Gas tank piston torsion and drift measurement method and system based on image recognition

Abstract

The present invention provides a method and system for measuring torsion and drift of a gas tank piston based on image recognition, which relates to the field of measurement technology, including using a binocular stereo camera to synchronously collect continuous images of left and right paths during the operation of the gas tank piston, transmitting the collected image pairs to a computer for stereo matching, and calculating the three-dimensional spatial coordinates of each identification point; comparing the current three-dimensional coordinates of the identification point with the reference coordinates, fitting the spatial plane equation of the piston top surface by the least squares method, calculating the Euler angle change of the plane normal vector relative to the reference state, and obtaining the torsion angle of the piston around the X-axis, Y-axis and Z-axis; establishing a trend prediction model for the piston operation state, and predicting the remaining service life and failure risk of the piston by analyzing the change rate of the piston torsion and drift and the dynamic characteristics of acceleration.

Description

Technical Field

[0001] The invention relates to measurement technology, and in particular to a method and system for measuring torsion and drift of a gas cabinet piston based on image recognition. Background Art

[0002] Reciprocating compressors are key equipment widely used in the petroleum and chemical industries. Their reliability and safety directly affect production efficiency and product quality. The gas tank is one of the core components of the reciprocating compressor, and the movement state of the piston inside it has an important impact on the performance and life of the compressor. During operation, the piston is prone to twisting and drifting due to the effects of factors such as gas pressure and inertia, resulting in an increase in the gap between the piston and the cylinder, increased leakage, and reduced compression efficiency. In severe cases, it may even cause the piston to collide with the cylinder, causing equipment damage and safety accidents. Therefore, real-time monitoring of the torsion and drift state of the gas tank piston is of great significance for the status assessment and fault diagnosis of the reciprocating compressor.

[0003] The existing gas tank piston torsion and drift measurement methods have problems such as low measurement accuracy, great difficulty in implementation, and poor applicability, which makes it difficult to meet the requirements of reciprocating compressor status monitoring and fault diagnosis. Summary of the invention

[0004] The embodiments of the present invention provide a method and system for measuring torsion and drift of a gas cabinet piston based on image recognition, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] Provides a method for measuring gas tank piston torsion and drift based on image recognition, including:

[0007] At least three non-collinear identification points are set on the top of the piston of the gas tank, and the identification points are markers with specific colors or patterns. The initial three-dimensional spatial coordinates of the identification points are recorded as reference coordinates; at the same time, a binocular stereo camera is installed at a fixed position of the gas tank so that its optical axis is parallel to the axis of the gas tank piston, and the internal and external parameters of the camera are determined by binocular stereo calibration; the binocular stereo camera is used to synchronously collect continuous images of the left and right paths during the operation of the gas tank piston, and the collected image pairs are transmitted to the computer for stereo matching. Through epipolar line constraints and similarity measurements, the corresponding identification points are identified in the left and right views, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle;

[0008] The current three-dimensional coordinates of the identification point are compared with the reference coordinates, and the spatial plane equation of the piston top surface is fitted by the least square method. The Euler angle change of the plane normal vector relative to the reference state is calculated to obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, the offset of the center of mass coordinates of the identification point relative to the reference position is calculated to obtain the drift distance of the piston in three directions;

[0009] The piston torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order. A trend prediction model for the piston operating status is established using time series analysis and machine learning algorithms. The remaining service life and failure risk of the piston are predicted by analyzing the dynamic characteristics of the changing rate of piston torsion and drift and acceleration. A series of strain gauges are installed on the pipe wall of the gas holder to measure the stress distribution and deformation of the gas holder during operation in real time. The strain data are correlated with the piston torsion and drift data for analysis, and a comprehensive assessment model for the safety status of the gas holder is constructed to comprehensively assess the health level and remaining service life of the gas holder.

[0010] In an optional embodiment,

[0011] The collected image pairs are transferred to the computer for stereo matching. The corresponding identification points are identified in the left and right views through epipolar constraints and similarity measurements, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle, including:

[0012] According to the projection center of the binocular stereo camera and the pixel coordinates of the points to be matched in the left and right views, using the essential matrix obtained by pre-calibration, an epipolar constraint equation is established for the points to be matched in the left and right views to satisfy the three-point coplanar constraint;

[0013] Using the established epipolar constraint equation, for any pixel point to be matched in the left view, determine the straight line where its corresponding point in the right view is located, and reduce the right view to be matched area to the area near the straight line;

[0014] Taking the matching point in the left and right views as the center, pixel blocks of the same size are intercepted respectively, and the normalized cross-correlation coefficient of the two pixel blocks is calculated as a measure of the similarity between the two pixel blocks. The calculation method of the normalized cross-correlation coefficient is:

[0015] ;

[0016] Where NCC(p,q) represents the normalized cross-correlation coefficient, p and q represent two pixel blocks to be matched, which come from the areas to be matched in the left and right views, respectively, and n represents the size of the pixel block. , Represents pixel block W p and W qThe grayscale value at the middle coordinate (i, j), where i and j are relative coordinates inside the pixel block, ranging from 0 to n-1; , Represents pixel blocks W p and W q The average gray value of

[0017] By setting a normalized cross-correlation coefficient threshold, the pixel points exceeding the normalized cross-correlation coefficient threshold are screened out from all candidate matching pixels to obtain the best matching pixel coordinates of the identification points in the left and right views;

[0018] The pixel coordinates of the matching points in the left and right views and the projection matrix of the binocular camera are substituted into the spatial coordinate solution equation group based on the triangulation principle, and the equation group is solved by least squares optimization to obtain the three-dimensional spatial coordinates of the identification points. The objective function of the least squares optimization is the sum of the squares of the Euclidean distances between the pixel coordinates of the matching point and its spatial coordinates obtained by back-projection of the projection matrix on the projection pixel coordinates of the left and right views.

[0019] In an optional embodiment,

[0020] Compare the current three-dimensional coordinates of the identification point with the reference coordinates, fit the spatial plane equation of the piston top surface by the least squares method, calculate the change in the Euler angle of the plane normal vector relative to the reference state, and obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, calculate the offset of the center of mass coordinates of the identification point relative to the reference position, and obtain the drift distance of the piston in three directions, including:

[0021] The three-dimensional spatial coordinates of multiple identification points in the current state are compared with the coordinates in the reference state, and the spatial plane equation of the piston top surface in the current state is fitted by the least squares method, where the plane equation is in the form of ax+by+cz+d=0, (a, b, c) is the plane normal vector, and d is the distance from the plane to the origin. Multiple linear equations are obtained by substituting the coordinates of the identification points into the plane equation and written in matrix form. The homogeneous linear equations are solved by singular value decomposition to obtain the plane normal vector (a, b, c) and the distance d;

[0022] Calculate the angle between the plane normal vector in the current state and the normal vector in the reference state, decompose the angle into rotation angles around the X-axis, Y-axis and Z-axis, namely, Euler angles, and establish an equation by equalizing the current normal vector to the product of the reference normal vector and the rotation matrix, where the rotation matrix is ​​represented by the multiplication of the rotation matrices around the three coordinate axes. By solving the relationship between the rotation matrix elements and the Euler angles, the torsion angles of the piston around the X-axis, Y-axis and Z-axis are obtained;

[0023] Calculate the center of mass coordinates of multiple identification points in the reference state and the current state, and obtain the drift distance of the piston in the X, Y, and Z directions by subtracting the reference center of mass coordinates from the current center of mass coordinates, where the center of mass coordinates are equal to the arithmetic mean of the coordinates of all identification points in the corresponding directions.

[0024] In an optional embodiment,

[0025] The piston torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order. The trend prediction model of the piston operation status is established by using time series analysis and machine learning algorithms. By analyzing the dynamic characteristics of the change rate and acceleration of the piston torsion and drift, the remaining service life and failure risk of the piston are predicted, including:

[0026] The piston torsion angle and drift distance data obtained from each measurement are stored in a database in chronological order to construct a time series data set containing the torsion angle and drift distance;

[0027] Based on the time series data set, the first-order differential features of the piston torsion angular velocity and drift velocity are obtained by calculating the difference between the torsion angle and the drift distance at adjacent sampling moments. The second-order differential features of the piston torsion angular acceleration and drift acceleration are further obtained by calculating the difference between the first-order differential features. A piston state feature data set including torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration is constructed.

