At a high level, each evaluation has five phases: review the workpiece samples provided, recreate the application setup, test whether to use bright field or dark field illumination, test for the exact form factor to use, and then finally the wavelength to use.

Here’s how the Application Engineer went through this process for this particular application:

Step 1: Review all samples

We go through the samples we've received from the customer and determine which are the good ones, which are the bad ones, how are they categorized, and where are the defects. The more samples we have to test, the better we can evaluate the accuracy of the system for judging good vs. bad, as well as barely good vs. barely bad.

“The more samples we have to test, the better we can evaluate the accuracy of the system for judging good vs. bad, as well as barely good vs. barely bad.”

The samples provided are decent in that there is a large variety of them. There are multiple types with foil packaging on the front AND types with matte paper. We can easily test for a light that can get a uniform image of both.

The drawback is that there aren't many defective samples. For example, some types do not have broken pills in their sample. So there is no way to confirm if the contrast is good enough for the vision system to detect a broken pill or not. 

If a certain defect is application-critical, we need samples with variations of that defect to make sure that the solution can detect it and not just assume it should work.

Step 2: Recreate the application setup

As mentioned before, the engineer switched to a 5 MP camera to get a resolution high enough to read the lot codes. The application did not have any restrictions on the working distance or light working distance, so we can use any lens that gets the necessary FOV.

If you do not have the exact lens or camera needed on hand, there are a few possible substitutions:

• If your camera is too big: modify the resolution of the camera that you need it to be. 

• If you don’t have the exact focal length lens: adjust the working distance to the right FOV, but make sure the light working distance doesn't interfere with what would be the actual working distance.

Otherwise, you need to borrow the specific lens model from the customer or the manufacturer.

Here our engineer starts with a 5 MP camera and a 35 mm lens. The FOV measures about 5 inches. The largest sample is confirmed to fit within the FOV, but there is not a lot of tolerance, so the customer will have to make sure that there is enough control on the machine to ensure the workpiece will always fall within the FOV.

Step 3: Test bright field vs. dark field

To test whether bright field or dark field works better for an application, we like to start with a white HPR2 ring light because it works at both high and low angles. (If you don't know the difference between bright field vs. dark field, watch this).

For samples with printed text on matte paper, we know right away not to use dark field because we want to read the text. However, one sample has an embossed code, which tends to need dark field illumination. This means there’s a good chance we can’t read all the lot codes in one inspection. The engineer can confirm this after determining the form factor in the next phase.

Although the ring light confirms that bright field will work to read printed codes, the samples with foil packaging do not get an even image. The contrast of the lot and date code could also be better. Knowing we need bright field but that direct illumination causes glare on the foil packaging, we move on to finding which form factor creates a more even, flat image.

Step 4: Determine form factor

From here, we circle between trying different form factors and wavelengths.

We recommend starting with your most difficult sample because if a form factor works, it likely works with the rest. The "hardest" sample in this application is a blister pack with rounded foil wrapping.

“We recommend starting with your most difficult sample because if a form factor works, it likely works with the rest.”

Our Application Engineer starts testing different form factors on the front side of the sample.

Here are the form factors we end up trying:

1. HPD2 dome light

2. LFXV flat dome light

3. LFV3 coaxial light

4. LDL2 bar light

5. HPR2 low-angle ring light

6. FPQ3 low-angle square light

The go-to light for removing 3D texture is a dome light. Although it gets a flat image of the sample, it is not a sustainable solution because the camera hole in the dome light creates a shadow in the image.

This problem can be solved with the LFXV. It acts like a dome light, but the camera looks through the light itself instead of through a hole in the light. At a 20 mm LWD, the flat dome gets an even image of our most difficult sample.

Once we have the solution for the hardest sample, we can start checking if it works for the rest. The flat dome gets good images of defects, but a different wavelength could improve the solution's overall reliability if there is damage in the same place as the dark print. (This can be confirmed in the next evaluation phase.) However, the sample with the embossed code does not get enough contrast to see it reliably, so we need to find another solution for this type of code.

Watch an engineer talk through the process of testing which form factor works on the embossed code:

Testing shows that two different stages will be needed if the customer wants to perform all the inspection tasks for all the samples. The embossed code needs a dedicated stage with an HPR2, while all other samples can be inspected with an LFXV. Trying these two form factors on the pill side, the LFXV is the solution for both types of samples.

Step 5: Determine wavelength

After determining form factor, the engineer can test if different wavelengths can increase contrast to improve the reliability of the image.

We start with testing infrared because it can remove print in some situations. But to "solve" this application, IR would have to remove the text while still keeping the lot codes visible. We find that an 850 nm light cannot do this on the relevant samples, and for some, it does not even remove the printed text. So we turn to the visible spectrum.

To test if a monochromatic wavelength can improve the contrast, we need to test different colors of the form factor we chose. If you don't have red, green, and blue versions of the form factor you need, we recommend using a white version with different filters. Fortunately, we had a full-color version of an older flat dome light in our lab.

Pro tip: Different LEDs have different intensities. Don't confuse an image that looks better because it uses the right color with an image that looks "better" because it happens to be brighter. If your imaging software tells you the grayscale value, pick a gray section in the image and get as close to 200 as possible so all images have the same intensity.

“If your imaging software tells you the grayscale value, pick a gray section in the image and get as close to 200 as possible so all images have the same intensity.”

This is the most time-consuming part of testing because we need to image every color with most, if not all, of the samples.

Overall, white light gets the best results across all the samples on both the front side and pill side. And will be the most flexible if new sample variations are added later.