0041] S424, Time domain recovery of thermal expansion strain: Perform inverse Fourier transform on the frequency domain component of thermal expansion to recover the time domain thermal strain tensor sequence.

[0042] Optionally, the axial displacement correction in S43 includes:

[0043] S431, Mechanical strain tensor extraction: Calculate the mechanical strain tensor of each point based on the frequency domain filtering results

[0044] S432, axial component extraction: extract each point in the valve stem axial unit direction z dir The strain projection component on ;

[0045] S433, calculate the true axial displacement along the centerline integral: Calculate the true axial displacement.

[0046] Optionally, the spatiotemporal domain closed-loop calibration in S5 includes:

[0047] S51, multi-frame axial displacement sequence construction and wavelet denoising: the axial displacement value of the k-th observation point in N consecutive frames is denoised using discrete wavelet transform (DWT);

[0048] S52, displacement mutation detection: calculate the displacement difference of adjacent frames after noise reduction like This triggers back-projection cross-validation, where θ L is the mutation determination threshold;

[0049] S53, back-projection verification mechanism: Let the current frame structured light 3D reconstruction point be x (t) =(x,y,z), using the inverse modeling function Estimated theoretical opening value and the measured value ΔL (t) Perform cross validation and calculate the difference between measured and theoretical displacements like If the value is less than 0, it is considered as a valid measurement, otherwise it is recorded as an abnormal frame and marked or corrected, where ε is the cross-validation error tolerance.

[0050] A regulating valve opening monitoring system based on image recognition is used to implement the above-mentioned regulating valve opening monitoring method based on image recognition, and includes the following modules:

[0051] Composite marking building module: a composite optical marking band is formed on the surface of the valve stem. The composite optical marking band is composed of staggered infrared light-absorbing grooves and fluorescent coating ridges.

[0052] Stereo imaging and fusion module: Through the oblique arrangement of line structured light projectors and high-speed CMOS cameras, multi-band images are acquired and fused point clouds are generated to reconstruct the three-dimensional shape of the valve stem;

[0053] Occlusion compensation module: Analyzes the curvature distribution of the composite marker band based on the fused point cloud set, and uses interpolation and reconstruction methods to generate a complete 3D point cloud model of the composite marker band when an occlusion area is detected;

[0054] Thermal expansion decoupling module: decomposes the deformation of the composite optical marker point cloud into axial displacement and thermal expansion components, and eliminates thermal response errors;

[0055] Closed-loop calibration module: performs wavelet noise reduction on N consecutive frames of axial displacement, and reversely infers the opening of the structured light 3D coordinates through the inverse modeling function when the displacement changes suddenly, realizing real-time cross-validation and error control.

[0056] Beneficial effects of the present invention:

[0057] The present invention significantly enhances the traceability and robustness of image recognition by constructing a composite optical marker tape with a regular spatial frequency distribution. It adopts multi-band structured light and short-wave infrared image fusion technology to effectively suppress the influence of environmental interference on image quality. Combining curvature interpolation and Bezier surface reconstruction algorithms, it achieves high-precision topological repair of occluded areas, ensuring the continuity and integrity of monitoring data. Furthermore, the thermal expansion decoupling mechanism can accurately separate the non-real displacement error caused by ambient temperature rise, thereby improving the measurement reliability of the system in high-temperature scenarios.

[0058] The present invention can effectively filter out erroneous signals caused by occasional disturbances such as image edge jitter and occlusion jump through the combined judgment based on wavelet denoising and displacement mutation detection. When abnormal behavior is detected, the system calculates the theoretical opening value in real time through the reverse geometric modeling function and performs cross-validation to realize automatic identification and correction of abnormal frames. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0060] Figure 1 Schematic diagram of the monitoring method flow in an embodiment of the present invention;

[0061] Figure 2 Schematic diagram of system function modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0063] like Figure 1 As shown, a method for monitoring the opening of a regulating valve based on image recognition includes the following steps:

[0064] S1, composite optical marking construction: annular grooves and longitudinal ridges are machined on the surface of the valve stem, the grooves are filled with infrared absorbing materials, and the ridges are coated with fluorescent tracers to form a composite marking band with distinguishable spatial frequency;

[0065] S2, anti-interference stereo imaging: A line structured light projector arranged at a 45° angle works in conjunction with a high-speed CMOS camera to calculate the three-dimensional topography of the valve stem surface through the phase offset of the structured light fringes. Visible light and short-wave infrared images are simultaneously collected to generate a fused point cloud.