[0028] Using machine learning algorithms, the piston state feature data set is used as input, and the torsion angle and drift distance at multiple future sampling moments are used as output to train a trend prediction model for the piston operation state.

[0029] The trained piston running state trend prediction model is used to predict the piston torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration at multiple sampling moments in the future;

[0030] Analyze and predict the changing trends of the piston torsion angular velocity, drift speed, torsion angular acceleration and drift acceleration. When the torsion angular velocity or drift speed continues to increase, or the torsion angular acceleration and drift acceleration fluctuate abnormally, it is determined that the piston is at risk of failure.

[0031] Based on the predicted piston torsion angle and drift distance, the estimated time until the piston reaches the preset failure threshold is calculated, and the estimated time is used as the remaining service life of the piston.

[0032] In an optional embodiment,

[0033] The method further comprises:

[0034] Using the trained piston torsion angle and drift distance trend prediction model, the piston torsion angle, drift distance and corresponding high-order differential features at multiple future sampling moments are predicted to obtain a series of prediction result sequences at future moments.

[0035] The predicted piston torsion angle, drift distance and their high-order differential characteristic sequences are trend analyzed, and the statistics in the prediction time domain are calculated, including the average value, variance and change rate of each state characteristic. The deterioration degree of the piston running state is judged by analyzing the change trends of the torsion angular velocity, drift velocity, angular acceleration and displacement acceleration.

[0036] Based on the prediction results and trend analysis, the failure risk of the piston is evaluated, and the warning thresholds and failure thresholds of the torsional angular velocity, drift velocity, angular acceleration and displacement acceleration are set. The predicted value is compared with the threshold. When the predicted value exceeds the warning threshold, a warning signal is issued. When the predicted value exceeds the failure threshold, it is determined that the piston has a high failure risk.

[0037] Combine the prediction results and the failure threshold to estimate the remaining service life of the piston. Calculate the time when the torsion angle and drift distance in the prediction results exceed their respective failure thresholds for the first time, and take the smaller value of the two as the estimated remaining service life of the piston.

[0038] Based on the piston failure risk assessment and remaining service life prediction results, the operation and maintenance strategy is dynamically adjusted. The monitoring frequency is increased for pistons at the warning level, maintenance or replacement is arranged for high-risk pistons, and maintenance time and priority are reasonably arranged for pistons with different remaining service lives. The piston failure data is fed back to the prediction model to continuously optimize the model performance.

[0039] In an optional embodiment,

[0040] A series of strain gauges are installed on the pipe wall of the gas cabinet to measure the stress distribution and deformation of the gas cabinet in real time during operation. The strain data is correlated with the piston torsion and drift data to build a comprehensive assessment model for the safety status of the gas cabinet. The health level and remaining life of the gas cabinet are comprehensively assessed, including:

[0041] A series of strain gauges are arranged in the high stress area of ​​the gas cabinet pipe wall and the position where the stress gradient is greater than the preset gradient threshold. The measurement direction of the strain gauge is consistent with the main stress direction of the gas cabinet. The stress distribution and deformation during the operation of the gas cabinet are measured in real time to obtain the strain data of the gas cabinet pipe wall;

[0042] Preprocess the gas cabinet pipe wall strain data and use filtering algorithm to remove high-frequency interference and low-frequency drift in the strain signal;

[0043] Extract multi-domain features of gas cabinet wall strain data, including extracting the statistical features of the mean, variance, and peak-to-peak value of the strain signal, extracting the frequency domain features of the spectral distribution and main frequency amplitude of the strain signal, extracting the wavelet coefficients of the strain signal, and the time-frequency features of the Hilbert-Huang transform;

[0044] Obtain the monitoring data of the torsion angle and drift distance of the gas tank piston, perform correlation analysis on the torsion angle and drift distance data with the gas tank pipe wall strain data, and calculate the correlation coefficient matrix between the strain characteristics and the torsion angle and drift distance characteristics;

[0045] Adopting the dimensionality reduction algorithm of principal component analysis or independent component analysis, the gas tank wall strain data and the piston torsion angle and drift distance data are integrated to extract low-dimensional key features that comprehensively reflect the safety status of the gas tank;

[0046] Based on the low-dimensional key features of the gas cabinet safety status, a gas cabinet safety status classification model is established using support vector machine or random forest algorithm to divide the gas cabinet status into normal, warning and dangerous levels;

[0047] Based on the low-dimensional key features of the gas tank safety status, a neural network or long short-term memory network algorithm is used to establish a gas tank remaining life prediction model. By learning the degradation trend in historical data, the future health status and failure time of the gas tank are predicted;

[0048] Taking into account the gasholder safety status classification results and remaining life prediction results, a quantitative assessment of the health level of the gasholder is conducted to generate a comprehensive assessment report on the gasholder safety status.

[0049] According to a second aspect of the embodiments of the present invention,

[0050] Provides gas tank piston torsion and drift measurement system based on image recognition, including:

[0051] The first unit is used to set at least three non-collinear identification points on the top of the piston of the gas tank, wherein the identification points are markers with specific colors or patterns, and the initial three-dimensional spatial coordinates of the identification points are recorded as reference coordinates; at the same time, a binocular stereo camera is installed at a fixed position of the gas tank so that its optical axis is parallel to the axis of the gas tank piston, and the internal and external parameters of the camera are determined by binocular stereo calibration; the binocular stereo camera is used to synchronously collect continuous images of the left and right paths during the operation of the gas tank piston, and the collected image pairs are transmitted to the computer for stereo matching, and the corresponding identification points are identified in the left and right views through epipolar constraints and similarity measurements, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle;

[0052] The second unit is used to compare the current three-dimensional coordinates of the identification point with the reference coordinates, fit the spatial plane equation of the piston top surface by the least square method, calculate the Euler angle change of the plane normal vector relative to the reference state, and obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, calculate the offset of the center of mass coordinates of the identification point relative to the reference position, and obtain the drift distance of the piston in three directions;

[0053] The third unit is used to store the piston torsion angle and drift distance data obtained from each measurement into a database in chronological order, and use time series analysis and machine learning algorithms to establish a trend prediction model for the piston operating status. By analyzing the changing rate of piston torsion and drift and the dynamic characteristics of acceleration, the remaining service life and failure risk of the piston can be predicted. A series of strain gauges are installed on the pipe wall of the gas holder to measure the stress distribution and deformation of the gas holder during operation in real time, and the strain data is correlated with the piston torsion and drift data for analysis to construct a comprehensive assessment model for the safety status of the gas holder to comprehensively assess the health level and remaining life of the gas holder.

[0054] According to a third aspect of the embodiments of the present invention,

[0055] An electronic device is provided, comprising:

[0056] processor;

[0057] a memory for storing processor-executable instructions;

[0058] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0059] A fourth aspect of the embodiments of the present invention is:

[0060] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0061] By performing stereo matching and triangulation on the collected images of the piston and cylinder wall mark points, the three-dimensional spatial coordinates of the mark points can be accurately reconstructed, and then the kinematic parameters such as the drift, deflection and torsion angle of the piston relative to the cylinder liner can be calculated. This method reduces the matching search range by introducing epipolar constraints, accurately evaluates the similarity of matching points using the normalized correlation coefficient, and solves the triangulation equations using the least squares optimization algorithm, thereby improving the measurement efficiency and robustness while ensuring accuracy.

