Method for detecting spatial position of steam turbine rotor during assembling process based on machine vision

Abstract

This invention relates to the field of assembly and inspection technology, and in particular to a machine vision-based method for detecting the spatial pose of a turbine rotor during assembly. The method includes installing and calibrating a vision imaging unit at a preset cross-section of the stator inner cylinder; acquiring images in real time during assembly, calculating the radial clearance and eccentricity vector of each cross-section, and constructing a comprehensive assembly quality index to evaluate the assembly quality; when the quality is substandard, selecting a warning or emergency stop detection mode based on the deviation coefficient; continuing to acquire images after an emergency stop, analyzing the radial clearance time-series data, constructing an emergency stop stability coefficient, and assessing the rotor sway risk; verifying the reliability of the detection state by comparing the consistency between the prediction before the emergency stop and the actual measurement after the emergency stop; and finally, adaptively optimizing the detection parameters based on the consistency deviation. This invention solves the problems of not being able to dynamically monitor the entire assembly process and the inability to capture the transient behavior of the rotor portion that has entered the stator after an emergency stop.

Description

Technical Field

[0001] This invention relates to the field of assembly and inspection technology, and in particular to a method for detecting the spatial pose of a steam turbine rotor during the assembly process based on machine vision. Background Technology

[0002] The installation of the turbine rotor is a critical process in the installation and overhaul of power generation equipment. The rotor needs to be slowly hoisted into the stator inner cylinder. Since the design gap between the rotor and the inner wall of the stator is usually only a few millimeters, any slight deviation in posture may cause the rotor to scrape against the stator, resulting in equipment damage. Currently, rotor orientation detection during the rotor mounting process mainly relies on mechanical point measurement or offline laser tracking in a stopped state, which has the following drawbacks: First, it is impossible to obtain dynamic-static clearance data in real time during continuous rotor advancement, and it can only "advance a point and stop once," which is inefficient and has measurement blind spots; Second, contact measurement requires the installation of sensors or targets on the rotor surface, which are easily scratched in narrow spaces, and the targets themselves increase the risk of mounting; Third, it can only obtain single-point data and cannot restore the spatial orientation of the rotor in the stator cavity, making it difficult to judge the overall tilt and the coordination of eccentricity of each section; Fourth, existing methods cannot capture the swaying process of the rotor due to inertia at the moment of emergency stop, and this transient behavior may cause the rotor to make instantaneous contact with the stator; Fifth, the detection results lack a self-verification mechanism, making it impossible to verify the reliability of the detection system itself, and it is difficult to detect and adjust in time when the detection parameters drift or the vision unit is displaced.

[0003] Therefore, there is an urgent need for an intelligent detection method that can achieve dynamic monitoring of the entire dressing process, capture transient behavior after sudden stop, self-verification of detection results, and adaptive optimization of parameters.

[0004] Chinese Patent Publication No. CN105526878A discloses a real-time dynamic measurement method for the radial clearance between a steam turbine rotor and stator. Its features include steps such as determining a ranging device for the radial clearance, determining the radial clearance between the rotor and stator, determining the distance from the emitter to the rotor surface, calculating the vertical eccentricity of the rotor when the rotor center shifts, calculating the lateral eccentricity of the rotor when the rotor center shifts, calculating the rotor eccentricity at any position, and calculating the radial clearance between the steam turbine rotor and stator. This method enables rapid and accurate dynamic measurement of the steam seal clearance of the turbine rotor and the blade tip steam seal clearance during operation, solving the problems of large measurement errors and the inability to accurately measure the radial clearance between the turbine rotor and stator dynamically online. This ensures the safe and economical operation of the unit and saves downtime for major overhauls.

[0005] Therefore, the real-time dynamic measurement method for the radial clearance between the turbine rotor and stator has the following problems: 1. This method requires installing a corner reflector on the rotor surface as a cooperative target, and measuring the distance by emitting modulated light to the reflector through a laser rangefinder. This measurement method is only suitable for rotating conditions during operation. However, in the narrow space of rotor installation, installing any external target will increase the risk of scratching, and the corner reflector itself is very easy to be scraped off by the inner wall of the stator during installation.

[0006] 2. This method is designed for steady-state measurement during operation and does not take into account the swaying process of the rotor due to inertia during an emergency stop. When an emergency stop is triggered during the fitting process, the instantaneous scraping that may be caused by this transient behavior is completely outside its detection blind zone. Summary of the Invention

[0007] To address this, the present invention provides a machine vision-based method for detecting the spatial pose of a turbine rotor during the installation process, thereby overcoming the problems in the prior art that it is impossible to dynamically monitor the entire installation process and that the transient behavior of the rotor portion that has entered the stator after an emergency stop cannot be captured.

[0008] To achieve the above objectives, this invention provides a machine vision-based method for spatial pose detection during the installation process of a steam turbine rotor. It includes: Several visual imaging units are fixedly installed at several predetermined cross-sections of the inner cylinder of the turbine stator, and the visual imaging units are calibrated. Establish the transformation relationship between the image coordinate system and the stator global coordinate system, and determine the pixel equivalent and reference datum position; During the dressing process, the visual imaging unit continuously acquires a sequence of images including the outer edge of the rotor and the reference reference of the inner wall of the stator, determines the real-time radial clearance vector and real-time eccentricity vector of the rotor at the current section, and determines the comprehensive dressing quality index to determine whether the dressing quality of the current dressing process meets the standard. Based on the clothing quality deviation coefficient, which represents the degree to which the current clothing quality deviates from the standard requirements under the current substandard clothing condition, determine whether to implement the early warning detection mode or the emergency stop detection mode. Under the condition of executing the emergency stop detection mode, the propulsion speed is reduced to zero, and the sequence images within a preset time after stopping are collected to obtain the radial clearance time change data after stopping, and an emergency stop stability coefficient is constructed to determine whether there is an instantaneous scraping risk or shaking risk to the stator after the part of the rotor that has entered the stator is stopped. Predict the expected stability coefficient under the current emergency stop state, and determine the consistency deviation coefficient between the predicted stability coefficient and the emergency stop stability coefficient, so as to determine whether the detection status of the current rotor dynamic mounting process meets the standard. Based on the difference between the consistency deviation coefficient and the preset consistency deviation coefficient, the weighting coefficient in the weighted calculation of the comprehensive clothing quality index is determined, the position of the calibrated visual imaging unit is adjusted, and the trigger preset value of the emergency stop detection mode is adjusted.

[0009] Furthermore, the process of determining whether the dressing quality of the current dressing process meets the standards includes, The real-time radial gap vectors are summarized, and the minimum value is selected as the global minimum gap value at the current moment; Based on the real-time eccentricity vector, determine the tilt angle of the rotor's spatial axis at the current moment; The real-time eccentricity vector is normalized to obtain the unit vector of eccentricity direction for each cross section, and the eccentricity consistency coefficient of each cross section is calculated. The overall dressing quality index is calculated by weighting the global minimum gap value, the spatial axis tilt angle, and the eccentricity consistency coefficient of each section. The comprehensive clothing quality index is compared with the preset comprehensive clothing quality index; Based on the fact that the overall dressing quality is less than the preset overall dressing quality index, it is determined that the dressing quality of the current dressing process is substandard.