[0066] S3, dynamic occlusion compensation: Based on the fused point cloud set, the local curvature features of the composite marker belt are extracted. When the handwheel spoke occlusion is detected, the occlusion area curvature interpolation algorithm is used to reconstruct the topological structure of the occluded marker and generate a 3D point cloud model of the complete composite marker belt;

[0067] S4, thermal expansion decoupling calculation: Based on the 3D point cloud model of the complete composite marker tape, the deformation of the composite marker tape is decomposed into an axial displacement component and a radial thermal expansion component to eliminate the visual displacement error caused by thermal expansion;

[0068] S5, spatiotemporal closed-loop calibration: Wavelet denoising is performed on the axial displacement components of N consecutive frames. When the displacement difference between adjacent frames exceeds a threshold, the reverse projection verification mechanism is triggered, and the opening value is reversely calculated using the structured light 3D coordinates for cross-validation.

[0069] The composite optical labeling construct in S1 includes:

[0070] S11, precision turning: Use a CNC lathe to machine multiple annular grooves on the valve stem surface. The groove depth is d = 0.25 ± 0.02 mm, and the groove bottom has a 120° V-shaped structure. A longitudinal ridge is machined between adjacent annular grooves. The ridge height is h = 0.15 ± 0.01 mm, and the ridge top width is w = 0.10 mm.

[0071] S12, surface microstructure treatment: laser cladding treatment is performed on the inner wall of the groove to form a serrated texture with a surface roughness Ra of Ra ≥ 3.2 μm, and ion beam sputtering technology is used to deposit a nanoscale yttrium oxide (Y2O3) coating on the surface of the ridge;

[0072] S13, filling with infrared absorbing material: Silicon carbide (SiC) powder and epoxy resin are mixed in a predetermined mass ratio (3:1) to form a slurry, which is then injected into the groove and cured at room temperature. The slurry is then secondary cured in a vacuum environment at 80°C to ensure that the shape of the absorbing layer closely matches the groove and forms a dense coating.

[0073] S14, fluorescent tracer coating: strontium aluminate (SrAl2O4) based Eu 3+ Doped phosphor and high-temperature resistant silica gel are mixed in a predetermined volume ratio (1:5), and a fluorescent layer with a thickness of t = 50 ± 5 μm is evenly coated on the ridge surface using micro-dispensing technology, and then cured by ultraviolet light to form a peeling-resistant coating;

[0074] S15, spatial frequency calibration: Use white light interferometer to scan and measure the surface of the marker tape to confirm that the spacing change between adjacent grooves satisfies the linear gradient distribution, and the spacing difference range is Δs∈[0.1mm,0.3mm]. Perform Fourier transform spectrum analysis on the marker tape structure to verify that it has a spatial frequency range of f∈[0.5,2.0]mm -1 There are identifiable characteristic peaks.

[0075] Interference-resistant stereo imaging in S2 includes:

[0076] S21, multi-band structured light calibration: Line structured light projectors with wavelengths of λ1 = 808nm and λ2 = 1550nm are symmetrically arranged on both sides of the valve stem at an oblique angle of 45°. The optical axis of the high-speed CMOS camera forms an elevation angle of 30° with the axis of the valve stem. Calibration is performed using a checkerboard calibration plate to determine the geometric relationship between the projector and the camera, establish the projection matrix P, and the camera intrinsic parameter matrix K;

[0077] S22, dual-frequency phase offset calculation: Project a sinusoidal fringe pattern onto the valve stem surface, collect a deformed fringe image, and perform a windowed Fourier transform on the collected deformed fringe image to calculate the phase distribution, which is expressed as:

[0078]

[0079] in, is the modulation phase (m = 1, 2 corresponds to λ1, λ2 respectively), I(u+k,v) is the image grayscale value at pixel (u+k,v), T is the fringe period, and N = 5 is the window radius;

[0080] S23, 3D topography reconstruction: using dual-frequency phase unwrapping algorithm to solve the absolute phase Φ m , and calculate the three-dimensional coordinates according to the phase-depth mapping model, expressed as:

[0081]

