Data Analysis, Estimation, and Fault detection of Large-Scale - - PowerPoint PPT Presentation

▶
data analysis estimation and fault detection of large
SMART_READER_LITE
LIVE PREVIEW

Data Analysis, Estimation, and Fault detection of Large-Scale - - PowerPoint PPT Presentation

Dec 17, 2022 239 likes •573 views

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Networks Presented by: Parsa Yousefi Supervisors: Dr. M. Jamshidi, Dr. P. Benavidez June 23 rd , 2017 The University of Texas at San

slide-1
SLIDE 1

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Networks

Presented by: Parsa Yousefi Supervisors: Dr. M. Jamshidi, Dr. P. Benavidez June 23rd, 2017

slide-2
SLIDE 2

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Data Analytics
  • Introduction
  • Clustering
  • Neural Networks
  • Long Short-Term Memory
  • Data prediction
  • Latency
  • Reconstructing Data Using LSTM
  • Fault Detection
  • Training Initial Model Using LSTM
  • Future Works
  • Acknowledgments
  • References

Outline

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

2

slide-3
SLIDE 3

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Introduction
  • Definition: Data analytics refers to qualitative and

quantitative techniques and processes used to enhance productivity and business gain. Data is extracted and categorized to identify and analyze behavioral data and patterns, and techniques vary according to

  • rganizational requirements.
  • Big Data

Data Analytics

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

3

https://www.neuralt.com/uploads/assets/BigData%20.png

slide-4
SLIDE 4

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • A Summary of Data Science Process:
  • Data Collection
  • Processing
  • Cleaning Data
  • Product Data
  • Communication

Data Analytics

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

4

https://upload.wikimedia.org/wikipedia/commons/b/ba/Data_visualization_process_v1.png

slide-5
SLIDE 5

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Processing Methods
  • Clustering
  • Neural Networks

Data Analytics

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

5

http://i0.wp.com/planningtank.com/wp-content/uploads/2014/01/Distribution-Process.jpg

slide-6
SLIDE 6

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Data Clustering
  • Definition
  • Centroid-based Clustering
  • Density-based Clustering

Processing Methods

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

6

https://upload.wikimedia.org/wikipedia/commons/c/c8/Cluster-2.svg

slide-7
SLIDE 7

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Centroid-based Clustering
  • Initiating centroids
  • Finding nearest members to

centroids

  • Calculating New Centroids
  • Repeating the method until

convergence

  • Advantage
  • Convergence Speed
  • Disadvantage
  • Number of Centroids as an

input

Data Clustering

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

7

http://www.mdpi.com/sensors/sensors-15-29056/article_deploy/html/images/sensors-15-29056-g004-1024.png

slide-8
SLIDE 8

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Density-based Clustering
  • Defining clusters as areas of higher

density

  • Advantage
  • No need to set the number of clusters as

input

  • Disadvantage
  • Not applicable for datasets with large

differences

Data Clustering

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

8

http://web.cse.ohio-state.edu/~belkin.8/clustering_workshop_SDM2010/clusters.jpg https://upload.wikimedia.org/wikipedia/commons/thumb/a/af/ DBSCAN-Illustration.svg/400px-DBSCAN-Illustration.svg.png

slide-9
SLIDE 9

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Neural Networks
  • What is a Neural Network?
  • Definition of Dr. Hecht-

Nielsen:

  • “A computing system made

up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs.”

  • The Basics
  • Applications
  • Long Short-Term Memory

Processing Methods

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

9

http://www.ndt.net/article/v05n07/spanner2/fig2.gif

slide-10
SLIDE 10

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Long Short-Term Memory
  • A type of Recurrent Neural Networks
  • Introduced by S. Hochreiter and J.

Schmidhuber in 1997

  • The Core Idea
  • Why LSTM?

