| Summary: | 碩士 === 國立成功大學 === 電機工程學系碩博士班 === 91 === Traditional inspection is manual, unreliable and inefficient. Therefore, automatic inspection would have great benefits and have been widely adopted for various automatic visual inspections in today’s industry. This thesis investigates three major issues concerning an automatic inspection system: the classification of BGA chips, the alignment of IC packages and the measurement of MLCC.
In classifying the BGA chip, the blob analysis method and the skeleton extraction method were used. In aligning the IC package, the blob analysis method and the Hough transform method were used for direction determination respectively. Linear regression analysis and region growing were used to find the center position and angle of an IC package. In measuring deviations of layer widths of the MLCC, 2nd order regression analysis was applied to obtain the datum curve. Fast Fourier transform was used to find the number of layers. Feature point alignment was used to align the edge points to the best positions. Finally, deviations from the datum curve to the positions of edge points were calculated.
In the experiments, two BGA products with total 102 images were used for classification. According to the results, all 102 images were correctly classified. 20 ms and 225 ms in average are required by blob analysis and skeleton extraction respectively and the average of image size is 226*228. An IC product with total 76 images were used for alignment. According to the results, the RMS errors of the horizontal and vertical positions are 0.316 and 0.459 pixel(s) respectively and the angle is 0.121'' . 81 ms in average is required and the average image size is 628*337. As for direction determination, the correct rate is 93.4% and 5 ms in average is required in the blob analysis method, and the correct rate is 97.4% and 310 ms in average is required by the Hough transform method. Two MLCC products with total 8 images were used for measurement. According to the results, the average deviation is 3.19 pixel(s) and the average error of edge point position is 1.433 pixel(s). 883.5 ms in average is required and the average image size is 768*576.
|
|---|
