Research On Intelligent Analysis Method Of Filtergram Images For Mechanical Equipment Safety

As the basis of industrial production,mechanical equipments can create significant benefits in safety and economy on the premise of ensuring safe operation,reducing breakdown rate and prolonging the service life.The lubricant oil used in equipment carries direct information which can reflect the running conditions of the specific components.The dynamic operation status can be obtained in a non-destructive mode through monitoring the suspended particles extracted from the oil in order to guarantee its long-term running in a safe,healthy,reliable and harmless state.With lubricant wear debris analysis technique as the core,this paper carries out a research about extracting and analyzing digitial information of wear debris images to improve efficiency and accuracy of the analysis.An intelligent analysis method of wear debris information is proposed which has complete procedure,high automation and can simultaneously fulfill qualitative and quantitative index extraction.Through the study,wear debris analysis work can realize systemized,standardization and intelligence.The main research contents and conclusions are as follows:(1)An intelligent extraction process of digital information about wear debris images was proposed.Taking the image acquisition as first step,the whole process was composed of macroscopic distribution information extraction aiming at whole wear particles and detailed information extraction aiming at individual wear debris.The development conditions,implementation methods and output results of each step were determined.(2)The macroscopic distribution characteristics of wear debris images were analyzed and an automatic extraction method was proposed.The threshold segmentation technique in HSV color space was used to separate metal wear debris and non-metal particles respectively.Then,a futher segmentation method based on the reconstructed watershed algorithm and the overlapping area ratio was applied to solve the problems of overlapping wear debris.Finally,two indicesN_d andA_d were selected to describe the macroscopic distribution of wear particles quantitatively.(3)Quantitative characterization and feature extraction methods of single metal wear debris were studied.Combining H-minima improved watershed algorithm and region auto-growing technique,the adaptive segmentation of metal abrasive image was realized,which avoided artificial interaction in the segmentation process and improved the segmentation efficiency significantly.Based on the segmented region and coutour,the color features,size-shape features,edge detail features and surface texture features were further extracted to form a twenty-dimensional parameter system,which fulfilled the quantitative characterization of wear debris morphology.(4)An automatic recognition model of wear debris based on Fuzzy Support Vector Machine was established.Selecting the characteristic parameter system of wear debris as input vector,a three-layer and seven-class classifier with simple structure and small cumulative error was indirectly constructed by combining one-against-all method and binary tree method.Wear debris would be automatically identified as one of the seven categories.The recognition rate of training samples and test samples was 90.71%and 92.86%respectively.(5)Hydraulic motors,reducers of rotary drilling rigs and wind power gearboxes were selected as monitoring points for oil sample collecting.The proposed analysis process was then applied to analyze the wear debris information distributed in filtergraphic image.The results indicated that the detection accuracy of hydraulic oil,vehicle gear oil and industrial gear oil in the process of macroscopic distribution characteristics extraction was 82.79%,95.79%and 97.04%respectively,and the recognition accuracy was 76.07%,83.33%and 80.47%in the automatic recognition of wear debris.There is a certain cumulative error during the implementation of whole process.