Neural Networks

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

10

https://tc.sinaimg.cn/maxwidth.800/tc.service.weibo.com/cdn_images_1_medium_com/58ad765e09ea cb5116c9dfc5897c7296.png

slide-11
SLIDE 11

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Thrust 1, Sub-thrust 1-1 of TECHLAV
  • Problem definition?
  • The latency in sending and receiving data by UAVs and

UGVs in the presence of heavy computation

  • The latency is an effect of limitation in communication

speed

  • Solution?
  • Prediction of future data based on the current data with

high accuracy and reconstructing it

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

11

slide-12
SLIDE 12

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

12

Long short-term memory Latency (Delay) Cloud Processor Unit

Feeding Data to UAVs Feeding Data to UGVs

Computation

slide-13
SLIDE 13

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Our Methodology
  • Receiving data from Reference [1], [2], and Creating

Dataset

  • 12,000 samples of angular error
  • Simulating Latency
  • Use 70% of Dataset as the current data
  • Feed the current data to the Neural Network (LSTM) for

Training

  • Finding the pattern of data
  • Forecasting next 30% of dataset by NN
  • Reconstructing Data by Adding 70% of the Original Data

and the 30% forecasted one

  • Evaluating the predicted data comparing with original

dataset

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

13

slide-14
SLIDE 14

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • The Objective

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

14

Latency

70% of Original Dataset Time Series Forecasting Predictive Model Forecasted 30% Original Dataset Original Dataset Reconstructed Data Reconstructed Data Reconstruction Method

slide-15
SLIDE 15

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • The Structure of LSTM
  • One input layer
  • One hidden layer with

four LSTM neurons

  • Input gate
  • Output gate
  • Forget gate
  • Current Condition of

the network

  • One output layer
  • Sigmoid Function

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

15

sig c(t) sig i(t)

  • (t)

x(t) h(t) f(t)

Input Gate Output Gate Forget Gate

slide-16
SLIDE 16

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • The Structure of LSTM
  • Sigmoid Function: Used as an activation function for all

LSTM Blocks

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

16

slide-17
SLIDE 17

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

17

  • Results
  • Root mean

square error for train: 0.01856

  • Root mean

square error for test: 0.02324

  • 98.15% accuracy

in training

  • 97.68% accuracy

in testing

slide-18
SLIDE 18

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Results

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

18

slide-19
SLIDE 19

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Results

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

19

slide-20
SLIDE 20

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Results

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

20

slide-21
SLIDE 21

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Results

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

21

slide-22
SLIDE 22

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Challenges:
  • Despite the regression forecasting, time series prediction can be

more challenging, due to the fact that sequence dependence increases the complexity of the problem.

  • Our approach should be able to handle long sequence dependencies.
  • Gaps:
  • When the available data is only a small portion of total data, it is

almost impossible to achieve a proper estimation of the lost data.

  • Having a dataset including noise

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

22

slide-23
SLIDE 23

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Future Works
  • Using LSTM for a noisy dataset and comparing the

results

  • Using bi-directional recurrent neural networks in case of

irregular dispersed data loss for using both past and future samples

  • Using

Centroid-based and Density-based clustering methods for classification the dataset

Data Prediction Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

23

slide-24
SLIDE 24

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Thrust 2, Sub-thrust 2-1 of TECHLAV
  • Problem definition?
  • Failure in sensors, actuators, and components of a robot
  • Propagating wrong data to the computation unit
  • Failure in completing tasks
  • Solution?
  • Using Another agent for observing fault
  • Using neural networks for forecasting fault in the future

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

24

slide-25
SLIDE 25

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Our Methodology
  • Using two “Kobuki TurtleBot 2”, the first as the remote
  • bserver “Helper”, and the second as the “Faulty”
  • Inducing mechanical fault on the odometry sensors of

Faulty by using electrical tape to modify the friction of the wheels

  • Measuring the sensors of the Helper and the Faulty
  • 19,000 testing datapoints
  • Training the model using LSTM
  • Validating the model with high accuracy and prediction
  • f future fault