[0010] Furthermore, the process of determining whether to execute the early warning detection mode or the emergency stop detection mode includes, Calculate the absolute difference between the comprehensive clothing quality index and the preset comprehensive clothing quality index; Divide the absolute difference by the preset comprehensive clothing quality index to obtain the clothing quality deviation coefficient; Based on the fact that the clothing quality deviation coefficient is less than or equal to the preset deviation coefficient, the early warning detection mode is determined to be executed. The warning detection mode maintains a normal image acquisition frame rate, visually identifies risk sections, increases data storage frequency, and continuously monitors the changing trend of the comprehensive clothing quality index. When the index continues to decline and the clothing condition continues to deteriorate, the warning level is automatically upgraded.

[0011] Furthermore, the process of determining whether to execute the early warning detection mode or the emergency stop detection mode also includes, Based on the fact that the clothing quality deviation coefficient is greater than the preset deviation coefficient, it is determined to execute the emergency stop detection mode. The emergency stop detection mode immediately outputs an emergency stop signal to stop the dressing process. The visual imaging unit automatically switches to a high frame rate acquisition mode and records the real-time radial gap vector, real-time eccentricity vector, and the comprehensive dressing quality index of each section at the moment the emergency stop is triggered.

[0012] Furthermore, the process of determining whether there is a risk of instantaneous scraping or shaking of the stator after an emergency stop of the rotor portion that has entered the stator includes, The radial gap time-series variation data are fitted to obtain the initial oscillation amplitude, oscillation attenuation coefficient, and oscillation frequency for each cross section; Extract the global minimum dynamic gap from the radial gap time-series variation data; The initial oscillation amplitude, the oscillation attenuation coefficient, and the global minimum dynamic gap are normalized to obtain the relative oscillation amplitude, the relative attenuation coefficient, and the gap safety factor. The relative oscillation amplitude is taken as the maximum value, the relative attenuation coefficient is taken as the minimum value, and the gap safety factor corresponding to the global minimum dynamic gap is taken as the overall gap safety factor. The three are weighted and summed to obtain the emergency stop stability coefficient. The emergency stop stability coefficient is compared with the preset emergency stop stability coefficient; Based on the fact that the emergency stop stability coefficient is greater than the preset emergency stop stability coefficient, it is determined that there is a risk of instantaneous scraping or shaking.

[0013] Furthermore, the process of determining whether the detection status of the current rotor dynamic mounting process meets the standard includes, Extract the comprehensive clothing quality index from multiple consecutive moments within a preset time period before the emergency stop to form a time-series data sequence; Linear regression analysis was performed on the time series data to fit the trend line of the comprehensive clothing quality index over time; The instantaneous comprehensive clothing quality index prediction value at the moment of emergency stop is calculated based on the trend line and the precise moment of emergency stop. The expected stability coefficient is obtained by mapping the instantaneous comprehensive clothing quality index prediction value. The consistency deviation coefficient is calculated based on the expected stability coefficient and the measured emergency stop stability coefficient. The consistency deviation coefficient is compared with the preset consistency deviation coefficient; Based on the fact that the consistency deviation coefficient is greater than the preset consistency deviation coefficient, it is determined that the detection status of the current rotor dynamic fitting process is substandard.

[0014] Furthermore, the process of determining and optimizing the detection parameters in the corresponding detection mode includes, Calculate the difference between the consistency deviation coefficient and the preset consistency deviation coefficient; The difference is compared with a preset difference. Based on the difference being less than or equal to a preset difference, the weighting coefficients in the comprehensive clothing quality index are determined and adjusted.

[0015] Furthermore, the process of determining and optimizing the detection parameters in the corresponding detection mode also includes, Based on the fact that the difference is greater than a preset difference, the position of the calibrated visual imaging unit is determined and the trigger preset value of the emergency stop detection mode is adjusted.

[0016] Furthermore, the process of adjusting the weighting coefficients in the weighted calculation of the comprehensive clothing quality index includes, Extract real-time eccentricity vector data of each section before the emergency stop occurs; Based on the real-time eccentricity vector of each section, the eccentricity consistency coefficient and the spatial axis tilt angle of each section at each moment are calculated to form two time series data sequences. Correlation analysis was performed on the two time-series data sequences to obtain the correlation coefficient values; The main source of the current deviation is determined based on the comparison between the correlation coefficient value and the preset correlation coefficient value. Based on the fact that the correlation coefficient is negative and its absolute value is greater than the preset correlation coefficient value, it is determined that the inconsistency of the eccentricity direction of each section and the increase of the tilt angle are negatively correlated. That is, the main source of the current deviation is the insufficient consistency of the eccentricity of each section, and the weight coefficient corresponding to the consistency factor is increased.

[0017] Furthermore, the process of adjusting the weighting coefficients in the weighted calculation of the comprehensive clothing quality index also includes, Based on the fact that the correlation coefficient is positive and the absolute value is greater than the preset correlation coefficient value, it is determined that the inconsistency of the eccentricity direction of each section is positively correlated with the increase of the tilt angle. That is, the main source of the current deviation is that the overall tilt angle is too large, so the weight coefficient corresponding to the tilt angle factor is increased.

[0018] Compared with existing technologies, the advantages of this invention are as follows: First, by installing vision units at a preset cross-section in the stator inner cylinder and establishing a global coordinate system, this invention achieves direct, continuous, and non-contact measurement of radial gaps and eccentric vectors across multiple cross-sections during the dressing process, solving the technical problem of blind spots caused by the need for stopping the machine for point measurement in traditional methods. Second, by constructing a comprehensive dressing quality index that integrates global minimum gap, spatial axis tilt angle, and cross-sectional eccentricity consistency, multi-dimensional pose features are quantified into a single evaluation index, achieving real-time quantitative assessment of dressing quality. Third, by introducing a dressing quality deviation coefficient, adaptive switching of the detection mode is achieved. When the quality is slightly substandard, a warning detection mode is activated for intensive monitoring; when the quality is severely substandard, an emergency stop detection mode is triggered and the image acquisition frame rate is automatically increased, thus improving the detection strategy. Fourth, by constructing an emergency stop stability coefficient and fitting the decaying oscillation of the radial gap time series data after stopping, a quantitative assessment of the rotor inertial sway during an emergency stop can be achieved, solving the problem that existing technologies cannot capture the transient behavior of the rotor that has entered the stator after an emergency stop and its impact on the stator; Fifth, by comparing the expected stability coefficient predicted by the comprehensive dressing quality index before an emergency stop with the measured emergency stop stability coefficient after an emergency stop, a consistency deviation coefficient is determined, and the detection results are self-verified, solving the problem that the reliability of the detection system itself cannot be verified; Finally, by adaptively adjusting the weight coefficient of the comprehensive dressing quality index or triggering the recalibration of the visual unit according to the degree of consistency deviation, the detection system can become more accurate with use, reducing the impact of error accumulation on the detection results.

