Retraction Retraction: Research on 3D reconstruction method of wear particle dynamic image based on multi contour space

. Aiming at the problems of current image monitoring methods of lubrication oil wear particles, this paper designs and builds a dynamic monitoring system for oil wear particles based on microfluidic microscopic images. A contour-based 3D reconstruction method of debris particle images is proposed. The image sequences of rotating wear particles tracked by a single target are used as data, and the contour of the wear particle is extracted and the data is stored. The minimum area external rectangle method is used to correct the rotation of the particle images for the problem of deflection. And an algorithm based on cylindrical coordinate space conversion is used to convert the discrete contour point data into three-dimensional space. Complete the 3D model reconstruction of microfluidic wear particles. The ability to analyse wear particles in oil online monitoring technology is improved, which also shows new ideas for wear status monitoring and fault diagnosis technology.


Introduction
Rotation mechanical equipment wear and failure will produce different wear particles, and the identification and detection of abrasive particles in the lubrication oil is particularly important. Oil wear particles contain a large amount of mechanical wear and failure information. Analysis of their size, quantity, color, morphological characteristics and contour shape, combined with the working status and wear mode of the system, can effectively determine the degree and type of wear, analyze the cause of the failure, and locate the worn area. As an important and efficient wear detection method, oil abrasive particle monitoring can determine the wear status of equipment or key parts after the wear particle information is clear, so as to realize the current performance analysis of the machine and equipment and the prediction of the next failure development trend, and finally make a specific decision.
At present, there are many researches in the direction of wear particle analysis, and there are many studies on the two-dimensional characteristics of wear particles, and they have been used in oil monitoring experiments. However, with the continuous development of image processing technology and high-speed vision technology, the three-dimensional image analysis method of wear particles has IOP Publishing doi: 10.1088/1757-899X/1207/1/012017 2 gradually become the focus of research in this direction. The three-dimensional reconstruction analysis method of wear particles is still in the preliminary research stage of studying two-dimensional plane parameters. In the aspect of wear particle analysis, systematic research on the methods of 3D reconstruction of wear debris particles has not been carried out, and there is no complete theory to characterize the morphological characteristics of 3D debris particles [1].
In recent years, scholars have begun to study the 3D reconstruction technology of wear particles. In 2005, He [2] studied the method of ordinary optical microscope to reconstruct the three-dimensional surface of abrasive particles, improved the ordinary optical microscope, and conducted preliminary three-dimensional ferrography analysis. However, limited to the speed and effect of the algorithm, the analysis of wear particles has not been carried out, so the follow-up in-depth research is of great significance. In 2016, Wu [3] studied on-line extraction of wear particle features based on video sequences and three-dimensional representation models. A video-based rapid 3D reconstruction system for two-dimensional wear particles is proposed, which completes the dynamic wear particle extraction based on the Gaussian mixture model, uses the acquired feature information to track the wear particles, and realizes the three-dimensional surface reconstruction of the simulated wear particles in multiple views. Because the system is affected by the image quality, image processing methods such as the motion recovery and enhancement of the wear image need to be studied in depth to improve the image quality of the wear image. In 2017, Peng [4] used the advantages of video image processing to propose a system and method for three-dimensional wear particle analysis, and obtained that the wear particles in the oil rotate under the action of torque, and images of different views can be used. Sequence to build the important conclusions of the three-dimensional model. And put forward that the future research direction is to optimize the experimental system, improve the accuracy of wear particle detection, and perform three-dimensional reconstruction in multiple views.
In order to obtain more comprehensive information on wear debris, the technology of on-line visual ferrograph (OLVF) technology is being developed [5]. Li designed a new reflecting light based on LED arrays to solve the problem of ferro-imaging quality that is affected by the lower-brightness and non-uniform illumination of the original reflecting light [6]. In order to automatically acquire threedimensional information of wear debris for online condition monitoring, Peng developed a microfluidic device consisting of an oil flow channel and a video imaging system. A microchannel is designed to ensure the continuous rotation of particles such that their three-dimensional features can be captured [7]. Liu put the application of magnetic fluid in ferrograph to improve the technique for separating nonferrous wear debris by magnetic means [8].
In image analysis and processing algorithms, Peng applied two-dimensional fast Fourier transform, power spectrum and angular spectrum analyses to describe wear particle surface textures in three dimensions, and developed a software for the image analysis of boundaries and surfaces. [9][10][11]. Wu proposed the description of wear debris from on line ferrograph images by their statistical color. The method provides a primary exploration on describing the color of wear debris by on-line ferrograph images [12]. Wu also investigated statistical dimension of wear debris in on-line ferrograph images, in order to descript debris images for wear characterization [13]. Wu proposed the method of incorporating gray and boundary based segmentation, and watershed-based morphological separation of wear debris chains for on-line ferrograph analysis [14,15]. Wu present the development of a videobased system to extend the particle information in 3-dimension (3-D). The proposed method contains three main procedures including: particle extraction using a Gaussian mixture model, multiple particle tracking with Kalman filter, and 3-D feature reconstruction by the shape-from-silhouette method [16,17]. Xi had researched the restoration of online video ferrograph images for out-of-focus degradations. The main idea is to extract object edges, magnify with a non-linear gain factor, then combine with the input image to produce an enhanced image to facilitate further analysis [18]. Liu proposed an unsupervised segmentation for wear particle's image using local texture feature [19]. To address the image degradation issue (debris images captured often suffer degradation due to debris motion blur and lubrication contamination), Wu developed a new method of wear debris image IOP Publishing doi:10.1088/1757-899X/1207/1/012017 3 restoration to reduce the effect of blur [20]. Peng used a Multi-SVDD method to model dynamic oxides identification [21].
Therefore, the three-dimensional reconstruction analysis of wear particle images has become an important research direction of oil online monitoring. The optimization of the experimental system of oil wear particle online monitoring has become the key to the study, and the processing algorithm of wear particle images also needs further research. In this paper, a set of simple and reliable 3D wear particle reconstruction methods for the whole process is proposed to meet the above needs, which provides a new and important means to obtain abundant debris information.