[0062] By constructing a time series feature data set of piston state parameters, using a machine learning algorithm to establish a trend prediction model, and analyzing the statistical characteristics and change trends of the prediction results, the remaining service life and failure risk of the reciprocating compressor piston can be quantitatively evaluated. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of a method for measuring torsion and drift of a gas cabinet piston based on image recognition according to an embodiment of the present invention;

[0064] Figure 2 It is a structural schematic diagram of a gas cabinet piston torsion and drift measurement system based on image recognition according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0067] Figure 1 FIG. 1 is a flow chart of a method for measuring gas tank piston torsion and drift based on image recognition according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0068] S101. Set at least three non-collinear identification points on the top of the piston of the gas tank, the identification points are markers with specific colors or patterns, and record the initial three-dimensional spatial coordinates of the identification points as reference coordinates; at the same time, install a binocular stereo camera at the fixed part of the gas tank so that its optical axis is parallel to the axis of the gas tank piston, and determine the internal and external parameters of the camera through binocular stereo calibration; use the binocular stereo camera to synchronously collect continuous images of the left and right paths during the operation of the gas tank piston, transmit the collected image pairs to the computer for stereo matching, identify the corresponding identification points in the left and right views through epipolar constraints and similarity measurements, and calculate the three-dimensional spatial coordinates of each identification point using the triangulation principle;

[0069] S102. Compare the current three-dimensional coordinates of the identification point with the reference coordinates, fit the spatial plane equation of the piston top surface by the least square method, calculate the change in the Euler angle of the plane normal vector relative to the reference state, and obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, calculate the offset of the centroid coordinates of the identification point relative to the reference position, and obtain the drift distance of the piston in three directions;

[0070] S103. The piston torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order. A trend prediction model for the piston operating status is established using time series analysis and machine learning algorithms. The remaining service life and failure risk of the piston are predicted by analyzing the dynamic characteristics of the change rate of piston torsion and drift and acceleration. A series of strain gauges are installed on the pipe wall of the gas holder to measure the stress distribution and deformation of the gas holder during operation in real time. The strain data are correlated with the piston torsion and drift data for analysis, and a comprehensive assessment model for the safety status of the gas holder is constructed to comprehensively assess the health level and remaining service life of the gas holder.

[0071] In an optional embodiment,

[0072] The collected image pairs are transferred to the computer for stereo matching. The corresponding identification points are identified in the left and right views through epipolar constraints and similarity measurements, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle, including:

[0073] According to the projection center of the binocular stereo camera and the pixel coordinates of the points to be matched in the left and right views, using the essential matrix obtained by pre-calibration, an epipolar constraint equation is established for the points to be matched in the left and right views to satisfy the three-point coplanar constraint;

[0074] Using the established epipolar constraint equation, for any pixel point to be matched in the left view, determine the straight line where its corresponding point in the right view is located, and reduce the right view to be matched area to the area near the straight line;

[0075] Taking the matching point in the left and right views as the center, pixel blocks of the same size are intercepted respectively, and the normalized cross-correlation coefficient of the two pixel blocks is calculated as a measure of the similarity between the two pixel blocks. The calculation method of the normalized cross-correlation coefficient is:

[0076] ;

[0077] Where NCC(p,q) represents the normalized cross-correlation coefficient, p and q represent two pixel blocks to be matched, which come from the areas to be matched in the left and right views, respectively, and n represents the size of the pixel block. , Represents pixel block W p and W q The grayscale value at the middle coordinate (i, j), where i and j are relative coordinates inside the pixel block, ranging from 0 to n-1; , Represents pixel blocks W p and W q The average gray value of

[0078] By setting a normalized cross-correlation coefficient threshold, the pixel points exceeding the normalized cross-correlation coefficient threshold are screened out from all candidate matching pixels to obtain the best matching pixel coordinates of the identification points in the left and right views;

[0079] The pixel coordinates of the matching points in the left and right views and the projection matrix of the binocular camera are substituted into the spatial coordinate solution equation group based on the triangulation principle, and the equation group is solved by least squares optimization to obtain the three-dimensional spatial coordinates of the identification points. The objective function of the least squares optimization is the sum of the squares of the Euclidean distances between the pixel coordinates of the matching point and its spatial coordinates obtained by back-projection of the projection matrix on the projection pixel coordinates of the left and right views.

[0080] Exemplarily, the collected images of the gas tank piston and cylinder wall identification points are transmitted to the industrial computer through a high-speed optical fiber network, and the three-dimensional spatial coordinates of the identification points are accurately reconstructed using computer vision and stereo matching algorithms. First, based on the internal and external parameters of the binocular stereo camera obtained by the Zhang calibration method in advance, the epipolar geometric constraint relationship between the left and right views is established. For any pixel coordinate of the identification point to be matched in the left view, the epipolar line equation can be used to determine that its corresponding matching point in the right view must fall on a straight line, thereby reducing the matching search range from a two-dimensional plane to a one-dimensional straight line.

[0081] Then, with the identification point to be matched as the center, image blocks with a side length of 64 pixels are intercepted in the left and right views respectively, and the normalized cross-correlation coefficient (NCC) of the two image blocks is calculated to evaluate their similarity. The calculation process of NCC is as follows: first, the average grayscale values ​​of all pixels in the left and right view image blocks are calculated respectively, and then the difference between the grayscale value of each pixel and the average value is calculated to obtain the de-averaged image block; secondly, the sum of the products of the pixel differences at the corresponding positions of the two de-averaged image blocks is calculated as the numerator; then the sum of the squares of the grayscale differences of all pixels in the two image blocks is calculated respectively, and the square root is taken and multiplied as the denominator; finally, the numerator is divided by the denominator, and then divided by the total number of pixels in the image block 64 to obtain the NCC value. The NCC value range is -1 to 1, and the larger the value, the more similar the two image blocks are.

[0082] By setting the NCC threshold to 0.8, the NCC values ​​of all candidate matching pixels are sorted and screened, and the pixel with the largest NCC value that exceeds the threshold is selected as the best matching point. For example, for an image block of a certain identification point in the left view, a total of 100 candidate matching image blocks are searched on the candidate matching line corresponding to the right view, and the calculated maximum NCC value is 0.95, which exceeds the preset threshold of 0.8. The pixel coordinates at the maximum NCC value are used as the coordinates of the best matching point in the right view with the identification point in the left view.

[0083] Finally, the pixel coordinates of the left and right view matching points are substituted into the triangulation equations, which are composed of the projection relationship between the camera projection matrix and the spatial point coordinates. There are 4 equations and 3 unknowns, which are overdetermined equations. In order to obtain the optimal spatial point coordinate solution, the least squares optimization method is used. First, the real spatial coordinates of the identification point are used as the variables to be optimized, and their coordinate values ​​are substituted into the left and right camera projection matrices respectively. After the projection transformation, the predicted pixel coordinates of the left and right views are obtained; then, the Euclidean distance between the predicted coordinates and the actual matching coordinates is calculated, and the least squares optimization objective function is constructed, that is, the sum of the squares of the two distance differences is minimized; the objective function is iteratively optimized and solved by the Levenberg-Marquardt algorithm, and the three-dimensional spatial coordinates when the objective function takes the minimum value are obtained as the final coordinate measurement value of the identification point.

[0084] Taking the five identification points arranged on the piston and cylinder sleeve of a reciprocating compressor gas tank as an example, the baseline length of the binocular stereo camera is 500 mm, the focal length is 16 mm, and the image resolution is 2048 × 1536. The projection matrix of the left and right cameras is obtained by calibration. Assume that the projection matrix of the left camera is [1500, 0, 1024, 0; 0, 1500, 768, 0; 0,0, 1, 0], and the projection matrix of the right camera is [1500, 0, 1024, -750000; 0, 1500, 768, 0; 0, 0, 1,0]. After stereo matching and triangulation of the captured left and right view images, the three-dimensional coordinates of the five identification points are (-50.3, 120.5, 1020.6), (48.2, 118.3, 1025.2), (63.8, 20.4, 1028.7), (-35.6, 22.1, 1019.8), (5.2, 118.7, 1022.3), all in millimeters. Comparing the coordinates of the five identification points with the coordinates of their initial installation positions, it is found that the axial drift of the piston is 2.5 mm, the radial deflection is 1.2 mm, and the torsion angle is 0.8 degrees. At the same time, by tracking the displacement changes of the identification points in 100 consecutive frames of images, it is estimated that the axial drift speed of the piston is 0.8 mm / s, the radial deflection speed is 0.3 mm / s, and the torsion angular velocity is 0.2 degrees / s.

[0085] In summary, by performing stereo matching and triangulation on the collected images of the piston and cylinder wall identification points, the three-dimensional spatial coordinates of the identification points can be accurately reconstructed, and then the kinematic parameters such as the drift, deflection and torsion angle of the piston relative to the cylinder liner can be calculated. This method narrows the matching search range by introducing epipolar constraints, accurately evaluates the similarity of matching points using the normalized correlation coefficient, and solves the triangulation equations using the least squares optimization algorithm, thereby improving the measurement efficiency and robustness while ensuring accuracy. Practical applications have shown that this method can achieve online non-contact measurement of the motion state of the gas tank piston, with a measurement accuracy better than 0.1 mm and 0.1 degrees, and a response speed better than 0.1 seconds, which can provide reliable data support for the state monitoring, fault diagnosis and predictive maintenance of reciprocating compressors.

