Intelligent ferrographic analysis method

Abstract

The invention relates to the field of mechanical equipment abrasion and oil abrasive particle detection, in particular to an intelligent ferrographic analysis method which comprises the following steps: shooting a plurality of original abrasive particle images at different horizontal positions and different heights by adjusting the horizontal position and height of an imaging system; performing depth-of-field fusion and mapsheet splicing on the original abrasive particle image to obtain an abrasive particle target map, extracting the contour of each abrasive particle, performing graphic calculation analysis on an abrasive particle instance map, inputting each abrasive particle instance map into a convolutional neural network, and revising the classification state of each abrasive particle; and comprehensively judging the equipment wear state based on the quantitative index and qualitative index component wear state evaluation model. According to the method, all abrasive particles in the deposition area are clearly imaged in one image through full-field-depth and full-area abrasive particle photographing assisted by a field depth fusion and mapsheet splicing algorithm, and a good abrasive particle image basis is provided for qualitative observation and quantitative analysis. The whole method is simple and quick to operate, and the influence of manual operation on detection is avoided.

Description

Technical Field

[0001] The present invention relates to the field of mechanical equipment wear and oil particle detection, and specifically to an intelligent ferrography analysis method. Background Technique

[0002] The detection and monitoring of oil are common means to evaluate the conditions of various machines and predict various faults. The detection of abrasive particles in oil has become an essential method in modern industrial maintenance activities, and it is a conventional detection method to ensure the long-term effective and stable operation of various mechanical equipment, which can reduce the accident rate and improve work efficiency.

[0003] The existing ferrography analysis method observes, takes pictures of, and analyzes and calculates the abrasive particles on the ferrogram by using a ferrographic microscope. However, restricted by the optical imaging principle and the actual situation of optical products engineering, on the one hand, the microscopic imaging field of view is limited, and only local observation and imaging can be carried out in separate regions, unable to observe and analyze the overall appearance of all deposited abrasive particles, and it is also easy to miss key and important abrasive particles. On the other hand, the ferrographic microscope cannot balance between large depth of field and high magnification, and mainly realizes the imaging of large and small abrasive particles by switching lenses with different magnifications, and it is very difficult to clearly image abrasive particles with sizes ranging from a few micrometers to hundreds of micrometers at the same time. When analyzing and judging the wear state, the traditional analytical ferrograph can only roughly give the percentage of the coverage area of large abrasive particles and the percentage of the coverage area of small abrasive particles, but these are sampling values for different regional measuring points on the ferrogram, and it is difficult to meet the requirements of quantitative analysis and judgment. With the development of technology, currently, some analytical ferrographs can give individual abrasive particle size data and the abrasive particle size distribution data of some regions, but the image basis is still the selected and sporadic abrasive particle range by humans, and it is difficult to guarantee the data consistency, repeatability, standardization, etc. There are obvious deficiencies when used as the basis for quantitative analysis and judgment.

[0004] In summary, the traditional ferrography analysis method can only realize the observation, imaging, and analysis of local and specific abrasive particles, and it is difficult to grasp the overall picture of the abrasive particles in the oil sample. The problems brought about by this are mainly manifested as: it is difficult to carry out quantitative analysis and evaluation, the operation is cumbersome, the time consumption is long, and it is greatly affected by the operation of personnel. Summary of the Invention

[0005] To overcome the problems existing in the prior art, the purpose of the present invention is to provide an intelligent ferrography analysis method that can achieve clear imaging of all large and small abrasive particles in the deposition area, retains local details while providing a global perspective, provides a good abrasive particle image basis for qualitative observation and quantitative analysis, greatly enriches the content of quantitative indicators, improves the standardization and convenience of data processing, provides a new solution and a relatively perfect system framework for the standardization, accuracy, and perfection of ferrography data analysis, and at the same time has simple and fast operation, realizes the automation and intelligence of ferrography analysis, and avoids the influence of human operation on detection.

