Browse Prior Art Database

# Algorithm for Detecting Solder Joints' Defects using K-L Transformation

IP.com Disclosure Number: IPCOM000106510D
Original Publication Date: 1993-Nov-01
Included in the Prior Art Database: 2005-Mar-21
Document File: 6 page(s) / 177K

IBM

## Related People

Fujita, T: AUTHOR [+2]

## Abstract

Disclosed is an algorithm that would be effective in detecting defects of solder joints for fine pitch QFP components using video images. First of all, several characteristic values (brightness of each joints) will be obtained from video images. The main mathematical treatment is to make linear combination of characteristic values being referred to as the value P. If P is greater than Pc (threshold value), this joint will be judged as bad joint. Characteristic values will be K-L transformed so as to calculate weighting factors of linear combination. This operation is carried out before the current inspection process. The threshold value Pc will be determined by carrying out statistical treatment of teaching card data. The results of judgement were well consistent with the results of inspection obtained by inspector.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 37% of the total text.

Algorithm for Detecting Solder Joints' Defects using K-L Transformation

Disclosed is an algorithm that would be effective in detecting
defects of solder joints for fine pitch QFP components using video
images.  First of all, several characteristic values (brightness of
each joints) will be obtained from video images.  The main
mathematical treatment is to make linear combination of
characteristic values being referred to as the value P.  If P is
greater than Pc (threshold value), this joint will be judged as bad
joint.  Characteristic values will be K-L transformed so as to
calculate weighting factors of linear combination.   This operation
is carried out before the current inspection process.  The threshold
value Pc will be determined by carrying out statistical treatment of
teaching card data.  The results of judgement were well consistent
with the results of inspection obtained by inspector.

The several video images of soldered joints are taken so as to
calculate characteristic values which are average brightness of each
windows.  Several images are taken under several illumination
patters.  Then several characteristic values (aj) can be obtained.
Making use of these values, each soldered joints are judged, i.e.,
being classified into two categories (good joints and bad joints).
The simplest way of judgement is to use one value (P), i.e., if this
value is greater than threshold value (Pc) it will be judged as a bad
joint.  To calculate this value (P) from several characteristic
values (aj), linear combination of characteristic values will be
calculated at first.  These procedures are well explained in Fig. 1,
where ajs represent each characteristic values being the same as each
windows' average brightness obtained from video images.

An example of ajs is illustrated in Fig. 2, where 11
characteristic values are obtained from four video images taken by
four video cameras.  Four video images are taken under different
illumination patters.  The weighting factors (Uj) could be
analytically calculated by using of K-L transformation of sample data
obtained from other experiments.  Determination of weighting factors
is equivalent of determination of plane which divide characteristic
space into two regions.  One is that for good joints and the other
for bad joints.  The way of determination of weighting factors (Uj)
will be explained in " 2 the determination of weighting factors ".
In Fig. 1 there is another unknown parameter Pc which is called as
threshold value for judgement.  The determination of Pc is somewhat
difficult if compared to that of Uj.  Because the setting of the
value Pc is flexible and it can be set empirically by man.  What
value we will make it would much influence on the effectiveness of
the judgement.  If it is set it at much lower value, the less leaking
mistakes occur.  However, overkill mistakes will take place much
more.  Once the all parameters Uj, Pcare determined, w...