Event
Texas Instruments Tech Talk
Monday, September 16, 2019
5:00 p.m.-6:00 p.m.
2460 AV Williams
Kara Stamets
301 405 4471
stametsk@umd.edu
Texas Instruments Tech Talk
Title: Application of Recurrent Neural Networks (RNN’s) to Anomaly Detection in Predictive Maintenance on the Edge
Presenter: Dr. Hongmei Gou, Texas Instruments, Inc., Germantown, MD
Join Dr. Hongmei Gou on September 16th for an exciting talk on Recurrent Neural Networks (RNN's) and anomaly detection in predictive maintenance on the edge. Following the talk, Dr. Gou will hold a Q&A and she will also discuss career/employment opportunities at TI. Pizza and drinks will be served.
Abstract: Promise of Industry 4.0 can be fulfilled if subsystems reliability is kept at maximum level. After decades of reactive and condition-based maintenance, predictive maintenance (PdM) has been introduced, which applies machine learning and analytics on the real-time operational data collected from end sensor nodes to draw meaningful insights that can more accurately predict machine failures.
PdM is comprised of three steps: sense, compute and act. Data is collected from sensors that are already available in machines, by adding new sensors, or by using control inputs. Depending upon the machine types and the required failure analysis, different sensor signals – such as temperature, sound, magnetic field, current, voltage, ultrasonic, vibration – are analyzed to predict the failure. The predicted information from sensor data analysis is used to generate an event, work order and notification.
TI’s Sitara based embedded platforms can do real-time acquisition and processing of multiple sensory inputs important for PdM tasks. PdM decision making is facilitated by advancement of Deep Learning technologies. In this talk, we will present a predictive maintenance demo which leverages Recurrent Neural Network (RNN) for anomaly detection over motor drive control. The demo builds upon TI’s Processor SDK Linux software and runs on TI’s Sitara devices for real-time inference on the edge. We will discuss the system model of using RNN for anomaly detection, the workflow of developing the demo, and the benchmarking results.
Presenter Bio: Hongmei Gou received the Ph.D. degree in electrical engineering from University of Maryland, College Park, in 2007. She received the Graduate School Fellowship from University of Maryland for the years 2002 to 2004. Since 2007, she has been a software engineer with Texas Instruments. Previously, she was a research intern with the R&D Department, Pitney Bowes, Shelton, CT, in 2005 and 2006.
At Texas Instruments, Dr. Gou is a recognized technical leader on developing multimedia, machine vision, robotics, and Linux user space industrial protocols on TI's embedded devices. Her recent roles focus on deep learning, analytics, and predictive maintenance on TI's Sitara processors.