[0006] To achieve the above object, the present invention provides the following technical solutions: An intelligent ferrography analysis method, comprising the following steps:

[0007] S1: By adjusting the horizontal position and height of the imaging system, starting from one edge of the abrasive particle field of view range to the other edge, capture a number of original abrasive particle images at different horizontal positions and different heights;

[0008] S2: Perform preprocessing on the original abrasive particle images, perform depth of field fusion and image stitching on the original abrasive particle images to obtain an abrasive particle target image;

[0009] S3: Feed the abrasive particle target image into the Mask R-CNN network, extract the abrasive particle regions through the segmentation network to obtain accurate abrasive particle instance images, and obtain the outer contour of each abrasive particle;

[0010] S4: Perform graphic calculation and analysis on the abrasive particle instance images to obtain quantitative indicators, where the quantitative indicators include the calculation indicators of each abrasive particle;

[0011] S5: Input each abrasive particle instance image into a convolutional neural network for initial classification of characteristic abrasive particles, and revise the classification status of each abrasive particle in combination with the calculation indicators of the abrasive particles obtained in step S4 to determine the category information of the characteristic abrasive particles;

[0012] S6: Based on steps S4 - S5, determine comprehensive analysis indicators, including quantitative and qualitative indicators of single abrasive particles and overall abrasive particles;

[0013] S7: Construct a wear state evaluation model of components to comprehensively evaluate the wear state of the equipment.

[0014] The present invention is further configured as follows: Step S1 specifically includes the following steps:

[0015] S11: The imaging system takes the first abrasive particle image with the edge of the abrasive particle field of view range as the horizontal starting position and the glass surface in contact with the abrasive particle field of view range as the vertical starting position;

[0016] S12: Keep the horizontal position of the imaging system unchanged, raise the vertical height, and the height increase is controlled within the imaging depth of field range. Take the next abrasive particle image, and repeat the aforementioned height increase step after completion until the total elevation increase covers the maximum detection range of the abrasive particles, thereby obtaining multiple abrasive particle images at different elevations at the same horizontal position;

[0017] S13: After taking abrasive grain images at different heights at a certain planar position, the imaging system lowers its height to the vertical starting position and horizontally translates from the horizontal starting position towards the other edge of the abrasive grain field of view. After moving into position, it takes the first image at this horizontal position, then keeps the horizontal position unchanged, and repeats step S12 to achieve layered image capture of abrasive grains in this area;

[0018] S14: Repeat step S13 until the abrasive grain area captured by the imaging system reaches the other end edge of the abrasive grain field of view, completing the original acquisition of all abrasive grain images.

[0019] The present invention is further configured such that: in step S13, the horizontal translation distance does not exceed the length of the single imaging field of view in the horizontal direction, and there is a certain overlapping part between two imaging.

[0020] The present invention is further configured such that: step S2 specifically is, for the layered images at all heights obtained in the same horizontal position area, apply the image depth of field fusion processing algorithm to extract the clear parts in each image and synthesize them into a depth of field target image; then starting from the depth of field target image corresponding to the horizontal starting position, apply the image stitching algorithm to stitch all the depth of field target images one by one in the order of shooting until all the depth of field target images are stitched into a final abrasive grain target image with the full field of view.

[0021] The present invention is further configured such that: step S3 specifically is, the Mask R-CNN network analyzes and identifies the abrasive grain target area and the background area based on the artificial intelligence semantic segmentation model, extracts the abrasive grain target area through the segmentation network to obtain an accurate abrasive grain instance map, and applies the edge detection algorithm to the abrasive grain target area to extract the outer contour of each abrasive grain.

[0022] The present invention is further configured such that: the calculation indexes of each abrasive grain in step S4 include the major axis size, major / minor axis size ratio, perimeter, circularity, and projected area of the abrasive grain.

[0023] The present invention is further configured such that: in step S5, the characteristic abrasive grain categories include normal abrasive grains, cutting abrasive grains, severe sliding abrasive grains, fatigue flake abrasive grains, spherical abrasive grains, adhesive abrasive grains, etc.

[0024] The definitions of different categories of abrasive grains often also include some statistical-based empirical features. For example, the length of severe sliding abrasive grains is greater than 20 μm, and these empirical data need to be used to check and revise the results of algorithm recognition.