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

25

slide-26
SLIDE 26

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Type of faults
  • No Fault
  • Right wheel fault
  • Left wheel fault
  • Fault in both wheels

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

26

slide-27
SLIDE 27

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Results
  • Training the initial fault detection model with LSTM and

Testing it

  • Using 6 inputs from the odometry sensors (x and y position,
  • rientation in quaternions)
  • 4 outputs (no fault, left fault, right fault, both faults)
  • The testing accuracy of 88% on ~19,000 datapoints

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

27

slide-28
SLIDE 28

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Results

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

28

slide-29
SLIDE 29

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Challenges
  • Sensing: For a remote observer robot “Helper” to be able

to detect a fault in the “Faulty” the sensors on the robot need to be able to capture the fault in sufficient detail. Cameras for instance need to be capable of recording at a rate high enough to capture the fault over more than one image frame.

  • Computation: The fault diagnosis algorithm should be

able to run near real-time in order to make it useful in live detection of faults in the system. Training of the models should be completed with sufficient amounts of training data to capture types and levels of faults.

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

29

slide-30
SLIDE 30

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • Future Works
  • Integrating JSON, the extended version of AgentSpeak

into our ROS system through the use of ROSJAVA and RSON, the ROS interface to JSON

  • Using camera on Helper for processing the images

captured from Faulty to detect the faults

Fault Detection Using LSTM

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

30

slide-31
SLIDE 31

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • These researches are supported by Air Force

Research Laboratory and OSD for sponsoring this research under agreement number FA8750-15-2- 0116.

  • The Thrust 1, Sub-thrust 1-1 was done with

collaboration with Nima Ebadi, Ph.D. student of ECE Department at UTSA

  • The Thrust 2, Sub-thrust 2-1 was done by Jonathan

Lwowski and Shubham Sarpal, Ph.D. and M.Sc. students of ECE Department at UTSA

Acknowledgments

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

31

slide-32
SLIDE 32

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

  • N. Gamez, P. Kolar and M. Jamshidi, “Impact of Time Delays on Networked Control of

Autonomous Systems," a chapter in “Beyond Traditional Probabilistic Data Processing Techniques: Interval and Fuzzy Logic Methods and Their Applications}", Springer-Verlag, Heidelberg, German, to appear in 2017.

  • N. Gamez, “Modeling, Simulation, and Design of a Time-Delayed Multi-Agent System of

Autonomous Vehicles, Masters Thesis, University of Texas at San Antonio, 2017.

  • J. Schmidhuber, D. Wierstra, and F. J. Gomez Evolino, “Hybrid Neuroevolution and

Optimal Linear Search for Sequence Learning”, Proceedings of the 19 th International Joint Conference on Artificial Intelligence (IJCAI), Edinburgh, pp. 853–858, 2005.

  • K. Greff, R. Srivastava, J. Koutník, B. Steunebrink and J. Schmidhuber, “LSTM: A Search

Space Odyssey”, Cornell University Library, March 2015.

  • L. N. Smith, “Best Practices for Applying Deep Learning to Novel Applications”, Navy

Center for Applied Research in Artificial Intelligence, 2017.

  • https://www.sas.com/en_us/insights/big-data/what-is-big-data.html
  • http://pages.cs.wisc.edu/~bolo/shipyard/neural/local.html
  • https://blog.dominodatalab.com/topology-and-density-based-clustering/
  • http://colah.github.io/posts/2015-08-Understanding-LSTMs/
  • https://www-users.cs.umn.edu/~kumar/dmbook/ch8.pdf

References

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

32

slide-33
SLIDE 33

The University of Texas at San Antonio – Department of Electrical and Computer Engineering

Thanks for your attention!

Data Analysis, Estimation, and Fault detection of Large-Scale Autonomous System of Vehicles Using Neural Network

33