Petrochemical equipment pitting corrosion identification and quantitative analysis method and system

Abstract

The invention discloses a petrochemical equipment pitting identification and quantitative analysis method and system, and belongs to the technical field of nondestructive testing of petrochemical equipment.The method comprises the steps that petrochemical equipment pitting image data are collected, a neural network model based on deep learning is built, and a pitting identification model is obtained through training; establishing a corresponding relation between pixel coordinates of the pitting corrosion pits and actual coordinates, and converting pixel size information of the pitting corrosion pits into actual size information; identifying a to-be-identified pitting corrosion image by using the pitting corrosion identification model, outputting pixel coordinates of the identified pitting corrosion, performing conversion between the pixel coordinates and real coordinates, and calculating the density, opening and depth of a pitting corrosion pit; and evaluating the pitting corrosion severity of the pitting corrosion pit. According to the method provided by the invention, the pitting inspection efficiency and scientificity are improved, and technical support is provided for petrochemical equipment integrity defect management.

Description

Technical Field

[0001] The invention belongs to the technical field of nondestructive testing of petrochemical equipment, and in particular relates to a method and system for identifying and quantitatively analyzing pitting of petrochemical equipment. Background Art

[0002] After the metal components of petrochemical equipment are pitted, their cross-sectional stress increases, the structural strength weakens, and the bearing capacity decreases. The safe operation of the equipment will be adversely affected. Therefore, it is necessary to regularly inspect and evaluate the safety of the relevant equipment. The common method of pitting detection is manual inspection, that is, visually inspecting whether the equipment is corroded and using corresponding measuring tools to measure the maximum depth of the corrosion pit. The manual inspection method is time-consuming and labor-intensive, with low accuracy, and it is difficult to achieve quantitative analysis of the location, number, size and overall distribution of pitting pits.

[0003] In recent years, computer vision technology based on deep learning has made great progress and has been widely used in the field of defect detection. Deep learning methods have strong nonlinear fitting capabilities and are suitable for target detection under complex environmental conditions. As for the problem of pitting recognition, the difficulty lies in the difficulty of obtaining equipment pitting images, and the pitting pits are densely distributed and difficult to label. The above phenomenon leads to less high-quality pitting image data, which is not conducive to the learning and optimization of the recognition model.

[0004] On this basis, the present invention provides a method for pitting identification and quantitative analysis of petrochemical equipment, which expands the amount of pitting image data through manual annotation and model-generated pseudo-annotation to improve the accuracy of pitting identification. By utilizing the nonlinear fitting ability of the deep learning model, the problems of cluttered data background and densely distributed defects are overcome, and the detection of dense small targets under strong environmental factors is achieved. The correspondence between the pixel coordinates of the pitting pits and the actual world coordinates is established, providing a basis for the statistical analysis of the location, size, number and distribution of the pitting pits. According to the density, opening and depth of the pitting pits, a quantitative analysis of the severity of pitting of the equipment is achieved, providing data support for the standardized management and preventive maintenance management of the equipment. Summary of the invention

[0005] In view of the problems existing in the above-mentioned prior art, the present invention proposes a method and system for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment, which has a reasonable design, overcomes the shortcomings of the prior art, and has good effects.

[0006] In order to achieve the above purpose 1, the present invention adopts the following technical solution:

[0007] A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment, characterized in that it comprises the following steps:

[0008] S1. Collect pitting image data of petrochemical equipment, build a neural network model based on deep learning, and train to obtain a pitting recognition model;

[0009] S2, establishing a corresponding relationship between the pixel coordinates of the pit and the actual coordinates, and converting the pixel size information of the pit into the actual size information;

[0010] S3, using the pitting corrosion recognition model obtained in S1 to recognize the pitting corrosion image to be recognized, and outputting the pixel coordinates of the recognized pitting corrosion, converting the pixel coordinates to the real coordinates, and calculating the density, opening and depth of the pitting corrosion pit;

[0011] S4. Evaluate the severity of pitting corrosion.

[0012] Furthermore, the S1 includes the following sub-steps:

[0013] S1.1. Manually screen the collected pitting images and divide them into pitting images that are easy to label and pitting images that are difficult to label. Images containing 30 or less pits are considered to be easy to label images, and the remaining images are considered to be difficult to label images. The easy to label pitting images are manually labeled to obtain first image data.

[0014] S1.2, training the neural network model with the first image data to obtain a label generation model;

[0015] S1.3, using the model to identify the part of the pitting image that is difficult to annotate, and using the identification result as a pseudo-annotation to obtain the second image data;

[0016] S1.4. Retrain the neural network model using the first image data and the second image data to obtain a new annotation generation model;

[0017] S1.5, repeat S1.3 and S1.4 until all the pitting images that are difficult to mark are identified and the final second image data is obtained; use the first image data and the final second image data to retrain the neural network model to obtain a pitting recognition model, which is used to output the pixel coordinates of the pitting pits in the image.

[0018] Furthermore, S1 includes the following sub-steps: the neural network model is a target detection model or an instance segmentation model.

[0019] Furthermore, the collected pitting image data is no less than 1,000 pieces, the resolution is no less than 200 dpi, and a ruler is provided around the pitting pits on the image as a size reference.