[0086] In an optional embodiment,

[0087] Compare the current three-dimensional coordinates of the identification point with the reference coordinates, fit the spatial plane equation of the piston top surface by the least squares method, calculate the change in the Euler angle of the plane normal vector relative to the reference state, and obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, calculate the offset of the center of mass coordinates of the identification point relative to the reference position, and obtain the drift distance of the piston in three directions, including:

[0088] The three-dimensional spatial coordinates of multiple identification points in the current state are compared with the coordinates in the reference state, and the spatial plane equation of the piston top surface in the current state is fitted by the least squares method, where the plane equation is in the form of ax+by+cz+d=0, (a, b, c) is the plane normal vector, and d is the distance from the plane to the origin. Multiple linear equations are obtained by substituting the coordinates of the identification points into the plane equation and written in matrix form. The homogeneous linear equations are solved by singular value decomposition to obtain the plane normal vector (a, b, c) and the distance d;

[0089] Calculate the angle between the plane normal vector in the current state and the normal vector in the reference state, decompose the angle into rotation angles around the X-axis, Y-axis and Z-axis, namely, Euler angles, and establish an equation by equalizing the current normal vector to the product of the reference normal vector and the rotation matrix, where the rotation matrix is ​​represented by the multiplication of the rotation matrices around the three coordinate axes. By solving the relationship between the rotation matrix elements and the Euler angles, the torsion angles of the piston around the X-axis, Y-axis and Z-axis are obtained;

[0090] Calculate the center of mass coordinates of multiple identification points in the reference state and the current state, and obtain the drift distance of the piston in the X, Y, and Z directions by subtracting the reference center of mass coordinates from the current center of mass coordinates, where the center of mass coordinates are equal to the arithmetic mean of the coordinates of all identification points in the corresponding directions.

[0091] Exemplarily, in order to evaluate the torsion and drift state of the piston, it is necessary to compare and analyze the three-dimensional coordinates of the identification point at the current moment with its coordinates in the reference state. First, the spatial plane equation of the piston top surface in the current state is estimated by the least squares plane fitting method. By substituting the three-dimensional coordinates of multiple identification points into the plane equation, multiple linear equations are constructed and written in matrix form. The linear equations are solved using the singular value decomposition algorithm to obtain three parameters representing the direction of the plane normal vector and a parameter representing the distance from the plane to the origin. The general form of the plane equation can be written as a linear combination of the three components of the plane normal vector and the coordinates of the space point equal to the negative distance parameter. For example, the plane equation of the piston top surface in the current state is obtained by fitting as 0.12x+0.35y+0.93z-1024.8=0, then the plane normal vector is (0.12, 0.35, 0.93), and the distance parameter is 1024.8.

[0092] Then, the angle between the current plane normal vector and the normal vector in the reference state is calculated and decomposed into the rotation angles around the three coordinate axes. Based on the principle that the current normal vector is equal to the product of the reference normal vector and a rotation matrix, the equation relationship of the rotation transformation is established, where the rotation matrix can be written as the product of three rotation matrices around the X-axis, Y-axis and Z-axis, and each rotation matrix is ​​a function of the Euler angle. By solving the trigonometric function relationship between the Euler angle and the elements of the rotation matrix, the torsion angle of the piston around the three coordinate axes is obtained. Taking the reference normal vector (0.08, 0.21, 0.97) and the current normal vector (0.12, 0.35, 0.93) as an example, it can be obtained that the piston is twisted 1.2 degrees around the X-axis, -2.8 degrees around the Y-axis, and 0.5 degrees around the Z-axis. The negative sign indicates that the torsion direction is opposite to the positive direction of the coordinate axis.

[0093] Finally, the center of mass coordinates of the marking points in the reference state and the current state are calculated, and the drift distance of the piston in the three coordinate directions is obtained by difference. The center of mass of the marking point refers to the average value of the coordinate values ​​of all the marking points in a certain coordinate direction, which reflects the overall position of the marking point. Suppose the coordinates of the five marking points in the reference state are (-46.5, 115.3,985.2), (52.8, 120.6, 986.7), (59.4, 18.9, 990.5), (-38.2, 25.6, 984.1), (1.7,116.2, 988.3), and the corresponding center of mass coordinates are (5.84, 79.32, 987.16); and the measured value of the center of mass coordinates in the current state is (8.26, 80.51, 989.73). By subtracting the reference center of mass from the current center of mass, it can be seen that the piston has drifted 2.42 mm, 1.19 mm, and 2.57 mm in the X, Y, and Z directions, respectively.

[0094] In summary, by fitting the identification points to obtain the spatial plane equation of the piston top surface, and then solving the Euler angle deviation between the plane normal vector and the reference state, the torsion of the piston around the three coordinate axes can be accurately calculated; and by comparing the changes in the center of mass coordinates, the drift of the piston in three directions can be obtained. This method makes full use of the spatial coordinate information of multiple identification points, avoids the calculation distortion caused by single-point measurement errors through plane fitting and center of mass calculation, and can fully reflect the motion state of the piston. In actual tests, using 4-8 identification points, an evaluation accuracy of better than 0.5 mm drift and 0.2 degrees torsion can be achieved. Correlating the changing trends of torsion and drift with the failure modes of reciprocating compressors is helpful to achieve state monitoring and early warning of gas tank pistons.

[0095] In an optional embodiment,

[0096] The piston torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order. The trend prediction model of the piston operation status is established by using time series analysis and machine learning algorithms. By analyzing the dynamic characteristics of the change rate and acceleration of the piston torsion and drift, the remaining service life and failure risk of the piston are predicted, including:

[0097] The piston torsion angle and drift distance data obtained from each measurement are stored in a database in chronological order to construct a time series data set containing the torsion angle and drift distance;

[0098] Based on the time series data set, the first-order differential features of the piston torsion angular velocity and drift velocity are obtained by calculating the difference between the torsion angle and the drift distance at adjacent sampling moments. The second-order differential features of the piston torsion angular acceleration and drift acceleration are further obtained by calculating the difference between the first-order differential features. A piston state feature data set including torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration is constructed.

[0099] Using machine learning algorithms, the piston state feature data set is used as input, and the torsion angle and drift distance at multiple future sampling moments are used as output to train a trend prediction model for the piston operation state.

[0100] The trained piston running state trend prediction model is used to predict the piston torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration at multiple sampling moments in the future;

[0101] Analyze and predict the changing trends of the piston torsion angular velocity, drift speed, torsion angular acceleration and drift acceleration. When the torsion angular velocity or drift speed continues to increase, or the torsion angular acceleration and drift acceleration fluctuate abnormally, it is determined that the piston is at risk of failure.

[0102] Based on the predicted piston torsion angle and drift distance, the estimated time until the piston reaches the preset failure threshold is calculated, and the estimated time is used as the remaining service life of the piston.

[0103] For example, in order to achieve trend prediction and life estimation of the piston operation status, it is necessary to perform time series analysis and machine learning modeling on the continuously collected torsion angle and drift distance data. First, the torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order to construct a time series data set reflecting the changes in the piston motion state. For example, a reciprocating compressor is continuously monitored for 1,000 hours, and the torsion angle of the piston around the X, Y, and Z axes and the drift distance along the X, Y, and Z directions are recorded every hour. A total of 1,000 six-dimensional state data are obtained, which are sorted into a data table according to the collection time.

[0104] Then, by performing differential operations on the state time series data, the first-order differential features reflecting the torsion angular velocity and drift velocity, as well as the second-order differential features reflecting the torsion angular acceleration and drift acceleration are obtained. The first-order differential features are calculated by subtracting the value corresponding to the previous sampling moment from the torsion angle or drift distance at the current sampling moment, reflecting the instantaneous rate of change of the state quantity; the second-order differential features are calculated by subtracting the first-order differential values ​​of two adjacent sampling moments, reflecting the instantaneous rate of change of the state quantity change rate. The original torsion angle and drift distance data are spliced ​​with the calculated first-order and second-order differential features by column to form a complete piston state feature data set, with each row corresponding to a sampling moment and each column corresponding to a state feature.

[0105] Next, the sequence prediction model in machine learning is used, and the piston state feature data set is used as the training sample to establish a trend prediction model for future states. The feature vector at each sampling moment is used as input, and the torsion angle and drift distance at several future sampling moments are used as output. The long short-term memory neural network and other algorithms are used to adjust the network weight parameters to minimize the error between the predicted value and the true value, and obtain a trained state trend prediction model. For example, take 100 consecutive hours of piston state data as the input sequence, predict the torsion angle and drift distance in the next 10 hours, and continuously slide the input sequence window to build samples and train the model until the prediction accuracy of the model on the verification data set meets the requirements.

