The Key Technology Research Of Online Visual Ferrography Image Wear Debris Segmentation

In modern production,the requirements for the stability,processing accuracy,and reliability of mechanical equipment has becomed higher and higher.It has maked traditional equipment maintenance methods no longer able to meet the demand.From the initial offline monitoring of the status of mechanical equipment to the real-time monitoring today,the requirements for the condition detection and fault diagnosis of mechanical equipment are also increasing.At present,the ferrograms analysis technology has been widely used in the wear detection of mechanical equipment.The quantitative and qualitative analysis of the wear particles in lubricating oil has used to realize the state detection and fault diagnosis of mechanical equipment.The combination of the ferrograms analysis technology and the image processing technology have maked it more widely used.The real-time monitoring can be performed for different working conditions and equipment.Therefore,in order to improve the level of automation and intelligence of the ferrograms analysis technology,this thesis has studied the accuracy of segmentation of the ferrograms.In this thesis,the corresponding solutions are proposed for the accumulation of wear particles,chain formation problems and unevenly illuminated ferrograms.The solutions has compared with some commonly used methods to verify and analyze.In this thesis,the research on the field of the ferrograms wear particles segmentation and recognition has been carried out.Firstly,the domestic and foreign research status of the ferrography analysis technology and the ferrograms processing technology,as well as some related theories in image processing have been analyzed.In order to facilitate the study of the ferrography analysis technology,a ferrograms wear particle segmentation recognition system has been built.The system is composed of hardware and software platform.The hardware part mainly includes friction and wear experimental platform and On-line Visible Ferrograph(OLVF).The software part mainly includes ferrograms processing and wear particles segmentation and recognition functions.For the problem of accumulation and chain formation of the wear particles in ferrograms,an on-line visible ferrograms wear particle chain segmentation method based on the nearest-neighbor method has been proposed.Using the nearest-neighbor method to match the ferrograms at different moments to obtain the deposition change sequence of the wear particle chains.Each wear particle area has been divided by comparing the two adjacent wear particle chains in the deposition change sequence.The improved distance transformation method has generated marks for each wear particle.And using the markercontrolled watershed algorithm to have segmented the wear particle chain to obtain the wear particles segmentation image.Finally,the segmentation results have been compared with the traditional watershed algorithm and multi-scale binary morphology segmentation algorithm to verify the accuracy of the proposed method.This thesis have solved the problem of over-segmentation and under-segmentation during the wear particles segmentation process.The wear particle chain has been separated.For the problem of uneven illumination of ferrograms,a method of ferrograms segmentation based on the lattice has been proposed.Introducing the lattice on the ferrograms to mark the wear particles.In order to reducing the influence of uneven illumination by converting the color space of the ferrograms.Then the method has used the clustering method to extract the marked wear particles to obtain the ferrography rough segmentation image.The operation process of morphology has been improved,and it has superimposed with the rough segmentation image after dilating once and eroding twice to ensure the integrity of the boundary information of the wear particles,and to obtain the binary wear particles segmentation recognition image.Finally,the recognition results have been compared with the Otsu method and the adaptive threshold method.The results have showed that the proposed method can more accurately identify the wear particles and lay the foundation for the subsequent feature extraction of the wear particles.