[0019] Furthermore, this invention solves the problem of difficulty in fusing and comparing measurement results of different sections in traditional methods by installing a visual imaging unit at a preset cross-section of the stator inner cylinder and establishing a global coordinate system, thus unifying multi-section image data under the same spatial reference. By calibrating pixel equivalents and reference reference positions, it achieves accurate conversion from image pixels to physical dimensions, providing a reliable quantitative basis for subsequent millimeter-level gap measurements. During the fitting process, by real-time calculation of the radial gap vector and eccentricity vector of each section, and further extracting the global minimum gap value, spatial axis tilt angle, and cross-sectional eccentricity consistency coefficient, the original image data is transformed into data with clear... The precise physical orientation parameters enable operators to intuitively grasp the gap state and overall posture of various parts of the rotor. Based on this, by weighting and integrating the gap factor, tilt angle factor, and consistency factor into a comprehensive dressing quality index, the multi-dimensional orientation features are quantified into a single evaluation index. This index can comprehensively reflect the most dangerous gap state, overall tilt degree, and eccentricity coordination of each section during the dressing process. Furthermore, it enables the determination of dressing quality through preset values, providing an objective decision-making basis for subsequent risk assessment and adaptive switching of detection modes. This fundamentally solves the technical defects of traditional methods that rely on manual experience and lack unified evaluation standards.

[0020] Furthermore, this invention quantifies and classifies substandard conditions by constructing a clothing quality deviation coefficient, achieving dynamic matching between detection modes and risk levels: when the deviation coefficient is small, an early warning detection mode is activated, visually marking risk sections and increasing data storage frequency while maintaining normal data acquisition. It also continuously monitors the changing trend of the comprehensive clothing quality index, automatically raising the early warning level when the index continuously declines, allowing operators sufficient warning and preparation time before the risk escalates; when the deviation coefficient exceeds a threshold, an emergency stop detection mode is immediately triggered, automatically increasing the image acquisition frame rate to capture the transient shaking process that may occur after the emergency stop, and completely recording the key pose data at the moment of the emergency stop; based on this, By fitting the radial clearance time-series data after an emergency stop with damped oscillations, three key parameters are extracted: initial oscillation amplitude, oscillation damping coefficient, and global minimum dynamic clearance. An emergency stop stability coefficient is then constructed to quantitatively evaluate the dynamic behavior after an emergency stop. The emergency stop stability coefficient represents the maximum amplitude of the sway through the relative oscillation amplitude, the rate of sway attenuation through the relative damping coefficient, and the safety margin of the minimum clearance through the clearance safety factor. The weighted fusion of these three parameters is compared with a preset threshold, enabling quantitative detection and risk assessment of the inertial sway of a large rotor after an emergency stop. This solves the technical blind spot of existing technologies being completely unable to capture transient behavior after an emergency stop, and provides a key dynamic safety assessment method for sudden working conditions during the fitting process.

[0021] Furthermore, this invention performs self-verification of the detection results by calculating the consistency deviation coefficient, thereby achieving a quantitative assessment of the reliability of the detection system itself. By extracting the changing trend of the comprehensive clothing quality index before the emergency stop and extrapolating the predicted stability coefficient, it compares the predicted stability coefficient with the measured emergency stop stability coefficient after the emergency stop. When the deviation exceeds a preset value, the detection state is deemed substandard, thus identifying potential systemic problems in the detection process. Based on this, a differentiated optimization strategy is adaptively executed according to the degree of deviation: when the deviation is small, correlation analysis is used to accurately locate the source of the deviation as insufficient eccentricity or excessive tilt angle, and the weight coefficients in the comprehensive clothing quality index are adjusted accordingly, making subsequent detection more focused on the actual dominant risks; when the deviation is large, the vision unit self-calibration function is activated to determine whether the hardware has shifted or loosened. If there is no hardware problem, the trigger threshold of the emergency stop detection mode is automatically shifted to a more conservative direction to obtain more emergency stop data for verification. This enables the detection system to become more accurate with use, solving the technical defects of existing detection methods that cannot verify the reliability of their own results and cannot automatically adjust the detection strategy according to the deviation, significantly improving the robustness and adaptability of clothing detection under complex working conditions. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the steps of the machine vision-based spatial pose detection method for turbine rotor installation process according to an embodiment of the present invention. Figure 2 This is a logic block diagram of an embodiment of the present invention for determining whether to execute a warning detection mode or an emergency stop detection mode based on the clothing quality deviation coefficient; Figure 3 This is a logic block diagram of an embodiment of the present invention for determining whether there is a risk of instantaneous scraping or shaking of the rotor after an emergency stop based on the emergency stop stability coefficient; Figure 4 This is a logic block diagram illustrating how the detection parameters in the corresponding detection mode are optimized based on the difference between the consistency deviation coefficient and the preset consistency deviation coefficient, according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0024] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0025] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0026] Please see Figure 1 The diagram shown is a step-by-step illustration of the spatial pose detection method for the turbine rotor installation process based on machine vision, according to an embodiment of the present invention.

[0027] This invention provides a machine vision-based method for spatial pose detection during the installation of a steam turbine rotor, comprising: Step S1: Fix at least two visual imaging units at at least two preset cross-sections of the turbine stator inner cylinder, and calibrate the visual imaging units; Step S2: Establish the transformation relationship between the image coordinate system and the static global coordinate system, and determine the pixel equivalent and reference datum position; Step S3: During the rotor fitting process, the vision imaging unit is controlled to continuously acquire a sequence of images including the outer edge of the rotor and the reference reference of the inner wall of the stator, and to calculate the real-time radial clearance vector and real-time eccentricity vector of the rotor at this section in real time. The global minimum clearance value, the spatial axis tilt angle of the rotor and the eccentricity consistency coefficient of each section are calculated at the current moment to determine the comprehensive fitting quality index, so as to determine whether the fitting quality of the current fitting process meets the standard. If the standard is not met, the non-compliance detection mark is output and the comprehensive fitting quality index at the current moment is recorded. Step S4: Under the condition that the current attire does not meet the standard, determine the attire quality deviation coefficient, which represents the degree to which the current attire quality deviates from the standard requirement, based on the comprehensive attire quality index and the target comprehensive attire quality index, so as to determine the early warning detection mode or the emergency stop detection mode. Step S5: Under the condition of determining to execute the emergency stop detection mode, reduce the propulsion speed to zero, continue to collect the sequence images within a preset time after stopping, and calculate the radial gap time sequence change data after stopping. Fit the radial gap time sequence change data, extract the oscillation amplitude, attenuation coefficient and minimum dynamic gap, construct the emergency stop stability coefficient, and determine whether there is an instantaneous scraping risk or shaking risk to the stator after the part of the rotor that has entered the stator is stopped by the emergency stop based on the emergency stop stability coefficient. Step S6: Based on the changing trend of the comprehensive dressing quality index time series data within a preset period before the emergency stop, predict the expected stability coefficient under the current emergency stop state, calculate the consistency deviation coefficient between the expected stability coefficient and the emergency stop stability coefficient, and determine whether the detection state of the current rotor dynamic dressing process meets the standard based on the consistency deviation coefficient. Step S7: If the current detection status is determined to be substandard, the detection parameters in the corresponding detection mode are optimized based on the difference between the consistency deviation coefficient and the preset consistency deviation coefficient.