[0106] Using the trained prediction model, the trend prediction of the newly collected piston state feature sequence is carried out, and the remaining life and failure risk of the piston are judged by analyzing the prediction results. By inputting the state data of the last 100 hours into the prediction model, the predicted value sequence of the piston torsion angle, drift distance and its first-order and second-order differential characteristics in the next 10 hours can be obtained. By calculating the statistical characteristics of the predicted values, the deterioration trend of the piston state is evaluated. When the mean or variance of the predicted torsion angular velocity or drift velocity exceeds the normal range, such as the mean increases from 0.1 mm per hour to 0.5 mm, and the variance increases from 0.05 to 0.2 per hour, it indicates that the torsion or drift process is accelerated, warning that the piston has a greater risk of failure. If the predicted torsion angular acceleration or drift acceleration has abnormal fluctuations, such as the acceleration value that is continuously greater than 0.01° / hour^2 or 0.01 mm / hour^2, it also indicates that the piston state is deteriorating.

[0107] In addition, by extrapolating the predicted value, the remaining service life of the piston can be estimated. Assuming that the torsional failure threshold of the piston around the X-axis is 5 degrees, the current torsion angle is 1.2 degrees, and it is predicted to continue to increase at a rate of 0.05 degrees / hour in the future, it can be estimated that the failure threshold will be reached after another 76 hours of operation, so the estimated remaining service life of the piston is 76 hours. The estimated life of the torsion angle and drift distance in three directions is combined, and the minimum value is taken as the overall remaining service life of the piston.

[0108] In summary, by constructing a time series feature data set of piston state parameters, using machine learning algorithms to establish a trend prediction model, and analyzing the statistical characteristics and change trends of the prediction results, the remaining service life and failure risk of the reciprocating compressor piston can be quantitatively evaluated. This method fully exploits the degradation laws contained in the continuous monitoring data, improves the reliability of the prediction through cross-validation of multiple state parameters, and can promptly detect abnormal signs of the piston's operating state, providing strong support for compressor state maintenance and fault prevention.

[0109] In an optional embodiment,

[0110] The method further comprises:

[0111] Using the trained piston torsion angle and drift distance trend prediction model, the piston torsion angle, drift distance and corresponding high-order differential features at multiple future sampling moments are predicted to obtain a series of prediction result sequences at future moments.

[0112] The predicted piston torsion angle, drift distance and their high-order differential characteristic sequences are trend analyzed, and the statistics in the prediction time domain are calculated, including the average value, variance and change rate of each state characteristic. The deterioration degree of the piston running state is judged by analyzing the change trends of the torsion angular velocity, drift velocity, angular acceleration and displacement acceleration.

[0113] Based on the prediction results and trend analysis, the failure risk of the piston is evaluated, and the warning thresholds and failure thresholds of the torsional angular velocity, drift velocity, angular acceleration and displacement acceleration are set. The predicted value is compared with the threshold. When the predicted value exceeds the warning threshold, a warning signal is issued. When the predicted value exceeds the failure threshold, it is determined that the piston has a high failure risk.

[0114] Combine the prediction results and the failure threshold to estimate the remaining service life of the piston. Calculate the time when the torsion angle and drift distance in the prediction results exceed their respective failure thresholds for the first time, and take the smaller value of the two as the estimated remaining service life of the piston.

[0115] Based on the piston failure risk assessment and remaining service life prediction results, the operation and maintenance strategy is dynamically adjusted. The monitoring frequency is increased for pistons at the warning level, maintenance or replacement is arranged for high-risk pistons, and maintenance time and priority are reasonably arranged for pistons with different remaining service lives. The piston failure data is fed back to the prediction model to continuously optimize the model performance.

[0116] For example, after establishing the piston operation state trend prediction model, the trained model is used to predict the torsion angle, drift distance and its high-order differential features in the future period of time, and a prediction result sequence reflecting the future operation state change of the piston is obtained. For example, using the trained long short-term memory neural network model, with the piston state data of the last 100 hours as input, the torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration values ​​every 1 hour in the next 50 hours are predicted, and a total of 50 groups of 6-dimensional prediction result vectors are obtained.

[0117] Statistical analysis is performed on the multiple state feature sequences obtained by prediction, and the mean, variance, and change rate of each feature in the prediction time domain are calculated to comprehensively evaluate the deterioration trend of the piston operation state. The mean reflects the average level of the feature, the variance reflects the degree of fluctuation of the feature, and the change rate reflects the speed of change of the feature. Taking the torsional angular velocity as an example, if the mean value is predicted to be 0.05° / hour, the variance is 0.01°^2 / hour^2, and the change rate is 0.001° / hour^2 in the next 50 hours, it means that the piston rotation tends to be stable; if the mean value of the angular velocity suddenly increases to 0.3° / hour, the variance rises to 0.05°^2 / hour^2, and the change rate is 0.015° / hour^2, then the warning torsional state deteriorates. By setting the normal range of angular velocity, angular acceleration, displacement velocity, and displacement acceleration, the abnormal characteristics of the deterioration of the piston state in the prediction time domain can be identified.

[0118] Furthermore, the prediction results, trend analysis and failure thresholds are comprehensively considered to quantitatively evaluate the failure risk and remaining service life of the piston. According to the statistical values ​​of the state parameters of the piston failure in the past, the warning thresholds and failure thresholds of the torsional angular velocity, drift velocity, torsional angular acceleration and drift acceleration are set. The statistical quantities such as the mean and variance of the predicted values ​​are compared with the warning thresholds. If the warning threshold is exceeded, such as the mean value of the torsional angular velocity is higher than 0.2° / hour and the mean value of the angular acceleration is higher than 0.01° / hour^2, a piston status warning signal is issued; if the failure threshold is exceeded, such as the mean value of the displacement velocity is greater than 1 mm / hour and the mean value of the displacement acceleration is higher than 0.1 mm / hour^2, it is determined that the piston has a high failure risk within the predicted time period.

[0119] At the same time, the estimated time when the piston exceeds the failure threshold for the first time in each state parameter is determined, and the minimum value is taken as the overall remaining service life of the piston. Assuming that the failure threshold of the torsion angle is 5° and the failure threshold of the drift distance is 5 mm, the prediction results show that the torsion angle will reach 5.2° in the next 30 hours and the drift distance will reach 5.4 mm in the next 35 hours. The estimated remaining service life of the piston is the smaller value of the two, that is, 30 hours. If the predicted torsion angle or drift distance does not reach the failure threshold within the prediction time domain, the remaining service life of the piston is greater than the prediction time. Taking 50 hours as an example, the remaining service life of the piston is estimated to be "greater than 50 hours". Giving an estimated result of "greater than XX hours" can avoid inflated life estimates due to insufficient prediction time span.

[0120] Finally, based on the failure risk assessment and remaining life prediction results, the condition monitoring and maintenance strategy of the piston is dynamically optimized. For pistons at the warning level, if the estimated remaining life is less than 100 hours, or indicators such as angular velocity and displacement speed exceed the warning threshold, the frequency of monitoring data collection will be doubled, such as from once an hour to once every half an hour, to track the accelerated process of piston condition deterioration; for high-risk pistons, if the estimated remaining life is less than 10 hours, or the status indicators exceed the failure threshold, the machine will be immediately shut down for inspection or replacement to prevent accidents caused by running with problems; for pistons with different remaining service lives, the maintenance time and priority are reasonably arranged. For example, pistons with an estimated life of less than 50 hours are listed as key maintenance objects in the near future, and pistons with an estimated life of more than 100 hours can have their maintenance cycle appropriately extended. The condition monitoring data and failure records before the piston fails are fed back to the prediction model, and the prediction accuracy is improved by retraining the model parameters to achieve a dynamic closed loop of piston condition prediction, life assessment and maintenance strategy optimization.

[0121] In summary, the machine learning trend prediction model is used to obtain the state prediction results of the piston at multiple moments in the future. By analyzing the statistical characteristics and change trends of the prediction results, combined with the warning threshold and failure threshold, the failure risk level and remaining service life of the piston can be quantitatively evaluated, and the monitoring and maintenance strategies can be dynamically adjusted accordingly to achieve predictive maintenance of the piston state. This method is based on big data drive, and by continuously accumulating the monitoring data of the piston throughout its life cycle and iteratively optimizing the prediction model, it continuously improves the timeliness and accuracy of fault warning, and provides a reliable decision-making basis for ensuring the safe and efficient operation of reciprocating compressors.

