2. Background
• CyberManufacturing System: Future vision of physical components fully
integrated with computational processes in a connected environment.
• Cyber Security: Information technology security that focuses on protecting
computers, networks, programs and data from unintended or unauthorized access,
change or destruction.
• Enables intelligent behaviors
3. Objective
• To address security issues in CyberManufacturing Systems
• To explore the application of machine learning in detecting malicious attacks in
CMS
5. Why 3D printers?
•Additive manufacturing are connected to
the computer and it is vunerable to cyber
attack.
Aerospace
Automotive
Biotechnology
11. 3D Printed Mounts and Setup (2nd Design)
Images
Captured
Captured over 500 images Below 70% Accuracy
12. 3D Printed Mounts and Setup (3rd Design)
Images
Captured
Captured over 500 images Above 90% Accuracy
13. Second Half of Summer (Real-Time Detection)
Previous Experiment
Image classification detects
malicious attacks on the
infill in 3D printing
technology
New Experiment
Internet of Things
Real-Time
Detection
14. Hardware & Software
• Raspberry Pi B+
• 5MP Raspberry Pi camera
• Flash drive and SD card
• Wi-Fi USB Dongle
• BitTorrent Sync
• Dropbox
16. Overview of System
Machine Learning &
Signal Malicious Attack
Sync
(Computer to Cloud)
Sync
(Device to Computer)
Camera Capture
Images
User(s)
3-D Print
Object
18. Real-Time Detection
• Once detected, the R program sends a message via text/email to the person after few
minutes
19. Summary
• Designed various camera mounts
• Collected real images using the hardware for machine learning
• Internet of things for real-time detection
• Monitoring and detection system
20. Conclusion
• Vision-based system and image classification using machine learning techniques can
detect malicious attacks in 3D printing.
• Require large data set in order to train the machine learning to accurately detect only
malicious attacks.
21. Thank you! Q&A
Detecting Malicious Defects in CyberManufacturing Systems Using Machine Learning
by
PhD Student Mingtao Wu
Mechanical and Aerospace Engineering
College of Engineering and Computer Science
Syracuse University
miwu@syr.edu
by
Undergraduate Heguang Zhou
Mechanical and Aerospace Engineering
College of Engineering and Computer Science
Syracuse University
hzhou11@syr.edu
by
PhD Student ZhengYi Song
Mechanical and Aerospace Engineering
College of Engineering and Computer Science
Syracuse University
zsong04@syr.edu
by
Undergraduate Bruno C Silva
Industrial and Manufacturing Engineering
College of Administration and Engineering
FAE University
brunocansi@gmail.com
By
Undergraduate Lucas Lin
Mechanical and Aerospace Engineering
College of Engineering and Computer Science
Syracuse University
llin104@syr.edu
by
Undergraduate Jackie Cheung
Mechanical and Aerospace Engineering
College of Engineering and Computer Science
Syracuse University
jcheun03@syr.edu
Professor Young B. Moon
Mechanical and Aerospace Engineering
College of Engineering and Computer Science
Syracuse University
ybmoon@syr.edu
Editor's Notes
Here is some background info about our research. CMS is a future vision where all machines are connected to the network inside a factory with the use of sensors and other technologies. The system uses advanced algorithms to allow the physical components communicate with each other.Increase efficiency in production. Since these physical components are connected to the network, it is important to ensure its security. The reason why the security is imporant in a CMS is because for example, in a airplane industry, a part of the airplane is being created using additive manufacturing and has been changed a little. If the change can’t be detected by the employees and is installed to an airplane, it could cause safety issues and can lose loyalty from the customers
As of now Cyber Manufacturing systems do not exist. But in order for U.S. to adopt to this new form of manufacturing system in the future, cyber secuirty is a huge issue that needs to be addressed.
With my team’s research, we are focusing on detecting malicious defects on manufactured parts with the use of image classifcation and acoustic sound classification. Security is large, but we focus on machine laerning
These are just 2 examples worked on. Explain wat simulated image and real images are. (machine learning using images) and using acoustic signals) images plural to all
3D printing technology will be a big thing in the future. And since the things being printed are used by people, we need to ensure they are in the best quality possible during production. We have to avoid defects in the manufactured part that could be caused by hackers or unauthorized changes. (add a slide about the process of machining learning – good part, natural and malicious attack
So how does machine learning work? First of all, these are simulated images of the infills of a 3d printed cube. According to the setting, the infills can be different and we chose honeycomb shaped infills for the machine learning process. And machine learning works by using the different types of images as training data. Then when we show it new images, it can identify the part to be good, naturally defected, or macliciously attacked.
Add arrows to go back to previous step
everyday objects have network connectivity, allowing them to send and receive data.