[0020] Furthermore, the S2 comprises the following sub-steps:

[0021] S2.1. Collect pitting image data with a ruler, mark the ruler and the ruler scale, build a target detection model, and iterate the model multiple times with the collected image data to obtain a ruler recognition model;

[0022] S2.2. Establish a pixel coordinate system with the upper left corner of the pitting image as the origin, use the ruler recognition model to identify the ruler and ruler scale in the pitting image, and output the pixel coordinates of all ruler scale prediction boxes, denoted as {(x 11 ,y 11 , x 12 ,y 22 ), (x 21 ,y 21 , x 22 ,y 22 ),...,(x k1 ,y k1 , x k2 ,y k2 )}, where (x k1 ,y k1 ) is the coordinate of the upper left corner of the k-th prediction box in the image, (x k2 ,y k2 ) is the coordinate of the lower right corner of the kth prediction box; calculate the area S of each prediction box k , and remove the scales corresponding to the two smallest areas, S k The calculation process is expressed as:

[0023] S k =(x k2 -x k1 )×(y k2 -y k1 );

[0024] S2.3, if k-2 ≥ 2, calculate the center points of the remaining prediction boxes, and sort the prediction boxes according to the horizontal coordinates of the center points. The actual distance between the center points of two adjacent prediction boxes is recorded as L'. The corresponding relationship between the pixels and the actual coordinates is calculated by the coordinates of the two center points (X'1, Y'1), (X'2, Y'2) and trigonometric functions. The pixel length l' corresponding to one unit actual length is expressed as:

[0025]

[0026] If k-2<2, execute step S2.4;

[0027] S2.4. Artificially select two points on the image, the actual distance between the two points is L″, and the corresponding relationship between the pixels and the actual coordinates is calculated by the pixel coordinates (X″1, Y″1), (X″2, Y″2) of the two points and trigonometric functions. The pixel length l″ corresponding to one unit of actual length is expressed as:

[0028]

[0029] Furthermore, the S3 includes the following sub-steps:

[0030] S3.1, use the pitting recognition model established in S1 to identify the pitting image to be identified, output the pixel coordinates of the identified pitting prediction box, and convert them into real coordinates, which are recorded as:

[0031] S3.2. Approximate the pitting pit to a circle and calculate the radius of each pitting pit, which is recorded as {r1, r2, ..., r n}, the average radius is recorded as r, and the average opening s of the pit is calculated using the following formula:

[0032]

[0033]

[0034] s=πr 2 :

[0035] Among them, r n is the approximate radius of the nth pit; r is the average radius of all pits measured; s is the average opening of all pits measured;

[0036] S3.3. According to the coordinates of the pits, a rectangular pitting area is delineated in the image with the leftmost, rightmost, topmost and bottommost pits as the reference, and after converting the pixel length to the actual length, the area m of the pitting area is calculated, and the density of the pits is calculated. The calculation formula is:

[0037]

[0038] Where, ρ is the density of pits; n is the number of pits;

[0039] S3.4. Measure the actual depth of the pit manually and record it as h.

[0040] Furthermore, a series of different thresholds are set for ρ, s and h to comprehensively reflect the severity of corrosion.

[0041] In order to achieve the above purpose 2, the present invention adopts the following technical solution:

[0042] A petrochemical equipment pitting corrosion identification and quantitative analysis system, using the petrochemical equipment pitting corrosion identification and quantitative analysis method as described above, including an image acquisition module, a pitting corrosion identification module, a statistical analysis module and a quantitative evaluation module;

[0043] The image acquisition module is used to collect image data from the equipment to be inspected and transmit it to the pitting recognition module;

[0044] The pitting recognition module recognizes the received image data, marks each pitting with a rectangular frame, transmits the marked image to the quantitative evaluation module, and transmits the upper left corner coordinates and the lower right corner coordinates of the rectangular frame to the statistical analysis module;

[0045] The statistical analysis module is used to complete the conversion of pixel coordinates to real coordinates, complete the calculation of pit density and opening, and output the above values ​​and measured depth to the quantitative evaluation module;

[0046] The quantitative evaluation module records the received data and makes a comprehensive analysis of the severity of pitting corrosion according to different data values.

[0047] Beneficial technical effects brought by the present invention:

[0048] The present invention realizes the identification, statistical analysis and quantitative evaluation of pitting corrosion of equipment under complex environmental conditions; by introducing a computer vision method based on deep learning, pitting corrosion detection in complex environments is realized; by using manual annotation and pseudo-annotation generation methods, the data volume of high-quality pitting corrosion images is expanded to improve the accuracy of pitting corrosion identification; and by statistical analysis of the location, size, number and distribution of pitting corrosion pits, a comprehensive evaluation of the severity of pitting corrosion is realized. The method provided by the present invention improves the efficiency and scientificity of pitting corrosion inspection and provides technical support for the integrity defect management of petrochemical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment in the present invention;

[0050] Figure 2 It is a schematic diagram of pitting identification in the present invention;

[0051] Figure 3 This is a schematic diagram of the scale and scale identification in the present invention;

[0052] Figure 4 It is a schematic diagram of the conversion between pixel length and actual length in the present invention;

[0053] Wherein, (a) is a schematic diagram under the condition of k-2≥2; (b) is a schematic diagram under the condition of k-2<2;

[0054] Figure 5 It is a schematic diagram of a rectangular pitting area defined in the present invention;

[0055] Figure 6 This is a flow chart of a system for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment in the present invention;