[0025] The present invention is further configured such that in step S6, the qualitative indicators include the shape, profile, texture, color, position and orientation of the abrasive grains, the presence or absence of abrasive grains with specific characteristic morphologies, the presence or absence of abrasive grains with specific colors, and the presence or absence of non-ferromagnetic abrasive grains; the category of the characteristic abrasive grains is judged according to the qualitative indicators, and the number of characteristic abrasive grains of each category is recorded;

[0026] The quantitative indicators further include an overall evaluation index of the abrasive grains, and the overall evaluation index of the abrasive grains includes the total content of the abrasive grains in the oil sample per unit volume, the number distribution of the abrasive grains in each size segment, the percentage of the covered area of the abrasive grains, the proportion of large abrasive grains, the proportion of small abrasive grains, the wear severity index, and the average size of the abrasive grains.

[0027] It should be noted that the boundaries between large abrasive grains and small abrasive grains vary according to the different friction pairs of the monitored object. Abrasive grains with sizes exceeding the clearance of the friction pair are defined as large abrasive grains, and abrasive grains with sizes smaller than the clearance of the friction pair are defined as small abrasive grains. That is, abrasive grain diameters that have a greater impact on the friction and lubrication state are defined as large abrasive grains, and abrasive grain diameters that have a smaller impact on the friction and lubrication state are defined as small abrasive grains.

[0028] The present invention is further configured such that the percentage of the covered area of the abrasive grains is the percentage of the total area of the abrasive grains identified in the image to the area of the image, and is used to reflect the wear amount;

[0029] The number distribution of the abrasive grains in each size segment is distinguished by the major axis size, and multiple size intervals are defined. The number of abrasive grains counted according to each size interval segment is used to reflect the wear amount and the severity of wear;

[0030] The number of characteristic abrasive grains of each category corresponds to different wear states respectively and is used to judge the wear cause;

[0031] The wear severity index is used to judge the severity of wear and is calculated by the following formula,

[0032] I S =D L ×(D L -D S )

[0033] where D L is the proportion of large abrasive grains in the target image of the abrasive grains, and D S is the proportion of small abrasive grains in the target image of the abrasive grains.

[0034] In summary, the beneficial effects of the above technical solutions of the present invention are as follows:

[0035] Through panoramic depth-of-field and full-area abrasive particle photography, supplemented by depth-of-field fusion and image stitching algorithms, the present invention achieves clear imaging of all large and small abrasive particles within the deposition area in a single image, preserving both local details and providing a global perspective, thus providing a good basis for abrasive particle images for qualitative observation and quantitative analysis. The overall method is simple, fast, and fully automated, completely avoiding the influence of human operation on detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0037] Figure 1 It is a flow chart of the intelligent ferrography analysis method described in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the following will clearly and completely describe the technical solutions of the present invention in conjunction with the drawings of the present invention. Based on the embodiments of the present invention, other similar embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0039] The present invention will be further described below in conjunction with the drawings and preferred embodiments.

[0040] Embodiment:

[0041] As Figure 1 shown, it is a preferred embodiment of the present invention. An intelligent ferrography analysis method includes the following steps:

[0042] S1: By adjusting the horizontal position and height of the imaging system, starting from one end edge of the abrasive particle field of view to the other end edge, capture a number of original abrasive particle images at different horizontal positions and different heights;

[0043] S11: The imaging system takes the edge of the abrasive particle field of view as the horizontal starting position and the glass surface in contact with the abrasive particle field of view as the vertical starting position to capture the first abrasive particle image;

[0044] S12: Keep the horizontal position of the imaging system unchanged, raise the vertical height, and control the raised height within the imaging depth-of-field range. Capture the next abrasive particle image, and repeat the above-mentioned raising step after completion until the total elevation raised covers the maximum detection range of the abrasive particles, thereby obtaining multiple abrasive particle images at different elevations at the same horizontal position;

[0045] S13: After taking abrasive grain images at different heights at a certain planar position, the imaging system lowers the height to the vertical starting position and horizontally translates from the horizontal starting position towards the other edge direction of the abrasive grain field of view. After moving in place, the first image at this horizontal position is taken, and then, while keeping the horizontal position unchanged, step S12 is repeated to achieve the layered image shooting of the abrasive grains in this area; the horizontal translation distance does not exceed the length of the single imaging field of view in the horizontal direction, and there is a certain overlapping part between two imaging operations.

