The Precision Thin Component Edge Measurement System In Micrometer Level:its Analysis And Design

The precision thin part is a kind of component with tiny dimension. It is widely used and plays an important role in many industry applications. The edge contour of this kind of part can highly affect its performance. Whereas, due to its small scale, usual method can hardly make good dection to such components.Photogrammetry is a modern industrial inspection method, and its core technique is digtial image processing. in this paper, we consider such situation that the dimension of the image pixel is close to the required detection precision, the preformance of the edge detection algorithm will be strongly influenced by even only one single pixel. In general, most traditional edge detection methods preform de-fuzzy operations, mapping each pixel into a binary set. However, under the pixel dimension, the information of the pixel is always fuzzy, such de-fuzzy operation will result in the loss of precision.Based on such consideration, a new approach using fuzzy set is proposed. We assume that every pixel in the image can be edge. Furthermore, we propose a full-directiongradient operation, use this gradient as the menbership degree of the element in the fuzzy set. Meanwhile, in order to reduce the number of the element in the fuzzy set, a filter based on pixel siganl energe is introduced. It regards the variance of the pixel block as the signal of how large the possiblity the pixel belongs to the edge of image. Then the pixel with larger siganl energe value will be chosen.Finally, the edge fuzzy set under pixel cordinate will be transformed to the polar cordinate. These data will be compared with the CAD data of the part to get the error value. The error will be picked when its value goes outside the pre-defined threshold.According to the above method, we can find out that in the sample image, the 69 th gear tip(with azimuth as 347.3229 352.0326 348.7766 350.7192 351.3807 347.6453 347.3832)exists obvious abnormality. After manually checking, it is shown that the miss rate of this algorithm is 0% while the false alarm rate is 60%. Thus it can detect the abnormality effectively.