[0082] Among them, round is the rounding function, z(u,v) is the depth of field (z axis) coordinate value at the pixel coordinate (u,v), that is, the height in the three-dimensional topography, ΔΦ m (u,v) is the phase change at the pixel coordinate (u,v), λ m is the wavelength of the structured light, θ = 45° is the projection angle of the structured light, that is, the angle between the structured light and the surface normal;

[0083] S24, Multispectral Image Registration: For Visible Light Image I v With shortwave infrared image I ir Perform an affine transformation, expressed as:

[0084]

[0085] Among them, A is the affine transformation matrix obtained by SIFT feature point matching;

[0086] S25, point cloud fusion generation: Map the visible light texture and infrared radiation intensity to the three-dimensional point cloud to obtain the fused point cloud set P fused , expressed as:

[0087] P fused ={(x,y,z)|(x,y,z)∈P3D ,c=α·I v (x,y)+β·I ir (x,y)};

[0088] Among them, P 3D is the point cloud set after 3D reconstruction, α = 0.6, β = 0.4 are the image fusion weights, and c is the composite intensity value used for color assignment or radiation analysis after fusion.

[0089] Dynamic occlusion compensation in S3 includes:

[0090] S31, curvature field calculation: perform local surface fitting on the point cloud data of the composite marker band in the fused point cloud set and calculate the principal curvature κ l (p i ), expressed as:

[0091]

[0092] Among them, p i is the coordinate of the i-th 3D point in the fused point cloud set, κ l (p i ) is point p i The lth principal curvature value at , respectively represents the minimum curvature and the maximum curvature, E i = <p u ,p u > is the inner product of the tangent vector of the point in the u direction, F i = <p u ,p v > is the inner product of the angle between the tangent vectors, G i = <p v ,p v > is the inner product of the tangent vector of the point in the v direction, p u 、p v is the local parameter direction tangent vector generated by the point cloud normal estimation, L i = <n,p uu > is the inner product of the unit normal vector and the second-order derivative in the u direction, M i = <n,p uv > is the inner product of the unit normal vector and the second-order derivative of the uv mixing direction, N i = <n,p vv > is the inner product of the unit normal vector and the second-order derivative in the v direction, where n is the point p i The unit normal vector of ;

[0093] S32, occlusion area detection: setting the curvature mutation threshold Δk th =0.15mm -1 , when |κ1(p i )-κ1(p i+1 )|>Δκth When , the area is judged as an occlusion area, where κ1(p i ) is point p i The principal curvature value at κ1(p i+1 ) is point p i+1 The principal curvature value at ;

[0094] S33, curvature field interpolation: extract the occlusion area boundary point set B = {b j}, where b j Take the jth 3D point in the boundary point set and construct a cubic Bezier surface for interpolation fitting to generate a 3D Bezier surface S(u,v), which is expressed as:

[0095]

[0096] Among them, S(u,v) is the three-dimensional Bezier surface point in the parameter space (u,v), B m′,3 (u), B n′,3 (v) is the Bernstein basis function, which is used to generate the weight coefficient of the Bezier surface. c m′,n′ is the (m′,n′)th control point in the control point matrix;

[0097] S34, topology reconstruction: discretize the 3D Bezier surface S(u,v) into a point cloud set P rec , and take the point cloud set as the optimization variable, introduce the energy minimization objective function including the Laplace operator, and the objective function is expressed as:

[0098]

[0099] Among them, ||P rec -S(u,v)|| 2 P rec The square of the Euclidean distance between each point in and the corresponding point on the Bezier surface, is the Laplace smoothing operator for reconstructing the point cloud, μ = 0.7 is the weight factor between the smoothing term and the fitting term;

[0100] S35, data fusion output: merge the reconstructed point cloud with the original unoccluded area to obtain a 3D point cloud model of the complete composite marker band, expressed as:

[0101] P complete =P mark ∪P rec ;

[0102] Among them, P complete is the final generated 3D point cloud model of the complete composite marker band, P markis the point cloud subset of the original unoccluded area in the composite marker band, from the fused point cloud set P fused It is extracted from the image and represents the area with clear observation and no compensation required.

