Research On Tool Wear Monitoring Technique Based On Micro-vision

Tool wear monitoring technique is important to realize automation and intelligent ofmodern production. The key techniques of tool wear monitoring based on micro-vision aredeep researched and discussed. The key techniques include microscope autofocus, thesegmentation of tool wear area and the analysis of workpiece texture which are deepresearched. The research provides a base to realize the system of tool wear monitoringpractically.The major innovations of this paper are as follows:(1) Aiming at an image with strong noise environment, the traditional focus evaluationfunction can not satisfy the focus need. An improved processing of focus evaluationfunction is provided. The image is preprocessed firstly. Then this image is over-segmentedinto many blocks with watershed technique.The gray values of pixels in a certain block arereplaced by the mean gray of this block which can reduce the noise influence on focusevaluation function. The experimental results indicate the availability and practicability ofthis algorithm.(2) Aiming at the misjudgment and poor real-time of the traditional searchingalgorithm, an improved hill climbing method with adaptive-step is proposed. Twothresholds are set in this search algorithm. The value of search step is dependent on therelationship among the slope of adjacent positions, the two thresholds and the factor oflocal extreme. The step value is divided into three conditions, large step, medium step andsmall step. This searching algorithm can not only reduce the situation that the local extremeposition viewed as the focal plane but also can reduce the situation of the focal planemissed when searching with large step. This algorithm has lower computation and betterreal-time performance.(3) The traditional Markov random field has huge computation and is sensitive tonoise when used for the segmentation of tool wear area. A new algorithm of self-adaptiveMarkov random field based on region is proposed. The preprocessed image is divided intoblocks with watershed technique. The mean gray and variance of regional block are viewedas feature parameters for initial segmentation. The connection parameter of potentialfunction is determined adaptively in terms of the close connection degree between the current block and its adjacent block. It satisfies the mechanism of image segmentation. Thisalgorithm is more accurate for the edge segmentation of tool wear area. Throughexperiments, accuracy and robustness are improved.(4) In view of the low contrast images, the traditional threshold segmentationalgorithm has performance, so an improved segmentation technology of gray levelco-occurrence matrix is proposed. The improved gray level co-occurrence matrix isconstructed by the pixel value and weighted average value of its neighborhood. Then thethreshold is obtained by this improved gray level co-occurrence matrix. The generating stepvalue is a key parameter to construct gray level co-occurrence matrix. The featureparameters of co-occurrence matrix are simulated with different steps. The extreme positionof simulation curve in the first cycle is the best step value. Experiments show that the beststep value is better for the analysis of tool wear.(5) In order to improve the accuracy and robustness of tool wear monitoring, the valueof tool wear area and the feature parameter of worpiece texture are comprehensivelyutilized. Compared with single criterion, this method has higher accurate and robust.