[0028] Specifically, this invention achieves direct, continuous, and non-contact measurement of radial gaps and eccentric vectors across multiple cross-sections during the dressing process by installing vision units at a preset cross-section within the stator cylinder and establishing a global coordinate system. This solves the technical problem of blind spots caused by the need for stopping the machine for point measurement in traditional methods. Secondly, by constructing a comprehensive dressing quality index that integrates global minimum gap, spatial axis tilt angle, and cross-sectional eccentricity consistency, multi-dimensional pose features are quantified into a single evaluation index, enabling real-time quantitative assessment of dressing quality. Thirdly, by introducing a dressing quality deviation coefficient, adaptive switching of the detection mode is achieved. A warning detection mode is activated for intensive monitoring when the quality is slightly substandard, and an emergency stop detection mode is triggered and the image acquisition frame rate is automatically increased when the quality is severely substandard, allowing the detection strategy to better align with risks. Fourth, by constructing an emergency stop stability coefficient and fitting the decaying oscillation of the radial gap time series data after stopping, a quantitative assessment of the rotor inertial sway during an emergency stop can be achieved, solving the problem that existing technologies cannot capture the transient behavior of the rotor that has entered the stator after an emergency stop and its impact on the stator; Fifth, by comparing the expected stability coefficient predicted by the comprehensive dressing quality index before an emergency stop with the measured emergency stop stability coefficient after an emergency stop, a consistency deviation coefficient is determined, and the detection results are self-verified, solving the problem that the reliability of the detection system itself cannot be verified; Finally, by adaptively adjusting the weight coefficient of the comprehensive dressing quality index or triggering the recalibration of the visual unit according to the degree of consistency deviation, the detection system can become more accurate with use, reducing the impact of error accumulation on the detection results.

[0029] Specifically, in steps S1 and S2, the preset cross-section is the front end shaft seal interface and the middle partition plate cross-section of the stator cavity of the high-pressure cylinder of the steam turbine, which are respectively used as the first preset cross-section and the second preset cross-section.

[0030] The vision imaging unit consists of two industrial cameras, each with a resolution of 5 megapixels, equipped with a 12mm fixed-focus lens and a ring LED light source, and installed in a sealed protective housing.

[0031] In this embodiment of the invention, the process of establishing the transformation relationship between the image coordinate system and the static global coordinate system, and determining the pixel equivalent and reference reference position, includes: Step S21: An observation hole is opened on the outer wall of the stator at the first preset section. The first visual imaging unit is fixed to the outside of the observation hole by a flange. The lens extends into the inner cavity of the stator through the observation hole. The camera posture is adjusted so that the optical axis is perpendicular to the rotor axis and parallel to the radial plane of the stator. At the second preset section, the second visual imaging unit is fixed to the inner wall of the stator by a special clamp using the gap of the partition structure. The optical axis direction is also adjusted to meet the measurement requirements. Step S22: Establish a global coordinate system for the stator with the mid-section of the stator cylinder as the XY plane, the intersection of the mid-section and the rotor design center line as the origin, and the direction from the rotor design center line to the generator end as the positive Z-axis. Arrange at least three non-collinear reference targets at each preset section and measure the precise coordinates of the targets in the global coordinate system using a laser tracker. Place a specially designed calibration target within the camera's field of view, and simultaneously measure the global coordinates of the target's feature points using a laser tracker, acquire the target image using a vision imaging unit, and extract the image coordinates of the feature points. Solve the rotation matrix and translation vector of each camera using the PnP algorithm to establish the transformation relationship between the image coordinate system of each section and the global coordinate system of the stator. Step S23: At each preset cross-section, a standard cylinder with a known diameter is placed at a fixed position inside the stator cavity. The vision imaging unit is controlled to acquire images of the standard cylinder, extract the cylinder edge and fit a circle, calculate its pixel diameter in the image, and determine the pixel equivalent of the cross-section based on the ratio of the actual physical diameter of the standard cylinder to the pixel diameter. The measurement is repeated multiple times in different areas of the field of view, and the average value is taken as the final pixel equivalent. Step S24: At the first preset section, using the stator split surface as a reference, acquire the split surface image and extract the straight line equation of the split surface in the image. Combine the known geometric relationship between the split surface and the stator centerline to calculate the theoretical position of the stator center in the image. At the second preset section, using the cross mark pre-etched on the inner wall of the stator as a reference, locate the coordinates of the cross intersection point in the image through template matching and record the coordinates as the reference reference position.

[0032] Specifically, during the rotor fitting process, the visual imaging unit continuously acquires a sequence of images including the outer edge of the rotor and the reference reference of the stator inner wall, and determines a comprehensive fitting quality index. Based on the comparison between the comprehensive fitting quality index and a preset comprehensive fitting quality index, it determines whether the fitting quality of the current fitting process meets the standard. If it does not meet the standard, a non-compliance detection flag is output and the comprehensive fitting quality index at the current moment is recorded. If the overall dressing quality is less than the preset overall dressing quality index, it is determined that the dressing quality of the current dressing process is substandard, a substandard detection mark is output, and the overall dressing quality index at the current moment is recorded. If the overall dressing quality index is greater than or equal to the preset overall dressing quality index, then the dressing quality of the current dressing process is determined to be up to standard.

[0033] In this embodiment of the invention, the process of determining whether the dressing quality of the current dressing process meets the standard includes: Step S31: Obtain the real-time radial clearance vector of the rotor at this cross-section. Specifically: For each preset cross-section, the Canny edge detection operator is used to perform edge detection on each frame of the acquired image, extracting the edge points of the rotor's outer circle. The contour of the rotor's outer circle is obtained by fitting using the least squares method, thereby determining the coordinates of the rotor's center in the image. Simultaneously, based on the reference datum position, different calculation methods are used to obtain the theoretical coordinates of the stator center in the image. When the reference datum is the stator bifurcation surface, the straight line equation of the bifurcation surface in the image is extracted by Hough transform. Combined with the known geometric relationship between the bifurcation surface and the stator centerline, where the stator center is located on the perpendicular bisector of the bifurcation surface and the distance from the bifurcation surface is the design value, the stator center is calculated. Center coordinates; when the reference datum is a cross mark, the coordinates of the cross intersection point are accurately located by template matching or grayscale centroid method, and the coordinates of the stator center are calculated according to the relative position relationship between the pre-calibrated cross mark and the stator center; the pixel offset between the rotor center and the stator center is calculated, including the horizontal offset and the vertical offset, and the pixel equivalent is used to convert the pixel offset into the horizontal eccentricity and vertical eccentricity in physical space; according to the design radius of the rotor and the design inner diameter of the stator at this section, combined with the horizontal eccentricity and the vertical eccentricity, the radial clearance values ​​in the top, bottom, left and right directions of this section are calculated respectively, and the radial clearance values ​​in these four directions constitute the real-time radial clearance vector of this section.