[0056] Among them, 1-pitting identification rectangular marking box; 2-ruler identification rectangular marking box; 3-scale identification rectangular marking box; 4-pitting area; DETAILED DESCRIPTION

[0057] The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0058] The terms used in this disclosure are for the purpose of describing characteristic embodiments only and are not intended to limit the disclosure. The singular forms "a", "the" and "the" used in this disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0059] The specific implementation of the present invention is further described below in conjunction with specific embodiments:

[0060] A method for identifying and quantitatively analyzing pitting corrosion in petrochemical equipment, such as Figure 1 As shown, the following steps are included:

[0061] S1. Collect pitting image data of petrochemical equipment, build a neural network model based on deep learning, and train to obtain a pitting recognition model, such as Figure 2 As shown, the model marks pitting pits in the form of rectangular frames, which are recorded as pitting identification rectangular marking frames 1;

[0062] S1 includes the following sub-steps:

[0063] S1.1. Manually screen the collected pitting images and divide them into pitting images that are easy to label and pitting images that are difficult to label. Images containing 30 or less pits are considered to be easy to label images, and the remaining images are considered to be difficult to label images. The easy to label pitting images are manually labeled to obtain first image data.

[0064] S1.2, training the neural network model with the first image data to obtain a label generation model;

[0065] S1.3, using the model to identify the part of the pitting image that is difficult to annotate, and using the identification result as a pseudo-annotation to obtain the second image data;

[0066] S1.4. Retrain the neural network model using the first image data and the second image data to obtain a new annotation generation model;

[0067] S1.5, repeat S1.3 and S1.4 until all the pitting images that are difficult to mark are identified and the final second image data is obtained; use the first image data and the final second image data to retrain the neural network model to obtain a pitting recognition model, which is used to output the pixel coordinates of the pitting pits in the image.

[0068] Specifically, the neural network model is a target detection model or an instance segmentation model.

[0069] Specifically, no less than 1,000 pitting image data are collected, with a resolution of no less than 200 dpi. A ruler is set around the pitting pits on the image as a size reference, and data augmentation operations are performed during training.

[0070] S2, establishing a corresponding relationship between the pixel coordinates of the pit and the actual coordinates, and converting the pixel size information of the pit into the actual size information;

[0071] S2 includes the following sub-steps:

[0072] S2.1. Collect pitting image data with a ruler, mark the ruler and the ruler scale, build a target detection model, and iterate the model multiple times with the collected image data to obtain a ruler recognition model; Figure 3 As shown, the model marks the ruler and the scale on the ruler in the form of a rectangular frame, which is denoted as a ruler identification rectangular marking frame 2 and a scale identification rectangular marking frame 3;

[0073] S2.2. Establish a pixel coordinate system with the upper left corner of the pitting image as the origin, use the ruler recognition model to identify the ruler and ruler scale in the pitting image, and output the pixel coordinates of all ruler scale prediction boxes, denoted as {(x 11 ,y 11 , x 12 ,y 22 ), (x 21 ,y 21 , x 22 ,y 22 ),...,(x k1 ,y k1 , x k2 ,y k2 )}, where (x k1 ,y k1 ) is the coordinate of the upper left corner of the k-th prediction box in the image, (x k2 ,y k2 ) is the coordinate of the lower right corner of the kth prediction box; calculate the area S of each prediction box k , and remove the scales corresponding to the two smallest areas, S k The calculation process is expressed as:

[0074] Sk =(x k2 -x k1 )×(y k2 -y k1 );

[0075] S2.3, such as Figure 4 As shown in (a), if k-2 ≥ 2, calculate the center points of the remaining prediction boxes and sort the prediction boxes according to the horizontal coordinates of the center points. The actual distance between the center points of two adjacent prediction boxes is recorded as L′, the pixel coordinates of the two center points are (X′1, Y′1) and (X′2, Y′2), and the pixel distance is recorded as d′. The correspondence between pixels and actual coordinates is calculated by using the coordinates of the two center points and trigonometric functions. The pixel length l′ corresponding to one unit actual length is expressed as:

[0076]

[0077] If k-2<2, execute step S2.4;

[0078] S2.4, such as Figure 4 As shown in (b), two points are manually selected on the image. The actual distance between the two points is L″, the pixel coordinates of the two center points are (X″1, Y″1) and (X″2, Y″2), and the pixel distance is recorded as d″. The correspondence between the pixel coordinates and the actual coordinates is calculated by using the two-point pixel coordinates and trigonometric functions. The pixel length l″ corresponding to one unit actual length is expressed as:

[0079]

[0080] S3, using the pitting corrosion recognition model obtained in S1 to recognize the pitting corrosion image to be recognized, and outputting the pixel coordinates of the recognized pitting corrosion, converting the pixel coordinates to the real coordinates, and calculating the density, opening and depth of the pitting corrosion pit;

[0081] S3 includes the following sub-steps:

[0082] S3.1, use the pitting recognition model established in S1 to identify the pitting image to be identified, output the pixel coordinates of the identified pitting prediction box, and convert them into real coordinates, which are recorded as:

[0083] S3.2. Approximate the pitting pit to a circle and calculate the radius of each pitting pit, which is recorded as {r1, r2, ..., r n}, the average radius is recorded as r, and the average opening s of the pit is calculated using the following formula:

[0084]

[0085]

[0086] s=πr 2 :

[0087] Among them, r n is the approximate radius of the nth pit;

[0088] r is the average radius of all pits measured;

[0089] s is the average opening of all pits measured;

[0090] S3.3, according to the coordinates of the pits, a rectangular pitting area 5 is delineated in the image with the leftmost, rightmost, topmost and bottommost pits as the reference, such as Figure 5 After converting the pixel length to the actual length, the area m of the pitting area 5 is calculated, and the density of the pitting pit is calculated. The calculation formula is:

[0091]

[0092] Where, ρ is the density of pits;

[0093] n is the number of pits;

[0094] S3.4. Measure the actual depth of the pit manually and record it as h.