[0046] S14: Repeat step S13 until the abrasive grain area captured by the imaging system reaches the other end edge of the abrasive grain field of view, completing the original acquisition of all abrasive grain images.

[0047] S2: Preprocess the original abrasive grain images, perform depth of field fusion and image stitching on the original abrasive grain images to obtain the abrasive grain target image;

[0048] For the layered images at all heights in the same horizontal position area, apply the image depth of field fusion processing algorithm to extract the clear parts in each image and synthesize them into a depth of field target image; then, starting from the depth of field target image corresponding to the horizontal starting position, apply the image stitching algorithm to stitch all the depth of field target images one by one in the order of shooting until all the depth of field target images are stitched into a final abrasive grain target image with the full field of view.

[0049] The so-called depth of field fusion specifically means that for the layered images at all heights in the same horizontal position area, feature extraction is performed on each captured image respectively, and methods such as gradient and Laplace transform are used to analyze the sharpness of each area in the image. According to the sharpness characteristics, the weight of each pixel in each image is calculated, and pixels with higher sharpness will be assigned higher weights. Compare the sharpness weights of the same area in each image, select the one with the highest weight and copy it to the newly created target image with the same size as the captured image, and repeat the above operations until the copying of all areas is completed, and finally save it as the depth of field fusion target image.

[0050] The specific process of image splicing is as follows: Read in the depth-of-field fusion target image that was shot and processed first as the process image, and then read in the second depth-of-field fusion target image in chronological order as the image to be spliced. Use the feature point detection algorithm on the process image and the image to be spliced respectively to find the feature points in the images. Use the feature descriptor to compare the feature points in the process image and the image to be spliced to find the matching feature points. Based on the matching feature points in the process image and the image to be spliced, calculate the size of the area outside the overlapping part of the image to be spliced and the process image. Expand the size of the process image and paste the extra area in the image to be spliced into the expanded process image to form a new spliced process image. Continue to read in the next depth-of-field fusion target image and repeat the above splicing steps until all depth-of-field fusion images are spliced and processed. Finally, the formed process image is trimmed at the field of view edge and saved as the spliced abrasive target image.

[0051] S3: Send the abrasive target image into the Mask R-CNN network, extract the abrasive region through the segmentation network to obtain an accurate abrasive instance image, and acquire the outer contour of each abrasive.

[0052] The Mask R-CNN network analyzes and identifies the abrasive target region and the background region based on the artificial intelligence semantic segmentation model, extracts the abrasive target region through the segmentation network to obtain an accurate abrasive instance image, and applies the edge detection algorithm to the abrasive target region to extract the outer contour of each abrasive.

[0053] S4: Conduct graphic calculation and analysis on the abrasive instance image to obtain quantitative indicators, where the quantitative indicators include the calculation indicators of each abrasive.

[0054] The calculation indicators of each abrasive in step S4 include the major axis size, major / minor axis size ratio, perimeter, circularity, and projected area of the abrasive.

[0055] Among them, the major axis size is calculated according to the length of the minimum circumscribed rectangle of a single abrasive graphic; the perimeter is the total length of the outer contour of a single abrasive identified in the image; the projected area is the planar area enclosed within the contour of a single abrasive; the circularity is e = (4π × projected area) / (contour perimeter × contour perimeter), and the major / minor axis size ratio is the ratio of the major axis size to the minor axis size of a single abrasive identified in the image.

[0056] S5: Input each abrasive instance image into the convolutional neural network for preliminary classification of characteristic abrasives, and revise the classification status of each abrasive in combination with the calculation indicators of the abrasives obtained in step S4 to determine the category information of the characteristic abrasives.

[0057] The categories of the characteristic abrasives include normal abrasives, cutting abrasives, severe sliding abrasives, fatigue flake abrasives, spherical abrasives, and adhesive abrasives.

