| 摘要 |
PROBLEM TO BE SOLVED: To acquire the accurate coverall quality information by applying a neural network that is learnt by a fuzzy network output vector to perform the test and the quality evaluation of a paste type solder brick. SOLUTION: When a printed circuit board 12 is detected by a sensor 26, a computer 28 collects the images of a prescribed group of paste type solder bricks through a camera 30. Then the computer 28 positions a marked solder brick image to apply the proper digital image processing to it, and the value of a prescribed pairs of brick computational complexity are extracted in a digital image. Then the clear input and output vectors are stored in a 1st learning file. The pair of these input and output vectors are turned into a fuzzy one based on a similar relation and written in a 2nd learning file. The backward learning is applied in a neural network by means of the vector contained in the 2nd learning file, and this learning is discontinued when the satisfactory performance is attained.
|