[0103] Thermal expansion decoupling calculations in S4 include:

[0104] S41, total strain tensor calculation: Based on the complete composite marker band 3D point cloud model after occlusion compensation and its initial calibration state, the local displacement field is constructed by comparing the displacement difference between the current coordinates and the initial coordinates of each point, and the total strain tensor of each point is calculated;

[0105] S42, frequency domain filtering separation: The total strain tensor is mapped to the frequency domain, and a band-stop filter is applied to isolate the thermal expansion component concentrated in the low-frequency band, thereby achieving decoupling of the thermal response from the actual structural deformation;

[0106] S43, axial displacement correction: Based on the results of frequency domain filtering separation, the influence of thermal expansion is removed from the total strain tensor, the mechanical strain is extracted, and integrated along the center line of the composite marker band to finally obtain the true axial displacement value.

[0107] The total strain tensor calculation in S41 includes:

[0108] S411, point cloud registration and displacement vector construction: select the complete composite marker band 3D point cloud model after occlusion compensation and pair it with its corresponding initial calibration point cloud model to construct a local displacement vector field, which is expressed as:

[0109] P complete ={r k =(x k ,y k ,z k )};

[0110]

[0111] Among them, P ref is the initial calibration point cloud model, r k =(x k ,y k ,z k ) is the position of the kth point in the current frame, is the position of the corresponding point in the initial frame, v k is the displacement vector of the kth point, (u k ,v k ,w k ) is the three-dimensional displacement vector of the k-th point;

[0112] S412, local neighborhood gradient estimation: For each point r k , construct its k-nearest neighbor point set in, is the neighborhood point set of the kth point, and the displacement field gradient tensor is estimated in the neighborhood by least squares fitting Expressed as:

[0113]

[0114] S413, total strain tensor construction: For each point r k , calculate the total strain tensor based on the displacement gradient Expressed as:

[0115]

[0116] in, is the transposed matrix of the displacement gradient tensor.

[0117] Frequency domain filtering separation in S42 includes:

[0118] S421, Time series construction of total strain tensor: For each point r in the complete composite marker 3D point cloud model k , construct the total strain tensor sequence on continuous time frames, expressed as:

[0119]

[0120] in, are points r k The total strain tensor in frames 1, 2, ..., T, where T is the total number of time frames, For point r k Total strain tensor time series;

[0121] S422, Fourier transform mapping to frequency domain: Perform discrete Fourier transform (DFT) on the strain tensor sequence of each frame point component by component, and map the strain signal from the time domain to the frequency domain, which is expressed as:

[0122]

[0123] in, For point r k The Fourier coefficient of the (i, j)th component of the strain tensor at frequency f is, is the (i, j)th component of the strain tensor of the tth frame;

[0124] S423, Bandstop Filter Construction and Application: Construct a bandstop filter transfer function and apply it to the frequency domain strain signal to generate the thermal expansion frequency domain component, expressed as:

[0125]

[0126] Where H(f) is the filter transfer function, is the frequency domain component of thermal expansion (filtering result), f c is the filter cutoff frequency, rect is the rectangular window function,

[0127] S424, Time domain recovery of thermal expansion strain: Perform inverse Fourier transform on the frequency domain component of thermal expansion to recover the time domain thermal strain tensor sequence, expressed as:

[0128]

[0129] in, is the thermal strain tensor component of point k in the tth frame.

[0130] Axial displacement correction in S43 includes:

[0131] S431, Mechanical strain tensor extraction: Calculate the mechanical strain tensor of each point based on the frequency domain filtering results Expressed as:

[0132]

[0133] in, is the total strain tensor of the point in the tth frame, is the thermal expansion strain tensor corresponding to the same position;

[0134] S432, axial component extraction: extract each point in the valve stem axial unit direction z dir The strain projection component on is expressed as:

[0135]

[0136] in, is the axial mechanical strain component of point p in the tth frame, is the mechanical strain tensor of the corresponding point, z dir is the unit vector in the direction of the valve stem centerline, z dir The transpose of

[0137] S433, calculate the true axial displacement along the centerline integral: On the axial plane, the true axial displacement is calculated as:

[0138]

[0139] Where, ΔL (t) is the true axial displacement value at the t frame, ds: The differential arc length unit on , is the axial strain component at the corresponding point.

