Image Processing and Data Analysis of X-Ray CT Scan of 3D-Printed Component

Bethany Krull Author

04/03/2021 Added

32 Plays

Description

Presentation of my UCARE work for the 2020-2021 academic year.

Searchable Transcript

Search:

[00:00:01.050]Hello, my name is Bethany Kroll.
[00:00:02.930]And today I'll be talking with you about my UCare prep project,
[00:00:07.950]titled image processing and data analytics.
[00:00:10.950]X-rays CT scans from 3d printed components. Um,
[00:00:14.700]I did this with Dr. Rao.
[00:00:17.910]It's just an outline of what we'll be talking about today.
[00:00:21.180]And my project background within 3d printed
[00:00:25.800]parts. There are a lot of different types of errors that can occur,
[00:00:29.250]but they can be difficult to detect because they are
[00:00:33.510]internal within the part that you can't really look at,
[00:00:37.470]unless you've cut the part open. Um,
[00:00:40.020]so CT scans are a great way to get a sneak peek on the internal
[00:00:44.940]structure of a part. So basically if you look at figure two down here,
[00:00:50.220]um, if you think of this as the component that you're printing,
[00:00:53.790]and this has the Z axis up and down, um,
[00:00:56.250]CT scans are x-rays that are taken at different Z
[00:01:00.930]depths along the component. And, um, they you're just like, Oh,
[00:01:04.980]a little x-ray of different cross sections as you go down.
[00:01:08.820]So it'd be like one here.
[00:01:10.680]And then one that goes around this box and then up higher as you go through.
[00:01:17.010]So specifically I looked into porosity detection with a
[00:01:21.990]laser powder bed fusion process.
[00:01:24.510]So that's when you have metal powder it's laid out and then a laser goes through
[00:01:28.740]and melts the powder into a solid, um,
[00:01:32.310]and so porosity occurs when,
[00:01:37.110]um, some of the powder doesn't get melted all the way,
[00:01:39.900]or there are gas bubbles within the,
[00:01:42.650]the melt process that caused these little gaps in the
[00:01:47.790]solid built. So this is actually the Z-axis right here. You can seek,
[00:01:52.370]see faintly some lines that's the layer filled. Um,
[00:01:57.060]and these are holes that have occurred within the part due to
[00:02:01.860]prosody. And this can have a lot of impact on the,
[00:02:06.780]uh, part at the end, if it is a high stress component,
[00:02:10.000]you need to have low porosity and, you know,
[00:02:12.990]some specific situations require specific amounts of properties.
[00:02:17.310]So it's really important to understand what's going on inside the part.
[00:02:22.770]So my method was basically identifying
[00:02:27.330]individual pores on these x-rays CT scans.
[00:02:31.200]And then once these individual pores have been identified,
[00:02:35.100]I can characterize them looking at their shape, their size,
[00:02:38.730]their proximity to other pores,
[00:02:40.620]and to the general scope of the part.
[00:02:45.030]And this plays a large role in the effect that it
[00:02:49.770]has on the parts performance.
[00:02:50.970]And then also it can close into the causes of these.
[00:02:55.860]So I'm just going to walk you through, this is a snippet of my logic. Um,
[00:03:00.700]right here, we actually see the original CT scan of the part.
[00:03:05.950]So this is what we get the raw data.
[00:03:08.050]And here I'm just basically taking this gray scale
[00:03:12.730]image and making the contrast more severe.
[00:03:16.270]So the darks are darker and the lights are lighter.
[00:03:18.370]It's still a gray scale image,
[00:03:19.570]but it's a lot more potent if you could say. And here,
[00:03:24.670]the reason I did that well,
[00:03:26.590]so that I could use this Piner icing function in MATLAB to turn it this
[00:03:31.480]gray scale image into a black and white image.
[00:03:34.240]So if this wasn't so severely, dark and severely light,
[00:03:39.220]we wouldn't be able to read very distinctly these intentional holes,
[00:03:44.170]the edges. And then also, if you look,
[00:03:46.840]you can see pretty clearly these are fours, these defects right here,
[00:03:52.150]you can see those are what we're trying to detect via the
[00:03:56.800]computer. It's pretty easy to see with your eyes,
[00:03:58.630]but we're trying to get the computer to recognize that. Um,
