Mechanical Power System Fault Diagnosis Method Based On Ferrography Image Processing Technology

Ferrography is through the analysis of wear particle morphology, number, size distribution and composition and other characteristics, to identify the machine lubrication and wear failure mode, extent and state of the machine in which to determine the reasons for failure, so as to further equipment maintenance decisions. Ferrography image recognition technology and computer graphics and image processing technology combined with Ferrography analysis, with great objectivity, the identification of wear particle can be fast, effective, quantitative analysis, is the wear areas of diagnosis and a hot Ferrography analysis research.This paper selects Ferrography color images of diesel engine, using digital image processing, pattern recognition theory , tribology knowledge and Wear Debris characteristic parameters of digital extraction and optimization, the support vector machine applied to the Wear Debris Image Recognition of Wear Debris in pattern recognition , verifying feasibility using support vector machine in diesel engine Ferrography Wear Debris of fault recognition, and also provides a new and efficient method for diesel engine Ferrography image recognition .This article firstly preprocessing the original diesel engine color images (including image geometric transformation, graying, enhanced filtering, image sharpening clarity, image segmentation, image contour extraction processing), Ferrography wear particle identification is applied directly after the pre-selected on the abrasive Ferrography method, select the target particle. For three particle types (sliding abrasive, abrasive cutting abrasive and fatigue), the first classification of the characteristic parameters (size parameters, profile shape parameters, structural parameters, color feature parameter), and through the sample particle training learned to get the size of the characteristic parameters; Whereas Wear Particle many parameters, this paper by feature optimized for this study to determine the type of recognition particle parameters required for Category 8, the experiment proved that support vector machine parameters in the particle identification accuracy aspects.