[0034] Step S32: Obtain the real-time eccentricity vector of the rotor at this cross section. Specifically: After obtaining the horizontal pixel offset and vertical pixel offset between the rotor center and the stator center, the horizontal pixel offset is converted into a horizontal eccentricity in physical space using the pixel equivalent, and the vertical pixel offset is converted into a vertical eccentricity in physical space. The horizontal eccentricity and vertical eccentricity together constitute the real-time eccentricity vector of this cross section at the current moment.

[0035] Step S33: Obtain the global minimum gap value at the current moment. Specifically: at each moment, summarize the gap values ​​in the four directions of the radial gap vector obtained by real-time calculation of each preset section to obtain a set of gap values ​​in all directions of all sections; select the gap value with the smallest value from the set as the global minimum gap value at the current moment.

[0036] Step S34: Obtain the spatial axis tilt angle of the rotor. Specifically: Based on the real-time eccentric vector of each preset section, obtain the eccentric position of the rotor center relative to the stator center at each section, that is, the two-dimensional eccentric coordinates at each section determined by the horizontal and vertical eccentricity; Combine the axial coordinates of each preset section, that is, the Z-axis coordinates of each section in the stator global coordinate system, and use the eccentric position of each section as a discrete point in three-dimensional space. Use the least squares method to fit these discrete points into a spatial straight line, which represents the spatial axis of the rotor at the current moment; Calculate the angle between the spatial straight line and the stator design center line, that is, the Z-axis of the stator global coordinate system, which is the spatial axis tilt angle of the rotor at the current moment.

[0037] Step S35: Calculate the eccentricity consistency coefficient for each cross-section. Specifically: Normalize the real-time eccentricity vector of each preset cross-section to obtain the unit vector of the eccentricity direction for each cross-section; calculate the average angle between the unit vectors of the eccentricity direction of all cross-sections. The specific calculation method is as follows: Combine all cross-sections in pairs and calculate the angle between each pair of unit vectors of the eccentricity direction. The angle is calculated using the vector dot product formula, that is, the cosine of the angle is the dot product of the two unit vectors, and then the angle value is obtained by the inverse cosine function; sum all the angles of the pairwise combinations and divide by the number of combinations to obtain the average angle value; the smaller the average angle value, the more consistent the eccentricity direction of each cross-section; according to the magnitude of the average angle value, convert it into a value between 0 and 1 through a linear mapping relationship. The linear mapping relationship is 1 - (average angle / 180°). This value is the eccentricity consistency coefficient of each cross-section. The larger the value, the better the eccentricity consistency.

[0038] Step S36: Calculate the comprehensive dressing quality index. Specifically: Divide the current global minimum gap value by the design gap value of the cross section to obtain the gap factor; divide the spatial axis tilt angle by the maximum allowable tilt angle, and then subtract the ratio from 1 to obtain the tilt angle factor; use the eccentricity consistency coefficient directly as the consistency factor, and sum the three factors by weight to obtain the comprehensive dressing quality index at the current moment.

[0039] The weighting coefficient of the gap factor is 0.2, the weighting coefficient of the tilt angle factor is 0.4, and the weighting coefficient of the consistency factor is 0.4.

[0040] In this embodiment of the invention, the preset comprehensive clothing quality coefficient ranges from 0.85 to 0.95, preferably 0.9. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0041] Specifically, this invention solves the problem of difficult fusion and comparison of measurement results from different sections in traditional methods by installing a visual imaging unit at a preset cross-section of the stator inner cylinder and establishing a global coordinate system, thus unifying multi-section image data under the same spatial reference. By calibrating pixel equivalents and reference reference positions, it achieves accurate conversion from image pixels to physical dimensions, providing a reliable quantitative basis for subsequent millimeter-level gap measurements. During the fitting process, by real-time calculation of the radial gap vector and eccentricity vector of each section, and further extracting the global minimum gap value, spatial axis tilt angle, and cross-sectional eccentricity consistency coefficient, the original image data is transformed into data with clear... The precise physical orientation parameters enable operators to intuitively grasp the gap state and overall posture of various parts of the rotor. Based on this, by weighting and integrating the gap factor, tilt angle factor, and consistency factor into a comprehensive dressing quality index, the multi-dimensional orientation features are quantified into a single evaluation index. This index can comprehensively reflect the most dangerous gap state, overall tilt degree, and eccentricity coordination of each section during the dressing process. Furthermore, it enables the determination of dressing quality through preset values, providing an objective decision-making basis for subsequent risk assessment and adaptive switching of detection modes. This fundamentally solves the technical defects of traditional methods that rely on manual experience and lack unified evaluation standards.

[0042] Please see Figure 2 As shown, it is a logic block diagram of an embodiment of the present invention for determining whether to execute a warning detection mode or an emergency stop detection mode based on the clothing quality deviation coefficient.

[0043] Specifically, when it is determined that the current attire is substandard, based on the comprehensive attire quality index and the target comprehensive attire quality index attire quality deviation coefficient, and based on the comparison result of the attire quality deviation coefficient and the preset deviation coefficient, it is determined whether to execute a warning detection mode or an emergency stop detection mode. If the clothing quality deviation coefficient is less than or equal to the preset deviation coefficient, then the early warning detection mode is determined to be executed; If the clothing quality deviation coefficient is greater than the preset deviation coefficient, then the emergency stop detection mode will be executed.

[0044] In this embodiment of the invention, the clothing quality deviation coefficient is the absolute difference between the comprehensive clothing quality index and the preset comprehensive clothing quality index divided by the preset comprehensive clothing quality index.

[0045] In this embodiment of the invention, the preset deviation coefficient ranges from 0.15 to 0.25, preferably 0.2. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0046] In this embodiment of the invention, the early warning detection mode is as follows: Maintaining the normal acquisition frame rate of the visual imaging unit, highlighting the current risk section in yellow on the display interface, increasing the data storage frequency of this section in the data record, and activating an audible and visual early warning signal to alert the operator to the current clothing status; the system continues to calculate the radial gap vector and eccentricity vector of each section in real time, and compares the current comprehensive clothing quality index with the value of the previous moment to monitor its changing trend; if the comprehensive clothing quality index shows a downward trend for three consecutive sampling moments under the early warning detection mode, it is determined that the situation is continuously deteriorating and approaching the triggering of the emergency stop detection mode, automatically upgrading the early warning level, that is, highlighting the risk section in red flashing on the display interface, and simultaneously suggesting the operator prepare for an emergency stop through voice prompts.