[0058] S6: Based on steps S4 - S5, determine comprehensive analysis indicators, including quantitative and qualitative indicators for individual abrasive grains and overall abrasive grains;

[0059] In step S6, the qualitative indicators include the shape, profile, texture, color, position and orientation of the abrasive grains, the presence or absence of specific characteristic morphology abrasive grains, the presence or absence of specific color abrasive grains, and the presence or absence of non - ferromagnetic abrasive grains; judge the categories of characteristic abrasive grains according to the qualitative indicators, and record the number of characteristic abrasive grains in each category;

[0060] The quantitative indicators also include the overall evaluation index of abrasive grains. The overall evaluation index of abrasive grains includes the total content of abrasive grains in the oil sample per unit volume, the number distribution of abrasive grains in each size segment, the percentage of the covered area of abrasive grains, the proportion of large abrasive grains, the proportion of small abrasive grains, the wear severity index, and the average size of abrasive grains.

[0061] The percentage of the covered area of the abrasive grains is the percentage of the total area of the identified abrasive grains in the image to the image area, which is used to reflect the wear amount;

[0062] The number distribution of abrasive grains in each size segment is distinguished by the major axis size, defining multiple size intervals, and the numerical value of the number of abrasive grains counted according to each size interval segment, which is used to reflect the wear amount and the severity of wear;

[0063] The number of characteristic abrasive grains in each category corresponds to different wear states respectively, and is used to judge the wear cause;

[0064] The wear severity index is used to judge the severity of wear and is calculated by the following formula. The wear severity index is calculated by the following formula,

[0065] I S =D L ×(D L -D S )

[0066] where D L is the proportion of large abrasive grains in the abrasive grain target image, and D S is the proportion of small abrasive grains in the abrasive grain target image.

[0067] S7: Component wear state evaluation model to comprehensively judge the wear state of the equipment.

[0068] The wear state evaluation model combines the quantitative index data automatically analyzed and calculated by the algorithm, and the subjective evaluation of the qualitative indicators by the operator after observing the abrasive grain image, and automatically analyzes and calculates the recommended evaluation conclusion, which is submitted to the operator as an aid, and the operator completes the final judgment result; the final judgment result includes the wear degree level, wear mechanism judgment, fault warning and life prediction.

[0069] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than limiting the protection scope of the present invention. Any simple modification or equivalent replacement made by those of ordinary skill in the art to the technical solution of the present invention shall not depart from the essence and scope of the technical solution of the present invention.

Claims

1. An intelligent ferrography analysis method, characterized in that: The following steps are involved: S1: By adjusting the horizontal position and height of the imaging system, a number of original wear particle images at different horizontal positions and heights are taken from one edge of the wear particle field of view to the other edge; S2: preprocessing the original wear particle image, performing depth of field fusion and image splicing on the original wear particle image, and obtaining the wear particle target image; S3: Send the wear particle target image to the Mask R-CNN network, extract the wear particle area through the segmentation network, obtain the accurate wear particle instance image, and obtain the shape contour of each wear particle; S4: performing graphic calculation analysis on the abrasive particle example diagram to obtain quantitative indicators, wherein the quantitative indicators include calculation indicators of each abrasive particle; S5: input each abrasive particle instance image into the convolutional neural network to perform preliminary classification of characteristic abrasive particles, revise the classification status of each abrasive particle in combination with the calculation index of the abrasive particles obtained in step S4, and determine the category information of the characteristic abrasive particles; S6: Based on steps S4-S5, determine comprehensive analysis indicators, including quantitative indicators and qualitative indicators of individual abrasive particles and overall abrasive particles; S7: Component wear status evaluation model, comprehensively evaluates the wear status of equipment.

2. An intelligent ferrography analysis method according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11: The imaging system takes the edge of the abrasive field of view as the horizontal starting position and the glass surface in contact with the abrasive field of view as the vertical starting position to capture the first abrasive particle image; S12: Keep the horizontal position of the imaging system unchanged, adjust the vertical height upward, control the upward height within the imaging depth of field range, take the next wear particle image, and repeat the above upward adjustment steps after completion, until the total height of the upward adjustment covers the maximum detection range of the wear particles, and obtain multiple wear particle images with different heights at the same horizontal position; S13: After taking images of abrasive particles at different heights at a certain plane position, the imaging system lowers its height to a vertical starting position, and starts to horizontally translate from the horizontal starting position toward the other edge of the abrasive particle field of view. After moving into position, the first image at the horizontal position is taken, and then the horizontal position is kept unchanged, and step S12 is repeated to achieve layered image shooting of abrasive particles in the area. S14: Repeat step S13 until the wear particle area photographed by the imaging system reaches the other edge of the wear particle field of view, completing the original acquisition of all wear particle images.