[0122] In an optional embodiment,

[0123] A series of strain gauges are installed on the pipe wall of the gas cabinet to measure the stress distribution and deformation of the gas cabinet in real time during operation. The strain data is correlated with the piston torsion and drift data to build a comprehensive assessment model for the safety status of the gas cabinet. The health level and remaining life of the gas cabinet are comprehensively assessed, including:

[0124] A series of strain gauges are arranged in the high stress area of ​​the gas cabinet pipe wall and the position where the stress gradient is greater than the preset gradient threshold. The measurement direction of the strain gauge is consistent with the main stress direction of the gas cabinet. The stress distribution and deformation during the operation of the gas cabinet are measured in real time to obtain the strain data of the gas cabinet pipe wall;

[0125] Preprocess the gas cabinet pipe wall strain data and use filtering algorithm to remove high-frequency interference and low-frequency drift in the strain signal;

[0126] Extract multi-domain features of gas cabinet wall strain data, including extracting the statistical features of the mean, variance, and peak-to-peak value of the strain signal, extracting the frequency domain features of the spectral distribution and main frequency amplitude of the strain signal, extracting the wavelet coefficients of the strain signal, and the time-frequency features of the Hilbert-Huang transform;

[0127] Obtain the monitoring data of the torsion angle and drift distance of the gas tank piston, perform correlation analysis on the torsion angle and drift distance data with the gas tank pipe wall strain data, and calculate the correlation coefficient matrix between the strain characteristics and the torsion angle and drift distance characteristics;

[0128] Adopting the dimensionality reduction algorithm of principal component analysis or independent component analysis, the gas tank wall strain data and the piston torsion angle and drift distance data are integrated to extract low-dimensional key features that comprehensively reflect the safety status of the gas tank;

[0129] Based on the low-dimensional key features of the gas cabinet safety status, a gas cabinet safety status classification model is established using support vector machine or random forest algorithm to divide the gas cabinet status into normal, warning and dangerous levels;

[0130] Based on the low-dimensional key features of the gas tank safety status, a neural network or long short-term memory network algorithm is used to establish a gas tank remaining life prediction model. By learning the degradation trend in historical data, the future health status and failure time of the gas tank are predicted;

[0131] Taking into account the gasholder safety status classification results and remaining life prediction results, a quantitative assessment of the health level of the gasholder is conducted to generate a comprehensive assessment report on the gasholder safety status.

[0132] For example, in order to comprehensively evaluate the health status and remaining life of the reciprocating compressor gas cabinet, it is necessary to arrange strain sensors on the gas cabinet pipe wall, monitor the stress distribution and deformation of the gas cabinet in real time, and integrate the strain monitoring data with parameters such as piston torsion angle and drift distance for analysis to build a comprehensive assessment model for the safety status of the gas cabinet. First, according to the finite element analysis results of the gas cabinet, a strain gauge array is arranged in the area where the von Mises equivalent stress of the gas cabinet pipe wall is large and the location where the stress gradient is significant. The measurement direction of the strain gauge is aligned with the circumferential and axial principal stress directions of the gas cabinet pipe wall, and the dynamic strain data of the gas cabinet during operation is collected in real time. For example, 8 strain gauges are arranged circumferentially in the middle of the gas cabinet, and 4 strain gauges are arranged axially to form a strain sensor array of 4 rows and 8 columns. The sampling frequency is set to 1kHz, and the data of 32 strain sensors are recorded every 1 second.

[0133] The collected gas cabinet pipe wall strain data is preprocessed, and the strain signal is smoothed by the median filter algorithm to remove the spike noise introduced by high-frequency mechanical vibration and electromagnetic interference, and then the baseline drift and temperature drift of the strain signal are eliminated by the polynomial fitting algorithm. After preprocessing, the stable gas cabinet pipe wall dynamic strain time series data is obtained. Then, the multidimensional features of the strain signal are extracted from the time domain, frequency domain and time-frequency domain. The time domain features include the mean, standard deviation, peak-to-peak value, root mean square value and other statistics of each strain sensor signal, reflecting the average stress level and dynamic stress fluctuation range of the gas cabinet pipe wall. The frequency domain features include the spectrum distribution of each strain signal, the amplitude corresponding to the main frequency point, the amplitude proportion of each harmonic component, etc., reflecting the frequency characteristics of the forced vibration of the gas cabinet. The time-frequency domain features extract the spectrum evolution characteristics at different time scales by performing short-time Fourier transform or wavelet packet decomposition on the strain signal, such as the change trend of the root mean square value of the wavelet coefficient over time, etc., reflecting the non-stationary characteristics of the dynamic response of the gas cabinet.

[0134] The monitoring data of the piston torsion angle and drift distance in the same time period are obtained, and the data are correlated with the strain signal of the gas cabinet wall. The Pearson correlation coefficient between the statistical characteristics of the strain signal, such as the mean and variance, and the torsion angle and drift distance characteristics is calculated, and the correlation between the strain characteristics and the torsion characteristics and the drift characteristics is determined through a significance test. The mapping relationship between the torsion angle, drift distance and the gas cabinet strain is constructed to analyze the cause of the local high stress on the gas cabinet wall. If the piston torsion angle is too large, it may cause local deformation of the gas cabinet, resulting in a significant increase in the amplitude and fluctuation of the strain signal; if the piston drift exceeds the standard, the gas cabinet may be unevenly stressed, causing asymmetric stress distribution. By analyzing the correlation between strain and torsion and drift parameters, the influencing factors of the health status of the gas cabinet are revealed.

[0135] In order to eliminate the information redundancy and noise interference between the piston motion characteristics and the gas tank strain characteristics, the principal component analysis algorithm is used to reduce the dimension of multi-dimensional parameters. The strain statistical characteristics, frequency domain characteristics, time-frequency characteristics of different positions on the gas tank wall, the piston torsion angle and its first-order difference and second-order difference characteristics, the piston drift distance and its first-order difference and second-order difference characteristics constitute a high-dimensional input vector, which is mapped to a low-dimensional feature space through principal component transformation, and the first principal components that have a significant impact on the health status of the gas tank are extracted as comprehensive evaluation indicators. For example, three principal components are obtained through principal component analysis, of which the first principal component is dominated by the mean strain value and the mean piston torsion velocity in the middle of the gas tank, the second principal component is dominated by the peak-to-peak value of the strain in the upper part of the gas tank and the piston drift acceleration, and the third principal component is dominated by the main frequency amplitude of the strain spectrum in the lower part of the gas tank and the piston torsion acceleration. The cumulative variance contribution rate of each principal component exceeds 85%.

[0136] Based on the extracted low-dimensional features of the gas cabinet status, the support vector machine algorithm is used to construct a gas cabinet safety status classification model. According to the time point when the abnormal working conditions of the gas cabinet occurred in the historical operation data, the gas cabinet status is divided into three levels: normal, warning and dangerous, and the corresponding low-dimensional feature space forms three clusters. The support vector machine algorithm is used to find the maximum interval hyperplane between the clusters of different states to obtain the gas cabinet status classification learning model. When the newly collected gas cabinet operation data falls into the normal state cluster range after principal component analysis and dimensionality reduction, the gas cabinet operation is diagnosed to be safe; if it falls into the warning or dangerous state cluster range, it indicates that the gas cabinet has potential abnormal risks or safety hazards.

[0137] Furthermore, a neural network algorithm is used to establish a prediction model for the remaining life of a gas cabinet. With the low-dimensional feature vector of the gas cabinet as input and the remaining service life of the gas cabinet at the corresponding moment as output, a feedforward neural network is constructed to learn the complex mapping relationship between the degradation law and life of the gas cabinet in historical data. According to the designed service life of the gas cabinet and the actual maintenance records, the real remaining life value of the gas cabinet at each historical moment is marked, and the neural network prediction model is trained. After the model training is completed, when new low-dimensional feature data of the gas cabinet is input, the neural network model can infer and predict the current remaining life of the gas cabinet. Combined with historical maintenance data, the prediction results are dynamically corrected to continuously optimize the performance of the prediction model. For example, a reciprocating compressor gas cabinet that has been in operation for 10 years has a designed service life of 20 years, and the neural network model predicts that its remaining life is 8.5 years. If subsequent maintenance finds that its actual use status is worse than expected and needs to be replaced 2 years in advance, the real life data will be fed back to the model, the network weight will be fine-tuned, and the prediction deviation will be corrected.