Experimental platform design
In order to collect two-dimensional wear particle image, combined with microfluidic chip technology and high-speed vision technology, an oil dynamic monitoring system based on microfluidic micro image analysis technology is designed. The system structure is shown in the figure 1. The system is mainly composed of dynamic injection system and microscopic imaging subsystem. The main functions of the two subsystems are to realize the oil supply injection of the system and acquisition and analysis of wear particle images in oil. The injection dynamic subsystem mainly includes oil pool, micro pump and related oil pipeline. The microscopic imaging subsystem includes a light source, a high magnification imaging lens, a high-speed visual digital camera, and a computer. When image data is acquired using this system, the uniform linear motion between the sensor and the scene generates the image blurring phenomenon. In this paper, we analyze and study the degradation model of the image, and apply a method based on the degradation function to recover the abrasive grain blurred image; and then use a target detection method based on visual saliency to achieve abrasive grain target detection.

Three dimensional reconstruction method of wear particle dynamic image
In this paper, the micro flow oil monitoring system introduced in the previous section is used to obtain the dynamic two-dimensional image of wear particles. A three-dimensional reconstruction method of wear particle dynamic image based on multi contour space mapping is proposed to realize the threedimensional reconstruction of wear particle. The overall process is shown in the figure.

Particle profile extraction
Image contour detection is an extremely important step in computer vision. Generally speaking, for the contour detection of irregular object image, firstly, the edge detection operator is used to extract the object edge. Then the redundant pixels are removed according to the contour characteristics of the object. Finally, these edge pixels are connected to determine the contour boundary of the object. Wear particle contour extraction is also a key process in wear particle 3D reconstruction. Because of the impurities and image background noise in the microchannel, the wear particle contour extraction must be able to effectively suppress the noise and accurately detect the position of the edge. In this paper, the wear particle image edge detection algorithm combined with background difference and Canny operator is used for contour extraction to realize the extraction process of wear particle contour in 3D reconstruction. For the debris particle image in micro flow, firstly, the particle image is segmented to suppress the interference of bubbles, impurity particles in background oil and chip pollutants that may appear in the particle image. And then the canny edge detection operator is used to extract the contour of the abrasive particle image to extract the abrasive particle contour accurately.