[0095] S4. Evaluate the severity of pitting corrosion;

[0096] A series of different thresholds are set for ρ, s and h to comprehensively reflect the severity of corrosion, as shown in the following table:

[0097]

[0098] A petrochemical equipment pitting corrosion identification and quantitative analysis system, using the petrochemical equipment pitting corrosion identification and quantitative analysis method as described above, including an image acquisition module, a pitting corrosion identification module, a statistical analysis module and a quantitative evaluation module;

[0099] The image acquisition module is used to collect image data from the equipment to be inspected and transmit it to the pitting recognition module;

[0100] The pitting recognition module recognizes the received image data, marks each pitting with a rectangular frame, transmits the marked image to the quantitative evaluation module, and transmits the coordinates of the upper left corner and the lower right corner of the rectangular frame to the statistical analysis module;

[0101] The statistical analysis module is used to complete the conversion of pixel coordinates to real coordinates, complete the calculation of pit density and opening, and output the above values ​​and measured depth to the quantitative evaluation module;

[0102] The quantitative evaluation module records the received data and makes a comprehensive analysis of the severity of pitting corrosion based on the different data values.

[0103] Example 1

[0104] A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment comprises the following steps:

[0105] S1. Collect pitting image data of petrochemical equipment, train Yolov5 target detection model, perform data enhancement operation during training, and obtain pitting recognition model after training;

[0106] S1 includes the following sub-steps:

[0107] S1.1. Manually screening the collected pitting images to divide them into pitting images that are easy to annotate and pitting images that are difficult to annotate, and manually annotating the pitting images that are easy to annotate to obtain first image data;

[0108] S1.2, training the Yolov5 target detection model with the first image data to obtain a label generation model;

[0109] S1.3, using the model to identify the part of the pitting image that is difficult to annotate, and using the identification result as a pseudo-annotation to obtain the second image data;

[0110] S1.4, retraining the Yolov5 object detection model using the first image data and the second image data to obtain a new annotation generation model;

[0111] S1.5. Repeat S1.3 and S1.4 until all the pitting images that are difficult to mark are identified and the final second image data is obtained; use the first image data and the final second image data to retrain the Yolov5 target detection model to obtain a pitting recognition model, which is used to output the pixel coordinates of the pitting pits in the image.

[0112] Specifically, the collected pitting image data is no less than 1,000 pieces, the resolution is no less than 200 dpi, and a ruler is provided around the pitting pits on the image as a size reference.

[0113] S2, establishing a corresponding relationship between the pixel coordinates of the pit and the actual coordinates, and converting the pixel size information of the pit into the actual size information;

[0114] S2 includes the following sub-steps:

[0115] S2.1. Collect pitting image data with a ruler, mark the ruler and the ruler scale, and use the collected image data to perform multiple iterations of training on the Yolov5 target detection model to obtain a ruler recognition model;

[0116] S2.2. Establish a pixel coordinate system with the upper left corner of the pitting image as the origin, use the ruler recognition model to identify the ruler and ruler scale in the pitting image, and output the coordinates of all ruler scale prediction boxes, denoted as {(x 11 ,y 11 , x 12 ,y 22 ), (x 21 ,y 21 , x 22 ,y 22 ),...,(x k1 ,y k1 , x k2 ,y k2 )}, where (x k1 ,y k1 ) is the coordinate of the upper left corner of the k-th prediction box in the image, (x k2 ,y k2 ) is the coordinate of the lower right corner of the kth prediction box; calculate the area S of each prediction box k , and remove the scales corresponding to the two smallest areas, S k The calculation process is expressed as:

[0117] S k =(x k2 -x k1 )×(y k2 -y k1 );

[0118] S2.3, if k-2 ≥ 2, calculate the center points of the remaining prediction boxes, and sort the prediction boxes according to the horizontal coordinates of the center points. The actual distance between the center points of two adjacent prediction boxes is recorded as L'. The corresponding relationship between the pixels and the actual coordinates is calculated by the coordinates of the two center points (X'1, Y'1), (X'2, Y'2) and trigonometric functions. The pixel length l' corresponding to one unit actual length is expressed as:

[0119]

[0120] If k-2<2, execute step S2.4;

[0121] S2.4. Artificially select two points on the image, the actual distance between the two points is L″, and the correspondence between the pixels and the actual coordinates is calculated by the pixel coordinates of the two points (X″1, Y″1), (X″2, Y″2) and trigonometric functions. The pixel length l″ corresponding to one unit of actual length is expressed as:

[0122]

[0123] S3, using the pitting corrosion recognition model obtained in S1 to recognize the pitting corrosion image to be recognized, and outputting the pixel coordinates of the recognized pitting corrosion, converting the pixel coordinates with the real coordinates, and calculating the density, opening and depth of the pitting corrosion pit;

[0124] S3 includes the following sub-steps:

[0125] S3.1, use the pitting recognition model established in S1 to identify the pitting image to be identified, output the pixel coordinates of the identified pitting prediction box, and convert them into real coordinates, which are recorded as:

[0126] S3.2. Approximate the pitting pit to a circle and calculate the radius of each pitting pit, which is recorded as {r1, r2, ..., r n}, the average radius is recorded as r, and the average opening s of the pit is calculated using the following formula:

[0127]

[0128]

[0129] s=πr 2 ;

[0130] Among them, r n is the approximate radius of the nth pit;

[0131] r is the average radius of all pits measured;

[0132] s is the average opening of all pits measured;

[0133] S3.3. According to the coordinates of the pitting pit, a rectangular pitting area is delineated in the image, and after converting the pixel length to the actual length, the area m of the pitting area is calculated, and the density of the pitting pit is calculated. The calculation formula is:

[0134]

[0135] Where, ρ is the density of pits;

[0136] n is the number of pits;

[0137] S3.4. Measure the actual depth of the pit manually and record it as h.

[0138] S4. Evaluate the severity of pitting corrosion;

[0139] A series of different thresholds are set for ρ, s and h to comprehensively reflect the severity of corrosion, as shown in the following table:

[0140] Risk Level <![CDATA[ρ / m -2 ]]> <![CDATA[s / mm 2 ]]> h / mm 1 <![CDATA[2×10 3 ]]> 0.5 0.4 2 <![CDATA[1×10 4 ]]> 2.0 0.8 3 <![CDATA[5×10 4 ]]> 8.0 1.6 4 <![CDATA[1×10 5 ]]> 12.5 3.2 5 <![CDATA[5×10 6 ]]> 24.5 6.4

[0141] A petrochemical equipment pitting corrosion identification and quantitative analysis system, using the petrochemical equipment pitting corrosion identification and quantitative analysis method as described above, including an image acquisition module, a pitting corrosion identification module, a statistical analysis module and a quantitative evaluation module;

[0142] The image acquisition module is used to collect image data from the equipment to be inspected and transmit it to the pitting recognition module;

[0143] The pitting recognition module recognizes the received image data, marks each pitting with a rectangular frame, transmits the marked image to the quantitative evaluation module, and transmits the coordinates of the upper left corner and the lower right corner of the rectangular frame to the statistical analysis module;

[0144] The statistical analysis module is used to complete the conversion of pixel coordinates to real coordinates, complete the calculation of pit density and opening, and output the above values ​​and measured depth to the quantitative evaluation module;

[0145] The quantitative evaluation module records the received data and makes a comprehensive analysis of the severity of pitting corrosion based on the different data values.

[0146] Example 2

[0147] A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment comprises the following steps:

[0148] S1. Collect petrochemical equipment pitting image data, train the Mask R-CNN instance segmentation model, perform data enhancement operations during training, and obtain a pitting recognition model after training is completed;

[0149] S1 includes the following sub-steps:

[0150] S1.1. Manually screening the collected pitting images to divide them into pitting images that are easy to annotate and pitting images that are difficult to annotate, and manually annotating the pitting images that are easy to annotate to obtain first image data;

[0151] S1.2, training the Mask R-CNN instance segmentation model using the first image data to obtain a label generation model;

[0152] S1.3, using the model to identify the part of the pitting image that is difficult to annotate, and using the identification result as a pseudo-annotation to obtain the second image data;

[0153] S1.4. Retrain the Mask R-CNN instance segmentation model using the first image data and the second image data to obtain a new annotation generation model;

[0154] S1.5. Repeat S1.3 and S1.4 until all the pitting images that are difficult to label are identified and the final second image data is obtained; the Mask R-CNN instance segmentation model is retrained using the first image data and the final second image data to obtain a pitting recognition model, which is used to output the pixel coordinates of the pitting pits in the image.

[0155] Specifically, no less than 1,000 pitting image data are collected, with a resolution of no less than 200 dpi. A ruler is set around the pitting pits on the image as a size reference, and data augmentation operations are performed during training.

[0156] S2, establishing a corresponding relationship between the pixel coordinates of the pit and the actual coordinates, and converting the pixel size information of the pit into the actual size information;

[0157] S2 includes the following sub-steps:

[0158] S2.1. Collect pitting image data with a ruler, mark the ruler and the ruler scale, build an instance segmentation model, and iterate the model multiple times with the collected image data to obtain a ruler recognition model;

[0159] S2.2. Establish a pixel coordinate system with the upper left corner of the pitting image as the origin, use the ruler recognition model to identify the ruler and ruler scale in the pitting image, and output the coordinates of all ruler scale prediction boxes, denoted as {(x 11 ,y 11 , x 12 ,y 22 ), (x 21 ,y 21 , x 22 ,y 22 ),...,(x k1 ,y k1 , x k2 ,y k2 )}, where (x k1 ,y k1 ) is the coordinate of the upper left corner of the k-th prediction box in the image, (x k2 ,y k2 ) is the coordinate of the lower right corner of the kth prediction box; calculate the area S of each prediction box k , and remove the scales corresponding to the two smallest areas, S k The calculation process is expressed as:

[0160] S k =(x k2 -x k1 )×(y k2 -y k1 );

