## Cathode surface inspection

The oxide cathode quality inspection is a very important issue for the overall quality of the cathode ray tube.

One conventional colour cathode ray tube contains 3 cathodes for red, green and blue light.

A cathode is not the most expensive part of the ray tube but it is indeed very crusial.
If one of these three cathode has a malfunction the cathode ray tube will be rejected.
A malfunction of a cathode can only be detected when the complete ray tube is assembled and a functional test is executed.

On the other hand it is very important that the working life of a cathode is at least 10 to 15 years.
To guarantee this lifespan the electron emitting surface has to meet a very high standard.
No damages or scratches are allowed. Also, emitting surface paste that is too thin or in-homogeneous is not allowed.
Damages, bad welding and malformed cathodes are also not allowed. For these inspections, read the separate descriptions.

## Vision inspection setup

There are two quality aspects that have to be inspected;## Damaged and thin paste.

For this inspection the ring light is positioned on a very high position, so the white surface (paste) will be showed as a dark area.The area with no paste (metal cap) or thin paste results in a total reflection.

The distance cathode ring light is about 120-130 mm and the diameter of the ring light is 70 mm. In this case an illumination top angle of 30 degrees is achieved. The grey level of the area outside the cap is relatively constant and black.

The grey level of the paste is very constant within a few grey levels, (“ grevel ”) but higher than outside the cap. The light will be reflected by the cap and a total reflection will occur at the places where there is no paste on the cap. A damaged part can be seen as a white area.

The first two images show a cathode cap almost completely covered with paste.

A small area with no paste at the edge of the cap is allowed.

These two images show damaged paste or a not ‘ fully ’ covered cap.

Of course the width and the area of the not covered parts can be measured. Compare this result with the accept/reject limits and the cathode can be accepted or rejected.

A combination of measurements and limits is also possible.

The left image shows a scratch. Only a small part of the scratch is visible and this illumination method is not sufficient enough to detect such scratches. The image on the right shows not only bright spots on the perimeter of the cathode cap but also in the centre. Thin paste results in this kind of images.

Extra software is needed to measure these separated bright spots to conclude that this cathode does indeed have a thin paste.

## Detection algorithm

The software is created with LabVIEW(National Instruments ™) and the vision library Promise from Philips Apptech Industrial Vision.For the white area detection rulers or edge detection algorithms are used. The rulers are positioned from the centre of the cap to the dark area outside the cap.

The first and last detected edges are used for determining the width of the damage. Graphical representation of the width (distance cap to paste) the damaged paste is given in the bottom picture.

The measured points (yellow) and the filtered points (red) (median filter, length= 5) so that single points or small areas will not be detected as damage.

The filter length is adjustable. In ‘delta points cap – paste’ (white) the width of the damage per ruler is displayed. The peak between 8 and 20 is the damaged white part the width is about 30 pixels.

The first ruler starts in +x direction representing measure point 0.

If the paste is not damaged; one or no edges may be found.

If one edge is found the width of the damage is 0. (see results, measuring points 120 and 121). If no edge is found (measuring points e.g. 112 till 119) the contrast between the area outside the cap and the paste is too low. this does not affect the detection reliability.

## Quality inspection of the paste.

For the paste quality inspection a different type of illumination is used.The best contrast is achieved with shearing illumination. This is a circular illumination which is positioned just above the surface of the cathode. This illumination accentuates the structure of the paste.

In the setup the ring light is positioned just above the cathode. A cathode with good paste results in a ‘rough’ texture. In technical jargon they call it a ‘cauliflower’ structure.

When the texture is rough the surface area for emitting electrons is as large as possible. These two images do have a thin paste so these have to be rejected. Cathodes with a thin surface give a dark area, scratches give a long small dark line.

The left image is a cathode with an overall thin paste. The intensity (grey level) is a little lower than that of an image with a good paste. The local deviation of the intensity (grey level) is much more (2 to 2.5 times normal deviation).

The paste ‘thickness’ is measured using the local intensity deviation.

On the right image the local intensity at the top left area is lower, the deviation in that area especially on the transition area is much higher than normal. So both local average (mean) and local deviation are used for detecting these cathode defects.

On the left, a scratched and a ‘wet’ paste image is given.

The scratch on the left image is a small dark line and is detectable with local average and local deviation.

If the area used for calculating the mean and deviation is too large then a thin scratch can’t be detected.

‘Wet’ paste is paste where the surface is not rough enough so it looks wet.

Measuring the local average gives no difference with respect to a normal structure.
Measuring the local deviation gives a good result; the ‘wet’ area’s gives a lower deviation.

The paste detection causes some problems on the edges of the paste or cap.
So only a certain circular area, with a smaller radius then the radius of the cap can be measured.