Image deflection correction
Under perfect conditions, the debris particles will rotate stably around a fixed axis. However, in the microfluidic chip channel of the system, the fluid movement is complex, and the abrasive particles will deflect at a certain angle. The deflected wear particle contour image is shown in the figure 2(a).
Before the next step of three-dimensional reconstruction of abrasive particles, the deflected particle image must be corrected first. Before image deflection correction, it is necessary to calculate the angle of abrasive particle deflection. In this paper, a minimum area circumscribed rectangle method is used to calculate the minimum area circumscribed rectangle of the abrasive particle profile, and the abrasive particle deflection angle is calculated through the rectangular axis of symmetry. The specific algorithm is as follows: Step 1: Store the data, input the abrasive particle contour image, and define the data   , xy as the coordinates of debris particle boundary points. Define the rectangular coordinates of the boundary as   , rectx recty .
Step 2: Search the boundary rectangle, classify according to the number of coincident points between the boundary rectangle and the debris boundary, round off the boundary rectangle with the number of coincident points of 0, 1 and 2 respectively, keep the boundary rectangle with the number of coincident points of 3 or more, and execute the cycle.
Step 3: Calculate the minimum boundary rectangular area, save   , rectx recty and obtain the minimum area circumscribed rectangle.
Step 4: Obtain the angle of each boundary of the minimum area circumscribed rectangle, and calculate the deflection angle of the rectangular symmetry axis.
Through the minimum area circumscribed rectangle method, the abrasive particle profile and the minimum circumscribed rectangle are obtained, as shown in the figure 2(b), in which the green rectangle is the minimum area circumscribed rectangle.

Particle boundary point extraction
Due to the uncertainty of the flow of wear particles in the oil, in order to accurately store the twodimensional data   , L x y of wear particle boundary points in the same coordinate system, it is necessary to establish a unified two-dimensional coordinate system. Firstly, the center of the abrasive image is determined, then the unified two-dimensional coordinate system is determined with the center In the binary image, the area of wear particles can be determined by the number of pixels in the image particle area. The center of the particle is the area center of the irregular abrasive particle. The calculation formula of the center coordinates ( , ) xy of the abrasive particle image with different attitudes is shown in the formula.
i i Where, N is the number of pixels in the wear particle area in the image. i x  is the sum of abscissa of pixels in the particle area, and i y  is the sum of ordinate of pixels in the particle area.

Three-dimensional space mapping
The abscissa and ordinate systems are established by using the contour image stored by the particle points. The two-dimensional debris contour image of the established coordinate system is shown in Figure 3. In the established plane rectangular coordinate system XOY, the two-dimensional coordinates   , A x y of any particle contour point have their corresponding coordinates   ,, A x y z  in threedimensional space. Because the particle rotates relative to the rotating axis, the abrasive particle contour is different in three-dimensional space. Therefore, this paper adopts an algorithm based on cylindrical coordinate transformation to realize the mapping from two-dimensional contour to threedimensional space. Firstly, a plane rectangular coordinate system is established on the contour image stored by the particle boundary points, and the coordinate origin is located in the upper left corner of the image. The coordinate data of particle contour points are determined by the plane rectangular coordinate system, and the two-dimensional coordinates of all particle contour points are stored.
Then convert particle  

,,
A x y z  coordinates to three-dimensional polar coordinate system, and store particle polar radius r, polar angle θ and height value z. The cylindrical coordinates   ,, rz  are used to determine the position of each contour point of particles in three-dimensional space. The cylindrical coordinate system is transformed into Cartesian coordinate system, and the mapping from polar coordinates and cylindrical coordinates to Cartesian coordinates is shown in the figure 4. The stored polar radius and polar angle of particle contour points are mapped to Cartesian coordinates, and the height value remains unchanged to form three-dimensional abrasive particle contour point coordinates   ,, x y z .  In this paper, the above algorithm based on cylindrical coordinate transformation can quickly and accurately map the wear particle from two-dimensional contour to three-dimensional space. The twodimensional abrasive particle contour image in Figure 4 is mapped to the particle contour image in three-dimensional space, as shown in Figure 5. In the figure, the edge pixels of abrasive particle contour in three-dimensional space are represented by blue dots.