[0138] Finally, the health level of the gasholder is quantitatively evaluated by combining the classification of gasholder safety status and the remaining life prediction results. For gasholders classified as normal and with a remaining life of more than 5 years, their health status is diagnosed as good, and regular maintenance can be carried out as planned; for gasholders classified as normal but with a remaining life of less than 5 years, their health status is assessed to be general, and their performance degradation trend must be closely monitored and the maintenance cycle should be appropriately shortened; for gasholders classified as warning status and with a remaining life of less than 3 years, their health status is assessed to be poor, and it is recommended to arrange maintenance as soon as possible, and adjust production operation parameters if necessary to avoid accelerated damage; for gasholders classified as dangerous status and with a remaining life of less than 1 year, their health status is assessed to be extremely poor, and they must be shut down for maintenance or scrapped and replaced immediately to prevent accidents caused by operation with illness. The results of the graded assessment of the health status of the gasholder are formed into a comprehensive safety status assessment report to provide a quantitative basis for compressor management decisions.

[0139] In summary, by arranging strain sensor arrays on the gas cabinet pipe wall, real-time monitoring of the dynamic stress distribution and deformation of the gas cabinet, and integrating piston motion parameters for correlation analysis, the key features of gas cabinet state degradation are extracted, and a gas cabinet safety state classification and remaining life prediction model is constructed, which can realize dynamic assessment and early warning of the health level of the gas cabinet. This method makes full use of multi-source heterogeneous sensor monitoring big data, comprehensively considers key factors affecting gas cabinet failure such as local stress of the gas cabinet and piston motion parameters, and directly establishes the intrinsic correlation between state characteristics and remaining life through end-to-end deep learning algorithms, overcoming the limitations of traditional gas cabinet state assessment methods that rely on manual experience, and can significantly improve the level of intelligent health management of reciprocating compressor gas cabinets.

[0140] Figure 2 FIG. 1 is a schematic diagram of the structure of a gas cabinet piston torsion and drift measurement system based on image recognition according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0141] The first unit is used to set at least three non-collinear identification points on the top of the piston of the gas tank, wherein the identification points are markers with specific colors or patterns, and the initial three-dimensional spatial coordinates of the identification points are recorded as reference coordinates; at the same time, a binocular stereo camera is installed at a fixed position of the gas tank so that its optical axis is parallel to the axis of the gas tank piston, and the internal and external parameters of the camera are determined by binocular stereo calibration; the binocular stereo camera is used to synchronously collect continuous images of the left and right paths during the operation of the gas tank piston, and the collected image pairs are transmitted to the computer for stereo matching, and the corresponding identification points are identified in the left and right views through epipolar constraints and similarity measurements, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle;

[0142] The second unit is used to compare the current three-dimensional coordinates of the identification point with the reference coordinates, fit the spatial plane equation of the piston top surface by the least square method, calculate the Euler angle change of the plane normal vector relative to the reference state, and obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, calculate the offset of the center of mass coordinates of the identification point relative to the reference position, and obtain the drift distance of the piston in three directions;

[0143] The third unit is used to store the piston torsion angle and drift distance data obtained from each measurement into a database in chronological order, and use time series analysis and machine learning algorithms to establish a trend prediction model for the piston operating status. By analyzing the changing rate of piston torsion and drift and the dynamic characteristics of acceleration, the remaining service life and failure risk of the piston can be predicted. A series of strain gauges are installed on the pipe wall of the gas holder to measure the stress distribution and deformation of the gas holder during operation in real time, and the strain data is correlated with the piston torsion and drift data for analysis to construct a comprehensive assessment model for the safety status of the gas holder to comprehensively assess the health level and remaining life of the gas holder.

[0144] According to a third aspect of the embodiments of the present invention,

[0145] An electronic device is provided, comprising:

[0146] processor;

[0147] a memory for storing processor-executable instructions;

[0148] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0149] A fourth aspect of the embodiments of the present invention is:

[0150] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0151] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for measuring gas tank piston torsion and drift based on image recognition, characterized in that: include: At least three non-collinear identification points are set on the top of the piston of the gas tank, and the identification points are markers with specific colors or patterns. The initial three-dimensional spatial coordinates of the identification points are recorded as reference coordinates; at the same time, a binocular stereo camera is installed at a fixed position of the gas tank so that its optical axis is parallel to the axis of the gas tank piston, and the internal and external parameters of the camera are determined by binocular stereo calibration; the binocular stereo camera is used to synchronously collect continuous images of the left and right paths during the operation of the gas tank piston, and the collected image pairs are transmitted to the computer for stereo matching. Through epipolar line constraints and similarity measurements, the corresponding identification points are identified in the left and right views, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle; The current three-dimensional coordinates of the identification point are compared with the reference coordinates, and the spatial plane equation of the piston top surface is fitted by the least square method. The Euler angle change of the plane normal vector relative to the reference state is calculated to obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, the offset of the center of mass coordinates of the identification point relative to the reference position is calculated to obtain the drift distance of the piston in three directions; The piston torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order. A trend prediction model for the piston operating status is established using time series analysis and machine learning algorithms. The remaining service life and failure risk of the piston are predicted by analyzing the dynamic characteristics of the changing rate of piston torsion and drift and acceleration. A series of strain gauges are installed on the pipe wall of the gas holder to measure the stress distribution and deformation of the gas holder during operation in real time. The strain data are correlated with the piston torsion and drift data for analysis, and a comprehensive assessment model for the safety status of the gas holder is constructed to comprehensively assess the health level and remaining service life of the gas holder.

2. The method according to claim 1, characterized in that The collected image pairs are transferred to the computer for stereo matching. The corresponding identification points are identified in the left and right views through epipolar constraints and similarity measurements, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle, including: According to the projection center of the binocular stereo camera and the pixel coordinates of the points to be matched in the left and right views, using the essential matrix obtained by pre-calibration, an epipolar constraint equation is established for the points to be matched in the left and right views to satisfy the three-point coplanar constraint; Using the established epipolar constraint equation, for any pixel point to be matched in the left view, determine the straight line where its corresponding point in the right view is located, and reduce the right view to be matched area to the area near the straight line; Taking the matching point in the left and right views as the center, pixel blocks of the same size are intercepted respectively, and the normalized cross-correlation coefficient of the two pixel blocks is calculated as a measure of the similarity between the two pixel blocks. The calculation method of the normalized cross-correlation coefficient is: ； Where NCC(p,q) represents the normalized cross-correlation coefficient, p and q represent two pixel blocks to be matched, which come from the areas to be matched in the left and right views, respectively, and n represents the size of the pixel block. , Represents a pixel block W p and W q The grayscale value at the middle coordinate (i, j), where i and j are relative coordinates inside the pixel block, ranging from 0 to n-1; , Represents pixel blocks W p and W q The average gray value of By setting a normalized cross-correlation coefficient threshold, the pixel points exceeding the normalized cross-correlation coefficient threshold are screened out from all candidate matching pixels to obtain the best matching pixel coordinates of the identification points in the left and right views; The pixel coordinates of the matching points in the left and right views and the projection matrix of the binocular camera are substituted into the spatial coordinate solution equation group based on the triangulation principle, and the equation group is solved by least squares optimization to obtain the three-dimensional spatial coordinates of the identification points. The objective function of the least squares optimization is the sum of the squares of the Euclidean distances between the pixel coordinates of the matching point and its spatial coordinates obtained by back-projection of the projection matrix on the projection pixel coordinates of the left and right views.