[0161] S2.3, if k-2 ≥ 2, calculate the center points of the remaining prediction boxes, and sort the prediction boxes according to the horizontal coordinates of the center points. The actual distance between the center points of two adjacent prediction boxes is recorded as L'. The corresponding relationship between the pixels and the actual coordinates is calculated by the coordinates of the two center points (X"1, Y"1), (X"2, Y"2) and trigonometric functions. The pixel length l' corresponding to one unit actual length is expressed as:

[0162]

[0163] If k-2<2, execute step S2.4;

[0164] S2.4. Artificially select two points on the image, the actual distance between the two points is L″, and the corresponding relationship between the pixels and the actual coordinates is calculated by the pixel coordinates (X″1, Y″1), (X″2, Y″2) of the two points and trigonometric functions. The pixel length l″ corresponding to one unit of actual length is expressed as:

[0165]

[0166] S3, using the pitting corrosion recognition model obtained in S1 to recognize the pitting corrosion image to be recognized, and outputting the pixel coordinates of the recognized pitting corrosion, converting the pixel coordinates to the real coordinates, and calculating the density, opening and depth of the pitting corrosion pit;

[0167] S3 includes the following sub-steps:

[0168] S3.1, use the pitting recognition model established in S1 to identify the pitting image to be identified, output the pixel coordinates of the identified pitting prediction box, and convert them into real coordinates, which are recorded as:

[0169] S3.2. Approximate the pitting pit to a circle and calculate the radius of each pitting pit, which is recorded as {r1, r2, ..., r n}, the average radius is recorded as r, and the average opening s of the pit is calculated using the following formula:

[0170]

[0171]

[0172] s=πr 2 :

[0173] Among them, r n is the approximate radius of the nth pit;

[0174] r is the average radius of all pits measured;

[0175] s is the average opening of all pits measured;

[0176] S3.3. According to the coordinates of the pitting pit, a rectangular pitting area is delineated in the image, and after converting the pixel length to the actual length, the area m of the pitting area is calculated, and the density of the pitting pit is calculated. The calculation formula is:

[0177]

[0178] Where, ρ is the density of pits;

[0179] n is the number of pits;

[0180] S3.4. Measure the actual depth of the pit manually and record it as h.

[0181] S4. Evaluate the severity of pitting corrosion;

[0182] A series of different thresholds are set for ρ, s and h to comprehensively reflect the severity of corrosion, as shown in the following table:

[0183] Risk Level <![CDATA[ρ / m -2 ]]> <![CDATA[s / mm 2 ]]> h / mm 1 <![CDATA[2×10 3 ]]> 0.5 0.4 2 <![CDATA[1×10 4 ]]> 2.0 0.8 3 <![CDATA[5×10 4 ]]> 8.0 1.6 4 <![CDATA[1×10 5 ]]> 12.5 3.2 5 <![CDATA[5×10 6 ]]> 24.5 6.4

[0184] A petrochemical equipment pitting corrosion identification and quantitative analysis system, using the petrochemical equipment pitting corrosion identification and quantitative analysis method as described above, including an image acquisition module, a pitting corrosion identification module, a statistical analysis module and a quantitative evaluation module;

[0185] The image acquisition module is used to collect image data from the equipment to be inspected and transmit it to the pitting recognition module;

[0186] The pitting recognition module recognizes the received image data, marks each pitting with a rectangular frame, transmits the marked image to the quantitative evaluation module, and transmits the coordinates of the upper left corner and the lower right corner of the rectangular frame to the statistical analysis module;

[0187] The statistical analysis module is used to complete the conversion of pixel coordinates to real coordinates, complete the calculation of pit density and opening, and output the above values ​​and measured depth to the quantitative evaluation module;

[0188] The quantitative evaluation module records the received data and makes a comprehensive analysis of the severity of pitting corrosion based on the different data values.

[0189] Example 3

[0190] A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment comprises the following steps:

[0191] S1. Collect pitting image data of petrochemical equipment, train the EfficientDet target detection model, perform data enhancement operations during training, and obtain the pitting recognition model after training is completed;

[0192] S1 includes the following sub-steps:

[0193] S1.1. Manually screening the collected pitting images to divide them into pitting images that are easy to annotate and pitting images that are difficult to annotate, and manually annotating the pitting images that are easy to annotate to obtain first image data;

[0194] S1.2, training the EfficientDet object detection model with the first image data to obtain a label generation model;

[0195] S1.3, using the model to identify the part of the pitting image that is difficult to annotate, and using the identification result as a pseudo-annotation to obtain the second image data;

[0196] S1.4. Retrain the EfficientDet object detection model using the first image data and the second image data to obtain a new annotation generation model;

[0197] S1.5. Repeat S1.3 and S1.4 until all the pitting images that are difficult to label are identified and the final second image data is obtained; use the first image data and the final second image data to retrain the EfficientDet target detection model to obtain a pitting recognition model, which is used to output the pixel coordinates of the pitting pits in the image.

[0198] Specifically, no less than 1,000 pitting image data are collected, with a resolution of no less than 200 dpi. A ruler is set around the pitting pits on the image as a size reference, and data augmentation operations are performed during training.