[0047] The emergency stop detection mode is as follows: an emergency stop signal is immediately output to prompt the operator to stop the dressing process; at the same time, the visual imaging unit automatically switches to a higher frame rate acquisition mode, increasing the frame rate to 2-3 times the original frame rate, in order to capture the shaking process that may occur after the emergency stop; the real-time radial gap vector, real-time eccentric vector, and the comprehensive dressing quality index at the current moment of the emergency stop trigger are recorded at each section, and these data are used as input parameters for subsequent steps; the display interface automatically switches to the emergency stop monitoring view, focusing on displaying the gap status and eccentric vector direction of each section.

[0048] Please see Figure 3 As shown, it is a logic block diagram of an embodiment of the present invention for determining whether there is a risk of instantaneous scraping or shaking of the rotor after an emergency stop based on the emergency stop stability coefficient.

[0049] Specifically, under the condition of executing the emergency stop detection mode, the propulsion speed is reduced to zero, and a sequence of images is continuously acquired within a preset time after the stop to construct an emergency stop stability coefficient. Based on the comparison between this emergency stop stability coefficient and a preset emergency stop stability coefficient, it is determined whether the rotor has a risk of instantaneous scraping or shaking after the emergency stop. If the emergency stop stability coefficient is less than or equal to the preset emergency stop stability coefficient, then it is determined that there is no risk of instantaneous scraping or shaking. If the emergency stop stability coefficient is greater than the preset emergency stop stability coefficient, then there is a risk of instantaneous scraping or shaking.

[0050] In this embodiment of the invention, the process of constructing the emergency stop stability coefficient includes, Step S51: Obtain the radial clearance time-series change data after stopping. For the radial clearance time-series change data, use the damped oscillation model to perform curve fitting to obtain three key parameters for each section: initial oscillation amplitude, which represents the maximum amplitude of rotor sway at the moment of emergency stop; oscillation damping coefficient, which represents the rate of sway decay over time; and oscillation frequency, which represents the periodic characteristics of sway. The damped oscillation model describes the periodic damped oscillation of the gap value around a certain steady-state value using a time variable.

[0051] Step S52: Extract the global minimum dynamic clearance from the time series data of all sections, that is, the minimum value of the radial clearance in all sections at all times after stopping, which characterizes whether the rotor and stator make instantaneous contact during the emergency stop process. Step S53: Divide the initial oscillation amplitude by the design gap value at the cross section to obtain the relative oscillation amplitude; set the oscillation attenuation coefficient to a preset attenuation coefficient to obtain the relative attenuation coefficient; divide the global minimum dynamic gap by the design gap value at the cross section to obtain the gap safety factor. The preset attenuation coefficient is preset based on the statistical values ​​of historical stable clothing data.

[0052] Step S54: Take the maximum value of the relative oscillation amplitude of all cross sections, take the minimum value of the relative attenuation coefficient of all cross sections, take the gap safety factor corresponding to the global minimum dynamic gap as the overall gap safety factor, and sum the three values ​​by weight to obtain the emergency stop stability coefficient. The weighting coefficient for the relative oscillation amplitude is 0.4, the weighting coefficient for the relative attenuation coefficient is 0.2, and the weighting coefficient for the gap safety factor is 0.4.

[0053] In this embodiment of the invention, the preset emergency stop stability coefficient ranges from 0.35 to 0.45, preferably 0.4. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0054] Specifically, this invention quantifies and classifies substandard conditions by constructing a clothing quality deviation coefficient, achieving dynamic matching between detection modes and risk levels: when the deviation coefficient is small, an early warning detection mode is activated, visually marking risk sections and increasing data storage frequency while maintaining normal data acquisition. It also continuously monitors the changing trend of the comprehensive clothing quality index, automatically raising the early warning level when the index continuously declines, allowing operators sufficient warning and preparation time before the risk escalates; when the deviation coefficient exceeds a threshold, an emergency stop detection mode is immediately triggered, automatically increasing the image acquisition frame rate to capture the transient shaking process that may occur after the emergency stop, and completely recording the key pose data at the moment of the emergency stop; based on this, By fitting the radial clearance time-series data after an emergency stop with damped oscillations, three key parameters are extracted: initial oscillation amplitude, oscillation damping coefficient, and global minimum dynamic clearance. An emergency stop stability coefficient is then constructed to quantitatively evaluate the dynamic behavior after an emergency stop. The emergency stop stability coefficient represents the maximum amplitude of the sway through the relative oscillation amplitude, the rate of sway attenuation through the relative damping coefficient, and the safety margin of the minimum clearance through the clearance safety factor. The weighted fusion of these three parameters is compared with a preset threshold, enabling quantitative detection and risk assessment of the inertial sway of a large rotor after an emergency stop. This solves the technical blind spot of existing technologies being completely unable to capture transient behavior after an emergency stop, and provides a key dynamic safety assessment method for sudden working conditions during the fitting process.

[0055] Specifically, based on the changing trend of the comprehensive fitting quality index time series data within a preset period before the emergency stop, the expected stability coefficient under the current emergency stop state is predicted. The consistency deviation coefficient between the expected stability coefficient and the emergency stop stability coefficient is calculated. Based on the comparison result between the consistency deviation coefficient and the preset consistency deviation coefficient, it is determined whether the detection status of the current rotor dynamic fitting process meets the standard. If the consistency deviation coefficient is less than or equal to the preset consistency deviation coefficient, then the detection status of the current rotor dynamic mounting process is determined to be up to standard. If the consistency deviation coefficient is greater than the preset consistency deviation coefficient, then the detection status of the current rotor dynamic fitting process is determined to be substandard.

[0056] In this embodiment of the invention, the process of obtaining the consistency deviation coefficient is as follows: Step S61: Extract the comprehensive clothing quality index of multiple consecutive moments within a preset time period before the emergency stop occurs, and form a time-series data sequence; The preset time period is set according to the stability of the dressing process and the sampling frequency, and usually selects data within 5 to 10 seconds before the emergency stop.

[0057] Step S62: Perform linear regression analysis on the time series data sequence, with time as the independent variable and the comprehensive clothing quality index as the dependent variable, fit the trend line of the comprehensive clothing quality index with time, and obtain the slope and intercept of the change within the preset time period. The slope of change represents the rate of change of the overall clothing quality index over time. A positive value indicates that the quality index is on the rise, meaning that the clothing condition is improving, while a negative value indicates that the quality index is on the fall, meaning that the clothing condition is deteriorating. The intercept is used to determine the specific position of the trend line on the time axis.

[0058] Step S63: Based on the trend line and the precise time of the emergency stop, substitute the time of the emergency stop into the trend line equation and extrapolate to calculate the predicted value of the instantaneous comprehensive clothing quality index at the moment of the emergency stop. The time of the emergency stop is determined based on the moment when the propulsion speed drops to zero as monitored in real time in step S2. Step S64: Call the pre-established mapping relationship model between the comprehensive clothing quality index and the emergency stop stability coefficient, input the extrapolated instantaneous comprehensive clothing quality index prediction value into the mapping relationship model, and obtain the expected stability coefficient under the current emergency stop state. The mapping relationship model is obtained through regression analysis based on historical clothing data. The specific method is as follows: collect the time series data of the comprehensive clothing quality index before each emergency stop event during the historical clothing process and its corresponding measured emergency stop stability coefficient to form a training sample set. Use the least squares method to perform regression fitting and establish the functional mapping relationship between the two.