3. The method according to claim 1, characterized in that: The current three-dimensional coordinates of the identification point are compared with the reference coordinates, and the spatial plane equation of the piston top surface is fitted by the least square method. The Euler angle change of the plane normal vector relative to the reference state is calculated to obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; At the same time, the offset of the center of mass coordinates of the identification point relative to the reference position is calculated, and the drift distances of the piston in three directions are obtained, including: The three-dimensional spatial coordinates of multiple identification points in the current state are compared with the coordinates in the reference state, and the spatial plane equation of the piston top surface in the current state is fitted by the least squares method, where the plane equation is in the form of ax+by+cz+d=0, (a, b, c) is the plane normal vector, and d is the distance from the plane to the origin. Multiple linear equations are obtained by substituting the coordinates of the identification points into the plane equation and written in matrix form. The homogeneous linear equations are solved by singular value decomposition to obtain the plane normal vector (a, b, c) and the distance d; Calculate the angle between the plane normal vector in the current state and the normal vector in the reference state, decompose the angle into rotation angles around the X-axis, Y-axis and Z-axis, namely, Euler angles, and establish an equation by equalizing the current normal vector to the product of the reference normal vector and the rotation matrix, where the rotation matrix is ​​represented by the multiplication of the rotation matrices around the three coordinate axes. By solving the relationship between the rotation matrix elements and the Euler angles, the torsion angles of the piston around the X-axis, Y-axis and Z-axis are obtained; Calculate the center of mass coordinates of multiple identification points in the reference state and the current state, and subtract the reference center of mass coordinates from the current center of mass coordinates to obtain the drift distance of the piston in the X, Y, and Z directions, where the center of mass coordinates are equal to the arithmetic mean of the coordinates of all identification points in the corresponding directions.

4. The method according to claim 1, characterized in that The piston torsion angle and drift distance data obtained from each measurement are stored in the database in chronological order. The trend prediction model of the piston operation status is established by using time series analysis and machine learning algorithms. By analyzing the dynamic characteristics of the change rate and acceleration of the piston torsion and drift, the remaining service life and failure risk of the piston are predicted, including: The piston torsion angle and drift distance data obtained from each measurement are stored in a database in chronological order to construct a time series data set containing the torsion angle and drift distance; Based on the time series data set, the first-order differential features of the piston torsion angular velocity and drift velocity are obtained by calculating the difference between the torsion angle and the drift distance at adjacent sampling moments. The second-order differential features of the piston torsion angular acceleration and drift acceleration are further obtained by calculating the difference between the first-order differential features. A piston state feature data set including torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration is constructed. Using machine learning algorithms, the piston state feature data set is used as input, and the torsion angle and drift distance at multiple future sampling moments are used as output to train a trend prediction model for the piston operation state. The trained piston running state trend prediction model is used to predict the piston torsion angle, drift distance, torsion angular velocity, drift velocity, torsion angular acceleration and drift acceleration at multiple sampling moments in the future; Analyze and predict the changing trends of the piston torsion angular velocity, drift speed, torsion angular acceleration and drift acceleration. When the torsion angular velocity or drift speed continues to increase, or when the torsion angular acceleration and drift acceleration fluctuate abnormally, it is determined that the piston is at risk of failure. Based on the predicted piston torsion angle and drift distance, the estimated time until the piston reaches the preset failure threshold is calculated, and the estimated time is used as the remaining service life of the piston.

5. The method according to claim 1, characterized in that The method further comprises: Using the trained piston torsion angle and drift distance trend prediction model, the piston torsion angle, drift distance and corresponding high-order differential features at multiple future sampling moments are predicted to obtain a series of prediction result sequences at future moments. The predicted piston torsion angle, drift distance and their high-order differential characteristic sequences are trend analyzed, and the statistics in the prediction time domain are calculated, including the average value, variance and change rate of each state characteristic. The deterioration degree of the piston running state is judged by analyzing the change trends of the torsion angular velocity, drift velocity, angular acceleration and displacement acceleration. Based on the prediction results and trend analysis, the failure risk of the piston is evaluated, and the warning thresholds and failure thresholds of the torsional angular velocity, drift velocity, angular acceleration and displacement acceleration are set. The predicted value is compared with the threshold. When the predicted value exceeds the warning threshold, a warning signal is issued. When the predicted value exceeds the failure threshold, it is determined that the piston has a high failure risk. Combine the prediction results and the failure threshold to estimate the remaining service life of the piston. Calculate the time when the torsion angle and drift distance in the prediction results exceed their respective failure thresholds for the first time, and take the smaller value of the two as the estimated remaining service life of the piston. Based on the piston failure risk assessment and remaining service life prediction results, the operation and maintenance strategy is dynamically adjusted. The monitoring frequency is increased for pistons at the warning level, maintenance or replacement is arranged for high-risk pistons, and maintenance time and priority are reasonably arranged for pistons with different remaining service lives. The piston failure data is fed back to the prediction model to continuously optimize the model performance.

6. The method according to claim 1, characterized in that A series of strain gauges are installed on the pipe wall of the gas cabinet to measure the stress distribution and deformation of the gas cabinet in real time during operation. The strain data is correlated with the piston torsion and drift data to build a comprehensive assessment model for the safety status of the gas cabinet. The health level and remaining life of the gas cabinet are comprehensively assessed, including: A series of strain gauges are arranged in the high stress area of ​​the gas cabinet pipe wall and the position where the stress gradient is greater than the preset gradient threshold. The measurement direction of the strain gauge is consistent with the main stress direction of the gas cabinet. The stress distribution and deformation during the operation of the gas cabinet are measured in real time to obtain the strain data of the gas cabinet pipe wall; Preprocess the gas cabinet pipe wall strain data and use filtering algorithm to remove high-frequency interference and low-frequency drift in the strain signal; Extract multi-domain features of gas cabinet wall strain data, including extracting the statistical features of the mean, variance, and peak-to-peak value of the strain signal, extracting the frequency domain features of the spectral distribution and main frequency amplitude of the strain signal, extracting the wavelet coefficients of the strain signal, and the time-frequency features of the Hilbert-Huang transform; Obtain the monitoring data of the torsion angle and drift distance of the gas tank piston, perform correlation analysis on the torsion angle and drift distance data with the gas tank pipe wall strain data, and calculate the correlation coefficient matrix between the strain characteristics and the torsion angle and drift distance characteristics; Adopting the dimensionality reduction algorithm of principal component analysis or independent component analysis, the gas tank wall strain data and the piston torsion angle and drift distance data are integrated to extract low-dimensional key features that comprehensively reflect the safety status of the gas tank; Based on the low-dimensional key features of the gas cabinet safety status, a gas cabinet safety status classification model is established using support vector machine or random forest algorithm to divide the gas cabinet status into normal, warning and dangerous levels; Based on the low-dimensional key features of the gas tank safety status, a neural network or long short-term memory network algorithm is used to establish a gas tank remaining life prediction model. By learning the degradation trend in historical data, the future health status and failure time of the gas tank are predicted; Taking into account the gasholder safety status classification results and remaining life prediction results, a quantitative assessment of the health level of the gasholder is conducted to generate a comprehensive assessment report on the gasholder safety status.

7. A gas cabinet piston torsion and drift measurement system based on image recognition, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to set at least three non-collinear identification points on the top of the piston of the gas tank, wherein the identification points are markers with specific colors or patterns, and the initial three-dimensional spatial coordinates of the identification points are recorded as reference coordinates; at the same time, a binocular stereo camera is installed at a fixed position of the gas tank so that its optical axis is parallel to the axis of the gas tank piston, and the internal and external parameters of the camera are determined by binocular stereo calibration; the binocular stereo camera is used to synchronously collect continuous images of the left and right paths during the operation of the gas tank piston, and the collected image pairs are transmitted to the computer for stereo matching, and the corresponding identification points are identified in the left and right views through epipolar constraints and similarity measurements, and the three-dimensional spatial coordinates of each identification point are calculated using the triangulation principle; The second unit is used to compare the current three-dimensional coordinates of the identification point with the reference coordinates, fit the spatial plane equation of the piston top surface by the least square method, calculate the Euler angle change of the plane normal vector relative to the reference state, and obtain the torsion angle of the piston around the X-axis, Y-axis and Z-axis; at the same time, calculate the offset of the center of mass coordinates of the identification point relative to the reference position, and obtain the drift distance of the piston in three directions; The third unit is used to store the piston torsion angle and drift distance data obtained from each measurement into a database in chronological order, and use time series analysis and machine learning algorithms to establish a trend prediction model for the piston operating status. By analyzing the changing rate of piston torsion and drift and the dynamic characteristics of acceleration, the remaining service life and failure risk of the piston can be predicted. A series of strain gauges are installed on the pipe wall of the gas holder to measure the stress distribution and deformation of the gas holder during operation in real time, and the strain data is correlated with the piston torsion and drift data for analysis to construct a comprehensive assessment model for the safety status of the gas holder to comprehensively assess the health level and remaining life of the gas holder.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images