2. 02
01
04
Problems & Solutions
Requirements
What the main deliverables
of this project are.
The main problems we
faced, and how we solved
them.
Mechanical Prototype
We show a small example of
the main project as a
prototype.
03
Next step
Table of contents
4. Requirements
of the Project!
We were asked to make part of the
production line of marble factory to preform
the following:
● Scan the marble plate
● Detects if there is any kind of fractures
in the plate
● Measure the length, width and height
of the plate
● Store all these data in a database to
make it easy to access.
All these points summarizes what we need
to do in our project.
5. The scanning algorithm
Mechanical Prototyping
We developed a python script that scan
the plate and its output is a jpeg image
that will be the input to other processes.
We managed to design a rough image of
what our project will look like when it’s done
and how we will manufacture the prototype.
What we’ve done till now.
Handling the output
We also developed a python script that
takes the output of the first part and sort
them inside excel sheet which will be used
further on.
7. The scanning element
● The problem we had first is the
scanning element, the
technique of scanning.
● The size of the scanning
element was a big deal.
● The price was a major
thing too.
8. The scanning element
We solved these problems with a
brilliant solution which is called
‘Line scan camera’
But we had a small problem
9.
10. The scanning element
We had this Idea of using an area camera and modify the output slightly to have the needed outcome
20. This product uses google’s TPU (tensor
processing unit) and This option seems
like it might have more machine learning
power but comes with weaker overall
specs than a raspberry pi 4b also TPU
can only use tensorflow lite frame work
while other boards are not limited and
can use any machine learning framework
1-Google coral board 2-beaglebone AI
We considered this option but while it has some
ai specific hardware it is priced similar to jetson
Nano and has a dual core and only 1 gb ram
2- other options
21. This board has intel CPU that actually runs x86 not
arm. And can use windows it also has very good
performance for the price at least in general tasks
however we could not find any benchmarks for
machine learning for it. since it relays on intel
integrated graphics for its gpu and there is no
documentation for ts use in machine learning
applications we decided not to choose it
3- Lattepanda 3 Delta
2- other options
22. 2- sourcing boards
We have chosen jetson family of
developer kits as the best choice for our
application but now we must choose
right kit for our workload out of the
options we have
There persists an issue in pricing and
availability of all boards we looked at
and all of them have prices heavily
increased from msrp.
For example All retail websites on
nvidia's site are out of stock and
amazon both here and in italy have
much higher prices for all models
1-Choosing right board for
our workload
3-next steps