[0059] Step S65: Calculate the absolute difference between the expected stability coefficient and the emergency stop stability coefficient measured in step S4. Divide the absolute difference by the expected stability coefficient to obtain the consistency deviation coefficient, which characterizes the degree of deviation between the two.

[0060] In this embodiment of the invention, the preset consistency deviation coefficient ranges from 0.1 to 0.2, preferably 0.15. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0061] Please see Figure 4 As shown, it is a logic block diagram of an embodiment of the present invention for determining the optimization of detection parameters in the corresponding detection mode based on the difference between the consistency deviation coefficient and the preset consistency deviation coefficient.

[0062] Specifically, when it is determined that the current detection status is substandard, the detection parameters in the corresponding detection mode are optimized based on the comparison between the difference between the consistency deviation coefficient and the preset consistency deviation coefficient and the preset difference. If the difference is less than or equal to a preset difference, then the first optimization strategy is determined to be applied to the detection parameters; If the difference is greater than a preset difference, then the second optimization strategy is determined to be applied to the detection parameters.

[0063] In this embodiment of the invention, the preset difference range is 0.05 to 0.10, preferably 0.07. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0064] In this embodiment of the invention, the first optimization strategy is to adjust the weighting coefficients in the weighted calculation of the comprehensive clothing quality index. Specifically, real-time eccentricity vector data of each cross-section before the emergency stop is extracted, and the eccentricity consistency coefficient and spatial axis tilt angle of each cross-section at each moment are calculated based on the real-time eccentricity vector of each cross-section, forming two time-series data sequences. Correlation analysis is performed on the two time-series data sequences to determine whether the main source of the current deviation is insufficient eccentricity consistency or excessive tilt angle. If the eccentricity consistency is insufficient, in the weighted calculation of the comprehensive clothing quality index, the weighting coefficient corresponding to the consistency factor is increased based on the current weighting coefficient, with an increase range of 0.05 to 0.10, preferably 0.06, and the weighting coefficient corresponding to the tilt angle factor is correspondingly decreased. If the tilt angle is too large, in the weighted calculation of the comprehensive clothing quality index, the weighting coefficient corresponding to the tilt angle factor is increased based on the current weighting coefficient, with an increase range of 0.04 to 0.08, preferably 0.06, and the weighting coefficient corresponding to the consistency factor is correspondingly decreased. The decrease should be consistent with the increase.

[0065] The specific process of performing correlation analysis on the two time-series data sequences is as follows: Step S71: Using the Pearson correlation coefficient calculation method, calculate the linear correlation between the eccentricity consistency coefficient of each section and the tilt angle of the spatial axis at the same moment, and obtain the correlation coefficient value. The Pearson correlation coefficient is a dimensionless index that characterizes the direction and strength of the linear correlation between two variables. The value range is -1 to 1. The closer the absolute value is to 1, the stronger the correlation. The closer it is to 0, the weaker the correlation. Step S72: Determine the main source of the current deviation based on the sign and magnitude of the correlation coefficient value, wherein... If the correlation coefficient is negative and its absolute value is greater than the preset correlation coefficient value, it indicates that the inconsistency of the eccentricity direction of each section and the increase of the tilt angle are negatively correlated, that is, the main source of the current deviation is the insufficient consistency of the eccentricity of each section. If the correlation coefficient is positive and its absolute value is greater than the preset correlation coefficient value, it indicates that the inconsistency of the eccentricity direction of each section is positively correlated with the increase of the tilt angle, that is, the main source of the current deviation is that the overall tilt angle is too large.

[0066] In this embodiment of the invention, the preset correlation coefficient value ranges from 0.5 to 0.7, preferably 0.6. The preferred range and preferred value can be determined according to the actual situation, and are not specifically limited here.

[0067] In this embodiment of the invention, the second optimization strategy is to calibrate the position of the visual imaging unit and adjust the trigger preset value of the emergency stop detection mode. Specifically, the online self-calibration function of the visual imaging unit is activated. By acquiring an image of a fixed reference reference on the inner wall of the stator, the image is compared with the reference reference position recorded in step S1 to determine whether the visual unit has shifted or become loose. If the self-test detects a change in the pose of the visual unit, an automatic prompt is issued, requiring the operator to re-execute the external parameter calibration and pixel equivalent calibration in step S1. If the self-test does not detect any hardware problems, the system automatically adjusts the preset deviation coefficient in step S3, shifting the trigger preset value of the emergency stop detection mode towards a more conservative direction, that is, reducing the preset deviation coefficient. The reduced preset deviation coefficient ranges from 0.1 to 0.2, preferably 0.15, which is more inclined to trigger the emergency stop detection mode to obtain more data after the emergency stop for verification.

[0068] Specifically, this invention performs self-verification of the detection results by calculating the consistency deviation coefficient, thereby achieving a quantitative assessment of the reliability of the detection system itself. It extracts the changing trend of the comprehensive clothing quality index before an emergency stop and extrapolates the predicted stability coefficient, comparing it with the measured emergency stop stability coefficient after the emergency stop. When the deviation exceeds a preset value, the detection state is deemed substandard, thus identifying potential systemic problems during the detection process. Based on this, a differentiated optimization strategy is adaptively executed according to the degree of deviation: when the deviation is small, correlation analysis accurately locates the source of the deviation as insufficient eccentricity or excessive tilt angle, and the weighting coefficients in the comprehensive clothing quality index are adjusted accordingly, making subsequent detection more focused on the actual dominant risks; when the deviation is large, the visual unit self-calibration function is activated to determine whether the hardware has shifted or become loose. If there is no hardware problem, the trigger threshold of the emergency stop detection mode is automatically shifted towards a more conservative direction to obtain more emergency stop data for verification. This enables the detection system to become more accurate with use, solving the technical defects of existing detection methods that cannot verify the reliability of their results and cannot automatically adjust the detection strategy according to deviations, significantly improving the robustness and adaptability of clothing detection under complex working conditions.