Experimental test
The collected lubricating oil samples were monitored on the experimental platform described in Section II, and nineteen original wear particle images were obtained as shown in the figure 6. Through the wear particle segmentation method based on oil background difference, 19 images of the same wear particle are segmented. Canny edge detection algorithm is used to detect the edge of 19 segmented images of the same wear particle, and finally realize the contour extraction of particle image. After edge detection, 19 contour image sequences of the same wear particle are shown in the figure 7. It can be seen from the figure that after the wear particle image edge detection algorithm based on oil background difference and Canny operator is used for contour extraction, the contour line of particles is very clear and fine, the boundary is completely closed. However, due to the deflection of wear particle contour image, further particle image processing needs to be completed.
After the minimum area circumscribed rectangle is obtained, the wear particle deflection angle can be obtained by using the rectangular symmetry axis, and then the wear particle contour image can be corrected according to the deflection angle. Then, the wear particle contour image is corrected according to the deflection angle to obtain the wear particle contour image after deflection correction.
After determining the wear particle image center, taking the particle image center as the reference point, a unified particle boundary point stored contour image is established for nineteen corrected wear particle contour images. The corrected contour images used for the storage of debris boundary points are shown in the figure 8, and these contours are stored in two-dimensional data. The two-dimensional contour to three-dimensional space mapping method is used to convert the wear particle image in Figure 8. After conversion, nineteen contour images of the same wear particle in the three-dimensional space are shown in Figure 12. Finally, use Geomagic Design X software for rendering to form the three-dimensional model surface of wear particles. The front view, side view, top view and oblique view of the threedimensional model of debris particles are shown in Figure 11    From the reconstruction results of the three-dimensional model of wear particles, it can be concluded that most of the morphological information of particles can be reflected, including the thickness, width, length and aspect ratio of abrasive particles. Under a certain measurement error, deeper parameters such as the volume of abrasive particles can be estimated. In this paper, a three-dimensional reconstruction method of wear particle image based on contour is proposed. For the same wear particle tracked in microchannel, the three-dimensional model reconstruction based on its contour image sequence is realized.

R e t r a c t e d
Firstly, the image sequence of rotating wear particles tracked by single target is extracted as data, and the relationship between the frame sequence of the particle image and the corresponding particle rotation step is analyzed in detail. Then, the particle image edge detection method combined with oil background difference and Canny operator is used to segment and detect the edge of 19 images of the same wear particle, and the particle contour is extracted. Aiming at the problem of particle deflection, a minimum area circumscribed rectangle method is used to calculate the minimum area circumscribed rectangle of abrasive particle contour. The deflection angle of abrasive particle image is calculated through the rectangular symmetry axis, and the deflection correction of abrasive particle image is completed. After determining the abrasive particle image center in the binary image, taking the particle image center as the reference point, a uniform size contour image for boundary point storage is established, and the two-dimensional data of particle boundary points are stored. An algorithm based on cylindrical coordinate transformation is used to convert the image data to three-dimensional space, realize the mapping from two-dimensional contour to three-dimensional space, and establish the three-dimensional scatter diagram of wear particles. Finally, the three-dimensional model reconstruction of micro flow abrasive particles is completed, which provides more abundant information for wear particle analysis technology.
In future work, the authors will investigate how to acquire clearer and more complete images of wear particles and construct more sophisticated 3-D models of the particles. At the same time, the quantitative evaluation of reconstruction results should be completed. Finally, continue to study the real impact of three-dimensional reconstruction of wear particles in lubricating oil on condition monitoring during mechanical rotation, such as condition prediction and wear analysis of mechanical structure.