1 / 14

2018 Summer Intern

2018 Summer Intern. Hang Xu. Outline. 1. Real-time system research 2. Asset watchdog project for ABB. Motivation. Real-Time performance suitability research for machine learning libraries on embedded system. Data analytics from cloud to edge

jclift
Télécharger la présentation

2018 Summer Intern

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 2018 Summer Intern Hang Xu

  2. Outline 1. Real-time system research 2. Asset watchdog project for ABB

  3. Motivation • Real-Time performance suitability research for machine learning libraries on embedded system. • Data analytics from cloud to edge • Embedded system(micro-computer, micro-controller) has relatively fewer results on ML direction.

  4. Real-time System Research • Trained model based on practical solar power control system data • Compared between the Keras library and the Caffe library • Modified the original Caffe library • Conducted experiments

  5. Experiments - Model converted using Microsoft model conversion tool - On Raspberry Pi 3B - Patched real-time preemptive linux kernel on Pi - 10 times of experiments each of 10,000 execution time samples

  6. Comparison Figures

  7. Max and Min Execution Time

  8. Average & Std. Deviation of Execution Time

  9. Conclusion • The original Caffe library and the keras library are not suitable for ML problems of high real-time predictability requirement on an embedded system. • The modified Caffe without redundant part of third party library calls and multi-core functionalities is suitable for such ML problems on an embedded system.

  10. Asset Watchdog Project A cyber security system is designed for embedded platforms. The cyber intrusion is detected using execution time information. A demo is presented to show the prototype of the entire system.

  11. Asset Watchdog System Diagram for Demo Monitor telemetry of physical data After discussion with the electrical and control engineers, we decided to use this system diagram for the execution time based anomaly detection demo. Software Simulated Cyber Attacker Execution time monitoring and anomaly detection Control command Ethernet Port Matlab/ Simulink deploy C code laptop to develop controller code in Texas Instrument (TI) Code Composer Studio(CCS) TI DSP controller board

  12. Asset Watchdog Simplified System Diagram A complete HIL(hardware in the loop) simulation involves the development and deployment of physical models onto real time hardware, which is not the main objective for our demo. For the achievability during this summer, we plan to use simplified control target problem: LED control. Software Simulated Cyber Attacker Execution time monitoring and anomaly detection Ethernet Port deploy C code Control plants (Could be simple LED control for the initial step of demo) laptop to develop controller code in Texas Instrument (TI) Code Composer Studio(CCS) TI DSP controller board

  13. Asset Watchdog Demo Local Area Network Setting up Router Intrusion Detector Attacker Watch demo DSP control board Development Host

  14. Thank You!

More Related