[0199] S2, establishing a corresponding relationship between the pixel coordinates of the pit and the actual coordinates, and converting the pixel size information of the pit into the actual size information;

[0200] S2 includes the following sub-steps:

[0201] S2.1. Collect pitting image data with rulers, mark the rulers and ruler scales, build the EfficientDet target detection model, and iterate the model multiple times with the collected image data to obtain the ruler recognition model;

[0202] S2.2. Establish a pixel coordinate system with the upper left corner of the pitting image as the origin, use the ruler recognition model to identify the ruler and ruler scale in the pitting image, and output the coordinates of all ruler scale prediction boxes, denoted as {(x 11 ,y 11 , x 12 ,y 22 ), (x 21 ,y 21 , x 22,y 22 ),...,(x k1 ,y k1 , x k2 ,y k2 )}, where (x k1 ,y k1 ) is the coordinate of the upper left corner of the k-th prediction box in the image, (x k2 ,y k2 ) is the coordinate of the lower right corner of the kth prediction box; calculate the area S of each prediction box k , and remove the scales corresponding to the two smallest areas, S k The calculation process is expressed as:

[0203] S k =(x k2 -x k1 )×(y k2 -y k1 );

[0204] S2.3, if k-2 ≥ 2, calculate the center points of the remaining prediction boxes, and sort the prediction boxes according to the horizontal coordinates of the center points. The actual distance between the center points of two adjacent prediction boxes is recorded as L'. The corresponding relationship between the pixels and the actual coordinates is calculated by the coordinates of the two center points (X'1, Y'1), (X'2, Y'2) and trigonometric functions. The pixel length l' corresponding to one unit actual length is expressed as:

[0205]

[0206] If k-2<2, execute step S2.4;

[0207] S2.4. Artificially select two points on the image, the actual distance between the two points is L″, and the corresponding relationship between the pixels and the actual coordinates is calculated by the pixel coordinates (X″1, Y″1), (X″2, Y″2) of the two points and trigonometric functions. The pixel length l″ corresponding to one unit of actual length is expressed as:

[0208]

[0209] S3, using the pitting corrosion recognition model obtained in S1 to recognize the pitting corrosion image to be recognized, and outputting the pixel coordinates of the recognized pitting corrosion, converting the pixel coordinates with the real coordinates, and calculating the density, opening and depth of the pitting corrosion pit;

[0210] S3 includes the following sub-steps:

[0211] S3.1, use the pitting recognition model established in S1 to identify the pitting image to be identified, output the pixel coordinates of the identified pitting prediction box, and convert them into real coordinates, which are recorded as:

[0212] S3.2. Approximate the pitting pit to a circle and calculate the radius of each pitting pit, which is recorded as {r1, r2, ..., r n}, the average radius is recorded as r, and the average opening s of the pit is calculated using the following formula:

[0213]

[0214]

[0215] s=πr 2 ;

[0216] Among them, r n is the approximate radius of the nth pit;

[0217] r is the average radius of all pits measured;

[0218] s is the average opening of all pits measured;

[0219] S3.3. According to the coordinates of the pitting pit, a rectangular pitting area is delineated in the image, and after converting the pixel length to the actual length, the area m of the pitting area is calculated, and the density of the pitting pit is calculated. The calculation formula is:

[0220]

[0221] Where, ρ is the density of pits;

[0222] n is the number of pits;

[0223] S3.4. Measure the actual depth of the pit manually and record it as h.

[0224] S4. Evaluate the severity of pitting corrosion;

[0225] A series of different thresholds are set for ρ, s and h to comprehensively reflect the severity of corrosion, as shown in the following table:

[0226] Risk Level <![CDATA[ρ / m -2 ]]> <![CDATA[s / mm 2 ]]> h / mm 1 <![CDATA[2×10 3 ]]> 0.5 0.4 2 <![CDATA[1×10 4 ]]> 2.0 0.8 3 <![CDATA[5×10 4 ]]> 8.0 1.6 4 <![CDATA[1×10 5 ]]> 12.5 3.2 5 <![CDATA[5×10 6 ]]> 24.5 6.4

[0227] A petrochemical equipment pitting corrosion identification and quantitative analysis system, using the petrochemical equipment pitting corrosion identification and quantitative analysis method as described above, including an image acquisition module, a pitting corrosion identification module, a statistical analysis module and a quantitative evaluation module;

[0228] The image acquisition module is used to collect image data from the equipment to be inspected and transmit it to the pitting recognition module;

[0229] The pitting recognition module recognizes the received image data, marks each pitting with a rectangular frame, transmits the marked image to the quantitative evaluation module, and transmits the coordinates of the upper left corner and the lower right corner of the rectangular frame to the statistical analysis module;

[0230] The statistical analysis module is used to complete the conversion of pixel coordinates to real coordinates, complete the calculation of pit density and opening, and output the above values ​​and measured depth to the quantitative evaluation module;

[0231] The quantitative evaluation module records the received data and makes a comprehensive analysis of the severity of pitting corrosion based on the different data values.

[0232] The above describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the above-mentioned specific embodiments, and the devices and structures that are not described in detail should be understood to be implemented in a common manner in the art; any technician familiar with the art can use the above-disclosed methods and technical contents to make many possible changes and modifications to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, or modify them into equivalent embodiments of equivalent changes, which does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention are still within the scope of protection of the technical solutions of the present invention.

Claims

1. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment, characterized in that: The steps include: S1. Collect pitting image data of petrochemical equipment, build a neural network model based on deep learning, and train to obtain a pitting recognition model; S2, establishing a corresponding relationship between the pixel coordinates of the pit and the actual coordinates, and converting the pixel size information of the pit into the actual size information; S3, using the pitting corrosion recognition model obtained in S1 to recognize the pitting corrosion image to be recognized, and outputting the pixel coordinates of the recognized pitting corrosion, converting the pixel coordinates to the real coordinates, and calculating the density, opening and depth of the pitting corrosion pit; S4. Evaluate the severity of pitting corrosion.

2. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 1, characterized in that: The S1 comprises the following sub-steps: S1.

1. Manually screen the collected pitting images and divide them into pitting images that are easy to label and pitting images that are difficult to label. Images containing 30 or less pits are considered to be easy to label images, and the remaining images are considered to be difficult to label images. The easy to label pitting images are manually labeled to obtain first image data. S1.2, training the neural network model with the first image data to obtain a label generation model; S1.3, using the model to identify the part of the pitting image that is difficult to annotate, and using the identification result as a pseudo-annotation to obtain the second image data; S1.

4. Retrain the neural network model using the first image data and the second image data to obtain a new annotation generation model; S1.5, repeat S1.3 and S1.4 until all the pitting images that are difficult to mark are identified and the final second image data is obtained; use the first image data and the final second image data to retrain the neural network model to obtain a pitting recognition model, which is used to output the pixel coordinates of the pitting pits in the image.

3. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 1, characterized in that: The S1 includes the following sub-steps: the neural network model is a target detection model or an instance segmentation model.

4. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 2, characterized in that: The collected pitting image data shall be no less than 1,000 pieces, with a resolution of no less than 200 dpi. A ruler shall be provided around the pitting pits on the image as a size reference.

5. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 1, characterized in that: The S2 comprises the following sub-steps: S2.

1. Collect pitting image data with a ruler, mark the ruler and the ruler scale, build a target detection model, and iterate the model multiple times with the collected image data to obtain a ruler recognition model; S2.

2. Establish a pixel coordinate system with the upper left corner of the pitting image as the origin, use the ruler recognition model to identify the ruler and ruler scale in the pitting image, and output the pixel coordinates of all ruler scale prediction boxes, denoted as {(x 11 ,y 11 ,x 12 ,y 22 ),(x 21 ,y 21 ,x 22 ,y 22 ),...,(x k1 ,y k1 ,x k2 ,y k2 )}, where (x k1 ,y k1 ) is the coordinate of the upper left corner of the k-th prediction box in the image, (x k2 ,y k2 ) is the coordinate of the lower right corner of the kth prediction box; calculate the area S of each prediction box k , and remove the scales corresponding to the two smallest areas, S k The calculation process is expressed as: S k (x) k2 -x k1 )×(and k2 -and k1 ); S2.3, if k-2 ≥ 2, calculate the center points of the remaining prediction boxes, and sort the prediction boxes according to the horizontal coordinates of the center points. The actual distance between the center points of two adjacent prediction boxes is recorded as L'. The corresponding relationship between the pixels and the actual coordinates is calculated by the coordinates of the two center points (X'1, Y'1), (X'2, Y'2) and trigonometric functions. The pixel length l' corresponding to one unit actual length is expressed as: If k-2<2, execute step S2.4; S2.

4. Artificially select two points on the image, the actual distance between the two points is L″, and the correspondence between the pixels and the actual coordinates is calculated by the pixel coordinates of the two points (X″1, Y″1), (X″2, Y″2) and trigonometric functions. The pixel length l″ corresponding to one unit of actual length is expressed as:

6. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 1, characterized in that: The S3 comprises the following sub-steps: S3.1, use the pitting recognition model established in S1 to identify the pitting image to be identified, output the pixel coordinates of the identified pitting prediction box, and convert them into real coordinates, which are recorded as: S3.

2. Approximate the pitting pit to a circle and calculate the radius of each pitting pit, which is recorded as {r1, r2, ..., r n }, the average value of the radius is recorded as r, and the average opening s of the pit is calculated; S3.3, according to the coordinates of the pits, a rectangular pitting area is delineated in the image with the leftmost, rightmost, topmost and bottommost pits as the reference, and after converting the pixel length to the actual length, the area m of the pitting area is calculated, and the density of the pits is calculated; S3.

4. Measure the actual depth of the pit manually and record it as h.

7. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 6, characterized in that: In S3.2, the calculation formula is: s＝πr 2 ； Among them, r n is the approximate radius of the nth pit; r is the average radius of all pits measured; s is the average opening of all pits measured.

8. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 6, characterized in that: In S3.3, the calculation formula is: Where ρ is the density of pits; n is the number of pits.

9. A method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment according to claim 8, characterized in that: A series of different thresholds are set for ρ, s and h to comprehensively reflect the severity of corrosion.

10. A system for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment, using a method for identifying and quantitatively analyzing pitting corrosion of petrochemical equipment as claimed in any one of claims 1 to 9, characterized in that: It includes image acquisition module, pitting recognition module, statistical analysis module and quantitative evaluation module; The image acquisition module is used to collect image data from the equipment to be inspected and transmit it to the pitting recognition module; The pitting recognition module recognizes the received image data, marks each pitting with a rectangular frame, transmits the marked image to the quantitative evaluation module, and transmits the upper left corner coordinates and the lower right corner coordinates of the rectangular frame to the statistical analysis module; The statistical analysis module is used to complete the conversion of pixel coordinates to real coordinates, complete the calculation of pit density and opening, and output the above values ​​and measured depth to the quantitative evaluation module; The quantitative evaluation module records the received data and makes a comprehensive analysis of the severity of pitting corrosion according to different data values.

Images