[0069] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A machine vision-based method for spatial pose detection during the installation process of a steam turbine rotor, characterized in that, include, Several visual imaging units are fixedly installed at several predetermined cross-sections of the inner cylinder of the turbine stator, and the visual imaging units are calibrated. Establish the transformation relationship between the image coordinate system and the stator global coordinate system, and determine the pixel equivalent and reference datum position; During the dressing process, the visual imaging unit continuously acquires a sequence of images including the outer edge of the rotor and the reference reference of the inner wall of the stator, determines the real-time radial clearance vector and real-time eccentricity vector of the rotor at the current section, and determines the comprehensive dressing quality index to determine whether the dressing quality of the current dressing process meets the standard. Based on the clothing quality deviation coefficient, which represents the degree to which the current clothing quality deviates from the standard requirements under the current substandard clothing condition, determine whether to implement the early warning detection mode or the emergency stop detection mode. Under the condition of executing the emergency stop detection mode, the propulsion speed is reduced to zero, and the sequence images within a preset time after stopping are collected to obtain the radial clearance time change data after stopping, and an emergency stop stability coefficient is constructed to determine whether there is an instantaneous scraping risk or shaking risk to the stator after the part of the rotor that has entered the stator is stopped. Predict the expected stability coefficient under the current emergency stop state, and determine the consistency deviation coefficient between the predicted stability coefficient and the emergency stop stability coefficient, so as to determine whether the detection status of the current rotor dynamic mounting process meets the standard. Based on the difference between the consistency deviation coefficient and the preset consistency deviation coefficient, the weighting coefficient in the weighted calculation of the comprehensive clothing quality index is determined, the position of the calibrated visual imaging unit is adjusted, and the trigger preset value of the emergency stop detection mode is adjusted.

2. The method for spatial pose detection during turbine rotor installation based on machine vision according to claim 1, characterized in that, The process of determining whether the dressing quality of the current dressing process meets the standards includes, The real-time radial gap vectors are summarized, and the minimum value is selected as the global minimum gap value at the current moment; Based on the real-time eccentricity vector, determine the tilt angle of the rotor's spatial axis at the current moment; The real-time eccentricity vector is normalized to obtain the unit vector of eccentricity direction for each cross section, and the eccentricity consistency coefficient of each cross section is calculated. The overall dressing quality index is calculated by weighting the global minimum gap value, the spatial axis tilt angle, and the eccentricity consistency coefficient of each section. The comprehensive clothing quality index is compared with the preset comprehensive clothing quality index; Based on the fact that the overall dressing quality index is less than the preset overall dressing quality index, it is determined that the dressing quality of the current dressing process is substandard.

3. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 2, characterized in that, The process of determining whether to execute the early warning detection mode or the emergency stop detection mode includes: Calculate the absolute difference between the comprehensive clothing quality index and the preset comprehensive clothing quality index; Divide the absolute difference by the preset comprehensive clothing quality index to obtain the clothing quality deviation coefficient; Based on the fact that the clothing quality deviation coefficient is less than or equal to the preset deviation coefficient, the early warning detection mode is determined to be executed. The warning detection mode maintains a normal image acquisition frame rate, visually identifies risk sections, increases data storage frequency, and continuously monitors the changing trend of the comprehensive clothing quality index. When the index continues to decline and the clothing condition continues to deteriorate, the warning level is automatically upgraded.

4. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 3, characterized in that, The process of determining whether to execute the early warning detection mode or the emergency stop detection mode also includes... Based on the fact that the clothing quality deviation coefficient is greater than the preset deviation coefficient, it is determined to execute the emergency stop detection mode. The emergency stop detection mode immediately outputs an emergency stop signal to stop the dressing process. The visual imaging unit automatically switches to a high frame rate acquisition mode and records the real-time radial gap vector, real-time eccentricity vector, and the comprehensive dressing quality index of each section at the moment the emergency stop is triggered.

5. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 4, characterized in that, The process of determining whether there is a risk of instantaneous scraping or shaking of the stator after an emergency stop of a portion of the rotor that has entered the stator includes... The radial gap time-series variation data are fitted to obtain the initial oscillation amplitude, oscillation attenuation coefficient, and oscillation frequency for each cross section; Extract the global minimum dynamic gap from the radial gap time-series variation data; The initial oscillation amplitude, the oscillation attenuation coefficient, and the global minimum dynamic gap are normalized to obtain the relative oscillation amplitude, the relative attenuation coefficient, and the gap safety factor. The relative oscillation amplitude is taken as the maximum value, the relative attenuation coefficient is taken as the minimum value, and the gap safety factor corresponding to the global minimum dynamic gap is taken as the overall gap safety factor. The three are weighted and summed to obtain the emergency stop stability coefficient. The emergency stop stability coefficient is compared with the preset emergency stop stability coefficient; Based on the fact that the emergency stop stability coefficient is greater than the preset emergency stop stability coefficient, it is determined that there is a risk of instantaneous scraping or shaking.

6. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 5, characterized in that, The process of determining whether the detection status of the current rotor dynamic fitting process meets the standard includes, Extract the comprehensive clothing quality index from multiple consecutive moments within a preset time period before the emergency stop to form a time-series data sequence; Linear regression analysis was performed on the time series data to fit the trend line of the comprehensive clothing quality index over time; The instantaneous comprehensive clothing quality index prediction value at the moment of emergency stop is calculated based on the trend line and the precise moment of emergency stop. The expected stability coefficient is obtained by mapping the instantaneous comprehensive clothing quality index prediction value. The consistency deviation coefficient is calculated based on the expected stability coefficient and the measured emergency stop stability coefficient. The consistency deviation coefficient is compared with the preset consistency deviation coefficient; Based on the fact that the consistency deviation coefficient is greater than the preset consistency deviation coefficient, it is determined that the detection status of the current rotor dynamic fitting process is substandard.

7. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 6, characterized in that, The process of determining and optimizing the detection parameters in the corresponding detection mode includes: Calculate the difference between the consistency deviation coefficient and the preset consistency deviation coefficient; The difference is compared with a preset difference. Based on the difference being less than or equal to a preset difference, the weighting coefficients in the comprehensive clothing quality index are determined and adjusted.

8. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 7, characterized in that, The process of determining and optimizing the detection parameters in the corresponding detection mode also includes... Based on the fact that the difference is greater than a preset difference, the position of the calibrated visual imaging unit is determined and the trigger preset value of the emergency stop detection mode is adjusted.

9. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 8, characterized in that, The process of adjusting the weighting coefficients in the weighted calculation of the comprehensive clothing quality index includes: Extract real-time eccentricity vector data of each section before the emergency stop occurs; Based on the real-time eccentricity vector of each section, the eccentricity consistency coefficient and the spatial axis tilt angle of each section at each moment are calculated to form two time series data sequences. Correlation analysis was performed on the two time-series data sequences to obtain the correlation coefficient values; The main source of the current deviation is determined based on the comparison between the correlation coefficient value and the preset correlation coefficient value. Based on the fact that the correlation coefficient is negative and its absolute value is greater than the preset correlation coefficient value, it is determined that the inconsistency of the eccentricity direction of each section and the increase of the tilt angle are negatively correlated. That is, the main source of the current deviation is the insufficient consistency of the eccentricity of each section, and the weight coefficient corresponding to the consistency factor is increased.

10. The machine vision-based spatial pose detection method for turbine rotor installation process according to claim 9, characterized in that, The process of adjusting the weighting coefficients in the weighted calculation of the comprehensive clothing quality index also includes... Based on the fact that the correlation coefficient is positive and its absolute value is greater than the preset correlation coefficient, it is determined that the inconsistency of the eccentricity direction of each section is positively correlated with the increase of the tilt angle. That is, the main source of the current deviation is that the overall tilt angle is too large, so the weight coefficient corresponding to the tilt angle factor is increased.

Images