## Detection algorithm

The software contains the next steps:1. Detection and measuring the centre and radius of the cap.

2. Acquire a sub image of the cap.

3. Detecting the valid sub areas (kernel).

4. Calculate the mean and deviation (variance) of the sub area.(kernel) 1. Detecting the position of the paste or cap is done by means of a circle fit. The result of this routine is the centre and the radius of the paste or cap.

2. The inspection area is the cap area which is a circular area. A sub image is created to get a square part out of the existing image with a length equal to the diameter of the cap. This sub image will be processed by means of a kernel.

The position of the kernel starts at the left top of the sub image and shifts stepwise to the right. The kernel, represented here as a small square shifted in horizontal direction (x direction) from left to right. The shift distance is half the length of the kernel (Two dimensional Nyquist-Shannon theorem).

The next step is a shift, over half length kernel distance, down in y direction followed by shifts in x- direction. To determine if a kernel is a valid one, the distance of every kernel corner with respect to the centre is calculated. Only the kernels with all 4 corners positioned within a circle with a radius of: ‘radius – delta r’ are valid kernels.

Statistical calculations are done only on the valid kernels.

In the experimental set up you can choose between three ‘analysis types’ :

1. histogram, deviation

2. square deviation

3. exponential deviation

These three analysis types are explained below.

## Histogram, deviation

The software calculates a histogram of the grey levels of every kernel followed by the calculates of the mean and deviation.The results are displayed in:

**mean histogram**(left) and

**st.dev. histogram**(centre). From the st.dev. histogram (right) a slice is displayed (right) in the graph:

**var.histogram**at level (row, horizontal line yellow).

Interpretation of the

__mean grey results__:

If the

__mean grey value__from

**mean histogram**(left)is above the threshold-high-limit (red value, left) the results in the graph turns to red.

If the

__mean grey value__from

**mean histogram**(left) is under the threshold-low-limit (yellow value, left) the results in the graph turns to yellow. Here a dark spot is located, this could probably be thin paste. Interpretation of the

__deviation results__is slightly different.

If the

__deviation value__from

**st.dev. histogram**(centre) is above the threshold-high-limit (red value, right) the results in the graph turns red, so a bright

**dark spot (or area) will be located.**

__OR__If the

__deviation value__from

**st.dev. histogram**(centre) is under the threshold-low-limit (yellow value, right) the results in the graph turns

__. In case of ‘wet paste’ the variance is very low which results in a__

**Black**__black spot__.

If the

__deviation value__is between the threshold-high-limit (red) and threshold-low-limit (yellow) the results turn

__yellow__.

As we see, a

__good__paste has a certain deviation, the

__area indicates a__

**yellow**__deviation and a__

**good**__cathode paste.__

**good**## Square deviation

The software calculates the mean and the variance of every kernel; tis is called “square deviation” :**“square deviation” = {SUM(xi - xmean)2}/(n-1)**From the square dev. a slice is displayed in the :

**square dev**(right) and at level (row, yellow horizontal line).

The interpretation of the results is comparable with the interpretation of the results in the histogram and deviation pictures. If the

__mean grey value__is red, the mean value is above the threshold-high-limit. The mean grey value is yellow when the mean value is under the threshold-low-limit.

This means yellow is thin paste.

Interpretation of the

__deviation results__is comparable with the interpretation of the results in the histogram and deviation pictures.

Yellow indicates a paste with enough roughness, red indicates too much roughness and black indicates too less roughness.

## Exponential deviation

From every kernel the mean and a ‘exponential variance’ (‘exponential deviation’) is calculated:**‘exponential deviation’= {SUM(xi - xmean)4}/(n-1)**

In this case for the exponent the value 4 used. The results are displayed as follows:

**mean exponential dev.**(left) and

**variance exponential dev.**(on the centre).

From the exponential dev., a slice is displayed in:

**variance exponential dev.**(right) and at level (row, yellow horizontal line).

The interpretation of the results is comparable with the interpretation of the results in the histogram and deviation pictures.

If the

__mean grey value__is red, the mean value is above the threshold-high-limit; The mean grey value is yellow when the mean value is under the threshold-low-limit. This means yellow is thin paste. Interpretation of the

__deviation results__is comparable with the interpretation above.

Yellow indicates a paste with enough roughness, red indicates too much roughness and black indicates too less roughness.

As we see, a

__good__paste has a certain deviation; the yellow area indicates a

__good__deviation and accordingly a

__good__cathode paste.

## Results of Wet paste, Scratch and Thin paste

Here, some results are displayed: wet paste (left), scratch (centre) and thin paste (right).The overall results show that the exponential deviation method is a very sensitive method.