3. An intelligent ferrographic analysis method according to claim 2, characterized in that: In step S13, the distance of the horizontal translation does not exceed the length of the single imaging field of view in the horizontal direction, and there is a certain overlap between the two imagings.

4. The intelligent ferrographic analysis method according to claim 2, characterized in that: Step S2 specifically comprises: for all layered images of all heights obtained in the same horizontal position area, an image depth of field fusion processing algorithm is applied to extract the clear part of each image and synthesize it into a depth of field target image; then starting from the depth of field target image corresponding to the horizontal starting position, all depth of field target images are stitched together one by one according to the order in which they are taken using an image stitching algorithm, until all depth of field target images are stitched together into a final abrasive target image with a full field of view.

5. The intelligent ferrography analysis method according to claim 1, characterized in that: Specifically, step S3 includes: the Mask R-CNN network analyzes and identifies the abrasive target area and background area based on the artificial intelligence semantic segmentation model, extracts the abrasive target area through the segmentation network to obtain an accurate abrasive instance map, applies the edge detection algorithm to the abrasive target area, and extracts the shape contour of each abrasive particle.

6. The intelligent ferrography analysis method according to claim 1, characterized in that: The calculation indexes of each abrasive grain in step S4 include the major axis size, major / minor axis size ratio, circumference, circularity and projected area of ​​the abrasive grain.

7. An intelligent ferrography analysis method according to claim 6, characterized in that: In step S5, the categories of the characteristic abrasive particles include normal abrasive particles, cutting abrasive particles, severe sliding abrasive particles, fatigue flaky abrasive particles, spherical abrasive particles and adhesive abrasive particles.

8. An intelligent ferrography analysis method according to claim 7, characterized in that: In step S6, the qualitative indicators include the shape, contour, texture, color, position and direction of the abrasive particles, the presence or absence of abrasive particles with specific characteristic morphology, the presence or absence of abrasive particles with specific colors, and the presence or absence of non-ferromagnetic abrasive particles; the category of the characteristic abrasive particles is determined according to the qualitative indicators, and the number of characteristic abrasive particles of each category is recorded; The quantitative indicators also include overall evaluation indicators of abrasive particles, which include the total content of abrasive particles in a unit volume of oil sample, the number distribution of abrasive particles in each size segment, the coverage area percentage of abrasive particles, the proportion of large abrasive particles, the proportion of small abrasive particles, the wear severity index, and the average size of abrasive particles.

9. An intelligent ferrography analysis method according to claim 8, characterized in that: The coverage area percentage of the abrasive particles is the percentage of the total area of ​​the abrasive particles identified in the image to the image area, which is used to reflect the amount of wear; The number distribution of abrasive particles in each size segment is distinguished by the major axis size, and multiple size intervals are defined. The number of abrasive particles counted in each size interval is used to reflect the amount of wear and the severity of wear; The number of characteristic abrasive particles of each category corresponds to different wear states and is used to determine the cause of wear; The wear severity index is used to judge the severity of wear. S , calculated by the following formula, I S =D L ×(D L -D S ) Among them, DL is the proportion of large abrasive particles in the abrasive target map, and DS is the proportion of small abrasive particles in the abrasive target map.

10. An intelligent ferrography analysis method according to claim 9, characterized in that: Step S7 specifically includes: the wear state evaluation model combines the quantitative index data automatically analyzed and calculated by the algorithm, and the operator's subjective evaluation of the qualitative indicators after observing the wear particle image, automatically analyzes and calculates the recommended evaluation conclusion, and submits it to the operator as an auxiliary, and the operator completes the final evaluation result; the final evaluation result includes the wear degree level, wear mechanism judgment, fault warning and life prediction.

Images