[0140] The spatiotemporal closed-loop calibration in S5 includes:

[0141] S51, multi-frame axial displacement sequence construction and wavelet denoising: The axial displacement value of the k-th observation point in N consecutive frames is denoised using discrete wavelet transform (DWT), which is expressed as:

[0142]

[0143] in, is the original axial displacement time series, DWT is the wavelet transform operation, SoftThresh is the soft threshold filter, λ L =0.15mm is the noise reduction threshold, is the axial displacement time series after noise reduction, IDWT is the inverse wavelet transform function;

[0144] S52, displacement mutation detection: calculate the displacement difference of adjacent frames after noise reduction like This triggers back-projection cross-validation, where θ L =0.5mm is the mutation determination threshold, expressed as:

[0145]

[0146] Where, ΔL (t+1) is the true axial displacement value of the t+1 frame, ΔL (t) is the true axial displacement value of the t-th frame;

[0147] S53, back-projection verification mechanism: Let the current frame structured light 3D reconstruction point be x (t) =(x,y,z), using the inverse modeling function Estimated theoretical opening value and the measured value ΔL (t) Perform cross validation and calculate the difference between measured and theoretical displacements like If the value is not equal to 0, it is considered as a valid measurement. Otherwise, it is recorded as an abnormal frame and marked or corrected. Here, ε = 0.2 is the cross-validation error tolerance, which is expressed as:

[0148]

[0149] Among them, a x 、a y 、a z is the linear mapping coefficient from the spatial component of the structured light system to the aperture, and b is the regression bias term.

[0150] like Figure 2As shown, a regulating valve opening monitoring system based on image recognition is used to implement the above-mentioned regulating valve opening monitoring method based on image recognition, including the following modules:

[0151] Composite marking building module: a composite optical marking band is formed on the surface of the valve stem. The composite optical marking band is composed of staggered infrared light-absorbing grooves and fluorescent coating ridges.

[0152] Stereo imaging and fusion module: Through the oblique arrangement of line structured light projectors and high-speed CMOS cameras, multi-band images are acquired and fused point clouds are generated to reconstruct the three-dimensional shape of the valve stem;

[0153] Occlusion compensation module: Analyzes the curvature distribution of the composite marker band based on the fused point cloud set, and uses interpolation and reconstruction methods to generate a complete 3D point cloud model of the composite marker band when an occlusion area is detected;

[0154] Thermal expansion decoupling module: decomposes the deformation of the composite optical marker point cloud into axial displacement and thermal expansion components, and eliminates thermal response errors;

[0155] Closed-loop calibration module: performs wavelet noise reduction on N consecutive frames of axial displacement, and reversely infers the opening of the structured light 3D coordinates through the inverse modeling function when the displacement changes suddenly, realizing real-time cross-validation and error control.

[0156] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0157] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for monitoring the opening of a regulating valve based on image recognition, characterized in that: The following steps are involved: S1, composite optical marking construction: annular grooves and longitudinal ridges are machined on the surface of the valve stem in staggered distribution, the grooves are filled with infrared absorbing material, and the ridges are coated with fluorescent tracers to form a composite marking band; S2, anti-interference stereo imaging: A line structured light projector arranged at a 45° angle works in conjunction with a high-speed CMOS camera to calculate the three-dimensional topography of the valve stem surface through the phase offset of the structured light fringes. Visible light and short-wave infrared images are simultaneously collected to generate a fused point cloud. S3, dynamic occlusion compensation: Based on the fused point cloud set, the local curvature features of the composite marker belt are extracted. When the handwheel spoke occlusion is detected, the occlusion area curvature interpolation algorithm is used to reconstruct the topological structure of the occluded marker and generate a 3D point cloud model of the complete composite marker belt; S4, thermal expansion decoupling calculation: Based on the 3D point cloud model of the complete composite marker tape, the deformation of the composite marker tape is decomposed into an axial displacement component and a radial thermal expansion component to eliminate the visual displacement error caused by thermal expansion; S5, spatiotemporal closed-loop calibration: Wavelet denoising is performed on the axial displacement components of N consecutive frames. When the displacement difference between adjacent frames exceeds a threshold, the reverse projection verification mechanism is triggered, and the opening value is reversely calculated using the structured light 3D coordinates for cross-validation.