[00:04:02.080]and then this is just a simple flip. Um, so read it the other way.
[00:04:06.250]Cause the function I use later reads to them in the inverted.
[00:04:11.530]So this right here is actually, um, a large part of,
[00:04:16.360]uh, identifying pores.
[00:04:17.830]So this region properties function allows me to input an
[00:04:22.750]image like the one I just showed you how I prepared it and then output basically
[00:04:27.520]any type of data that I would want about, um, the,
[00:04:32.070]the regions inside of that image. So region properties is identifying the,
[00:04:37.410]the, like the shapes inside of you can see,
[00:04:40.810]like this would be a region. This is a region,
[00:04:44.290]all like the different identifiable shapes.
[00:04:46.660]So here's all the different types of things we can ask it for.
[00:04:50.560]But if you look through it more clearly, you can see it.
[00:04:53.920]There are issues with this unfiltered data. So up here,
[00:04:57.040]you can see this really large area with a huge, um,
[00:05:01.990]access length and like weird circularity. That's actually the background.
[00:05:07.060]So we cut that out and we're also cut out these very
[00:05:11.380]specific holes that could be pores to the computer,
[00:05:15.310]but we know that they're intentional parts of the design. So we cut those out.
[00:05:20.440]And then if you look here,
[00:05:22.120]all of these things with very small area, um,
[00:05:26.980]it's only a small screenshot of the large table,
[00:05:30.490]but anything with like these ones,
[00:05:32.500]these twos and threes and filled areas are actually most likely.
[00:05:36.730]If we look deeper into this image,
[00:05:39.250]just these pieces of noise in the skin and not actually for us,
[00:05:43.270]the computer can recognize that,
[00:05:44.950]but we can go through and filter out any regions that are way too small to
[00:05:49.510]possibly be pores.
[00:05:51.490]So once we've filtered out to things that we know as, um,
[00:05:56.530]programmers, isn't a poor background, these intentional holes,
[00:06:01.100]all this noise,
[00:06:02.210]then we can help the computer identify what is a port.
[00:06:06.170]So if we look here's one here, that's pretty easy to see.
[00:06:10.880]There's a kind of a cluster right here and another cluster down here. Um,
[00:06:16.400]and, and it doesn't read all of these.
[00:06:19.810]And this background as well.
[00:06:22.960]So, um, moving forward,
[00:06:25.840]refining for identification is important. Um,
[00:06:29.740]this isn't a perfect process and more should be done to
[00:06:34.030]verify that what we're seeing are fours is actually a poor. Um,
[00:06:37.720]and then add, identify occasion that extends beyond a single slice.
[00:06:43.150]So this part has, I think 1600 slices in it.
[00:06:48.220]Um, 16,000, I'm sorry. And, um, a poor can be,
[00:06:53.140]you know, the, the,
[00:06:53.950]the Z length of a poor can take up more than one slot,
[00:06:58.660]one more than one slice.
[00:07:00.130]So if you have multiple consecutive slices,
[00:07:04.450]they could be seeing the poor from multiple, uh,
[00:07:09.190]views along the Z axis.
[00:07:10.480]And we don't want to count one poor 10 times because 10
[00:07:15.010]slices, uh, intercepted that poor along the Z axis. So,
[00:07:19.930]uh, increasing our awareness of what identified the
[00:07:24.880]support and what's a repeat for, uh, is important.
[00:07:28.210]And then going back into that region properties function,
[00:07:31.240]I talked about earlier,
[00:07:32.380]characterizing these pores and determining their cause and their impact and all
[00:07:37.360]of that.
[00:07:38.140]So thank you for listening to my talk on my Euchre project.
[00:07:43.210]I would just like to acknowledge Dr. Rao,
[00:07:46.570]my advisor for all of his health and also plant veterans, a graduate student,
[00:07:51.280]also under Dr. Rob, who has been immensely helpful. And in this presentation,
[00:07:56.500]thank you for listening.

The screen size you are trying to search captions on is too small!

You can always jump over to MediaHub and check it out there.

Comments

0 Comments

Image Processing and Data Analysis of X-Ray CT Scan of 3D-Printed Component

Description

Searchable Transcript

Comments icon comment

Related Channels

Comments