2. The method for monitoring the opening of a regulating valve based on image recognition according to claim 1, characterized in that: The composite optical labeling construction in S1 includes: S11, precision turning: Use a CNC lathe to machine multiple annular grooves on the valve stem surface, with the groove bottom in a V-shaped structure, and longitudinal ridges between adjacent annular grooves; S12, surface microstructure treatment: laser cladding treatment is performed on the inner wall of the groove to form a serrated surface texture, and ion beam sputtering technology is used to deposit a nanoscale yttrium oxide coating on the surface of the ridge; S13, filling with infrared absorbing material: mixing silicon carbide powder and epoxy resin in a predetermined mass ratio to form a slurry, injecting the slurry into the groove and curing it at room temperature, and performing secondary curing in a vacuum environment; S14, fluorescent tracer coating: strontium aluminate-based Eu 3+ Doped phosphor and high-temperature resistant silica gel are mixed in a predetermined volume ratio, and a fluorescent layer is evenly coated on the ridge surface using micro-dispensing technology, which is then cured by UV light to form a peeling-resistant coating; S15, spatial frequency calibration: Use white light interferometer to scan and measure the surface of the marker tape to confirm that the spacing change between adjacent grooves satisfies the linear gradient distribution, and the spacing difference range is Δs∈[0.1mm, 0.3mm]. Perform Fourier transform spectrum analysis on the marker tape structure to verify that it has a spatial frequency range of f∈[0.5, 2.0]mm. -1 There are identifiable characteristic peaks.

3. The method for monitoring the opening of a regulating valve based on image recognition according to claim 1, characterized in that: The anti-interference stereoscopic imaging in S2 includes: S21, multi-band structured light calibration: Line structured light projectors with wavelengths of λ1 = 808nm and λ2 = 1550nm are symmetrically arranged on both sides of the valve stem at an oblique angle of 45°. The optical axis of the high-speed CMOS camera forms an elevation angle of 30° with the axis of the valve stem. Calibration is performed using a checkerboard calibration plate to determine the geometric relationship between the projector and the camera, establish the projection matrix P, and the camera intrinsic parameter matrix K; S22, dual-frequency phase offset calculation: project a sinusoidal fringe pattern onto the valve stem surface, collect a deformed fringe image, and perform a windowed Fourier transform on the collected deformed fringe image to calculate the phase distribution; S23, 3D shape reconstruction: using dual-frequency phase unwrapping algorithm to solve the absolute phase φ m , and calculate the three-dimensional coordinates based on the phase-depth mapping model; S24, Multispectral Image Registration: For Visible Light Image I v With shortwave infrared image I ir Perform affine transformation; S25, point cloud fusion generation: Map the visible light texture and infrared radiation intensity to the three-dimensional point cloud to obtain the fused point cloud set P fused .

4. The method for monitoring the opening of a regulating valve based on image recognition according to claim 3, characterized in that: The dynamic occlusion compensation in S3 includes: S31, curvature field calculation: perform local surface fitting on the point cloud data of the composite marker band in the fused point cloud set and calculate the principal curvature κ l (p i ); S32, occlusion area detection: setting the curvature mutation threshold Δκ th , when |κ1(p i )-κ1(p i+1 )|>Δκ th When , the area is judged as an occlusion area, where κ1(p i ) is point p i The principal curvature value at κ1(p i+1 ) is point p i+1 The principal curvature value at ; S33, curvature field interpolation: extract the occlusion area boundary point set B = {b j }, where b j Take the jth 3D point in the boundary point set and construct a cubic Bezier surface for interpolation fitting to generate a 3D Bezier surface S(u, v); S34, topology reconstruction: discretize the 3D Bezier surface S(u, v) into a point cloud set P rec And taking the point cloud set as the optimization variable, the energy minimization objective function including the Laplace operator is introduced; S35, data fusion output: merge the reconstructed point cloud with the original unoccluded area to obtain a three-dimensional point cloud model of the complete composite marker band.

5. The method for monitoring the opening of a regulating valve based on image recognition according to claim 4, characterized in that: The thermal expansion decoupling calculation in S4 includes: S41, total strain tensor calculation: Based on the complete composite marker band 3D point cloud model after occlusion compensation and its initial calibration state, the local displacement field is constructed by comparing the displacement difference between the current coordinates and the initial coordinates of each point, and the total strain tensor of each point is calculated; S42, frequency domain filtering separation: The total strain tensor is mapped to the frequency domain, and a band-stop filter is applied to isolate the thermal expansion component concentrated in the low-frequency band, thereby achieving decoupling of the thermal response from the actual structural deformation; S43, axial displacement correction: Based on the results of frequency domain filtering separation, the influence of thermal expansion is removed from the total strain tensor, the mechanical strain is extracted, and integrated along the center line of the composite marker band to finally obtain the true axial displacement value.

6. The method for monitoring the opening of a regulating valve based on image recognition according to claim 5, characterized in that: The total strain tensor calculation in S41 includes: S411, point cloud registration and displacement vector construction: select the complete composite marker band 3D point cloud model after occlusion compensation and pair it with its corresponding initial calibration point cloud model to construct a local displacement vector field; S412, local neighborhood gradient estimation: For each point r k , construct its k-nearest neighbor point set, and estimate the displacement field gradient tensor in the neighborhood by least squares fitting S413, total strain tensor construction: For each point r k , calculate the total strain tensor based on the displacement gradient 7. The method for monitoring the opening of a regulating valve based on image recognition according to claim 6, characterized in that: The frequency domain filtering separation in S42 includes: S421, Time series construction of total strain tensor: For each point r in the complete composite marker 3D point cloud model k , construct the total strain tensor sequence on continuous time frames; S422, Fourier transform mapping to frequency domain: performing discrete Fourier transform on the strain tensor sequence of each frame point component by component to map the strain signal from the time domain to the frequency domain; S423, Band-stop filter construction and application: Construct a band-stop filter transfer function and apply it to the frequency domain strain signal to generate the thermal expansion frequency domain component; S424, Time domain recovery of thermal expansion strain: Perform inverse Fourier transform on the frequency domain component of thermal expansion to recover the time domain thermal strain tensor sequence.

8. The method for monitoring the opening of a regulating valve based on image recognition according to claim 7, characterized in that: The axial displacement correction in S43 includes: S431, Mechanical strain tensor extraction: Calculate the mechanical strain tensor of each point based on the frequency domain filtering results S432, axial component extraction: extract each point in the valve stem axial unit direction z dir The strain projection component on ; S433, calculate the true axial displacement along the centerline integral: Calculate the true axial displacement.

9. The method for monitoring the opening of a regulating valve based on image recognition according to claim 1, characterized in that: The spatiotemporal closed-loop calibration in S5 includes: S51, multi-frame axial displacement sequence construction and wavelet denoising: the axial displacement value of the k-th observation point in N consecutive frames is denoised using discrete wavelet transform; S52, displacement mutation detection: calculate the displacement difference of adjacent frames after noise reduction like Then the back-projection cross-validation is triggered, where θ L is the mutation determination threshold; S53, back-projection verification mechanism: Let the current frame structured light 3D reconstruction point be x (t) =(x, y, z), using the inverse modeling function Estimated theoretical opening value and the measured value ΔL (t) Perform cross validation and calculate the difference between measured and theoretical displacements like If the value is less than 0, it is considered as a valid measurement, otherwise it is recorded as an abnormal frame and marked or corrected, where ε is the cross-validation error tolerance.

10. A regulating valve opening monitoring system based on image recognition, used to implement a regulating valve opening monitoring method based on image recognition according to any one of claims 1 to 9, characterized in that: Includes the following modules: Composite marking building module: a composite optical marking band is formed on the surface of the valve stem. The composite optical marking band is composed of staggered infrared light-absorbing grooves and fluorescent coating ridges. Stereo imaging and fusion module: Through the oblique arrangement of line structured light projectors and high-speed CMOS cameras, multi-band images are acquired and fused point clouds are generated to reconstruct the three-dimensional shape of the valve stem; Occlusion compensation module: Analyzes the curvature distribution of the composite marker band based on the fused point cloud set, and uses interpolation and reconstruction methods to generate a complete 3D point cloud model of the composite marker band when an occlusion area is detected; Thermal expansion decoupling module: decomposes the deformation of the composite optical marker point cloud into axial displacement and thermal expansion components, and eliminates thermal response errors; Closed-loop calibration module: performs wavelet noise reduction on N consecutive frames of axial displacement, and reversely infers the opening of the structured light 3D coordinates through the inverse modeling function when the displacement changes suddenly, realizing real-time cross-validation and error control.

Images