Virtual system faults for training fault identifiers

1
Virtual system faults for training fault
identifiers
2003 VTB Users and Developers Conference 17-18
September 2003

• F. Ponci
• Dept. of Electrical Engineering
• University of South Carolina

2
Topics

• Virtual Testing on the VTB platform
• VTB-LabView link the VTB platform for monitoring
  and diagnostics (MD)
• Simulated faulty system for the training of a
  fault identifier
• Distributed remote monitoring and diagnostics

3
Introduction

• Problem
• The monitoring and diagnostics of complex
  systems realized by the integration of several
  sub-systems require suitable tools
• Solution
• Availability of a general purpose Virtual
  Environment that allows simulation, measurement
  and testing of single components as if they were
  already part of the whole system

4
Aspects of a GoodVirtual Environment for MD

• Incorporates other environments through
  multi-formalism and co-simulation
• Diagnostic algorithms developed within
  specialized environments (advanced signal
  processing, fuzzy, neural, neuro-fuzzy)
• Provides a high level visualization of data (2D
  and 3D)
• Interacts with the physical environment, with
  special consideration for acquisition systems
  widely used for MD
• LabView

5
VTB-LabView in Test Automation basic options

• VTB model validation against the physical system
• The input of the real system is acquired with
  LabView
• The acquired data are fed to the VTB model
  through the VTB-LabView interface
• The comparison between real and simulated outputs
  is the feedback for model tuning
• VTB model simulation for detection of abnormal
  behavior of the physical system
• The input of the real system is acquired with
  LabView
• The acquired data are fed to the VTB model
  through the VTB-LabView interface
• The comparison between real and simulated outputs
  is used to identify abnormal behavior of the real
  system

6
VTB-LabView in Test Automation advanced options

• The VTB model validation and the anomalies
  detection as a design approach
• The mismatch between expected and simulated data
  as feedback for design
• Remote testing
• Simulation platform location far from real
  equipment location

7
Model Validation I
Simulated and measured outputs
Physical System Active filter Input from power
supply
The input of the simulated system is the input of
the physical system
System Input
LabView Acquisition Platform Acquired data input
and output of the filter
8
Model Validation I Remote Operation
Physical System Active filter Input from power
supply
Simulated and measured outputs
The input of the simulated system is the input of
the physical system
LabView Acquisition Platform Acquired data input
and output of the filter
Internet/Intranet (TCP/IP protocol)
System Input
USC Swearingen 3rd floor
USC Swearingen 2nd floor
9
Model Validation II
Simulated and measured outputs (superimposed)
Physical System 1-phase transformer (no
load) Input from the mains
The input of the simulated system is the input of
the physical system
System Input
LabView Acquisition Platform Acquired data input
and output of the transformer
10
Model Validation II Remote Operation
PoliMi
Simulated and measured outputs (superimposed)
Physical System 1-phase transformer (no
load) Input from the mains
The input of the simulated system is the input of
the physical system
System Input
Internet/Intranet (TCP/IP protocol)
USC
LabView Acquisition Platform Acquired data input
and output of the transformer
11
VTB-LabView in MD Test Automation

• Testing of the acquisition and diagnostic system
  on the VTB model
• Acquisition and diagnostic system implemented in
  LabView
• System simulation running in VTB
• LabView acquires the simulated data through the
  VTB-LabView interface
• VTB model simulation for data collection
• System simulation running in VTB
• Simulated data are used for training of fault
  identification system

12
Acquired and simulated data integration

• Case study AC motor drive
• Diagnostic approach based on a trained system
  (neuro-fuzzy)
• The training requires an extensive set of data
  collected under normal and faulty conditions
• The capability to integrate data collected from
  real measurements and from simulation results in
• Cost and risk reduction
• Data collected in a variety of operating
  conditions
• Data collected with the target subsystem
  interacting with the rest of the system

13
AC Drive Monitoring and Diagnostics
Physical system
Wavelet processing
Data acquisition
Neuro-fuzzy system
Fault
Non fault
14
AC Drive Virtual Monitoring and Diagnostics
VTB simulated system
Wavelet processing
Data acquisition
Neuro-fuzzy system
Fault
Non fault
15
Virtual MonitoringFeatures

• Design and validation of the acquisition system
• Sensor distribution
• Design and validation of the monitoring system
• Visualization of measured data
• Training of the operators
• Design and validation of the diagnostic algorithm

16
Diagnostic system training the physical AC drive
AC drive

• Limits on the variety of operating conditions
• No interaction with the rest of the system

Neuro-fuzzy system in training
Fault
Check and update
Non fault
17
Diagnostic system training on the simulated AC
drive

• Virtually no limits on the variety of operating
  conditions
• Easy test automatization
• Interaction with the rest of the system

Training of the Neuro-fuzzy system
Fault
Check and update
Non fault
18
Virtual and physical measurement integration
Physical system
Fault
Check and update
Non fault
Simulated system
19
The experimental setup

• AC motor drive with wounded rotor
• The line voltage and current acquisition
  Analog-to-Digital conversion board (ADC), 8 input
  channels with simultaneous sampling up to 500 kHz
  sampling rate on a single channel, ?10V range,
  12-bit resolution and offset, gain and
  non-linearity error in the range ?½ LSB.
• Voltage and current transducers have been
  specially realized in order to ensure an adequate
  insulation level between channels and between the
  supply and measuring devices over a wide band.
• According to the input signal range (230 V rms
  for the voltages and up to 20 A rms for the
  currents) a non-inductive, resistive voltage
  divider followed by an isolation amplifier was
  used as voltage transducer, and a closed-loop
  Hall effect transducers was used as current
  transducer
• Voltage transducers relative standard
  uncertainty on the gain of 0.02
• Current transducers relative standard
  uncertainty on the gain of 0.03 up to 5 kHz The
  time delay at 50 Hz between the voltage and
  current channels is 20 ?s and constant up to 5
  kHz 4.
• Voltage and current sampling rate 12.8 kHz, (256
  sampled/period for a fundamental frequency of 50
  Hz)

20
Operating conditions of the drive during
acquisitions
Normal operating conditions
Faulty operating conditions
Open rotor phase
Open stator phase
No-load
Nominal load
No-load
Nominal load
Nominal load
f1
f1
f1
f2
f2
f2



frated
frated
frated
21
Integration of acquired data

• The set of data experimentally acquired were
  integrated with data obtained from simulation
• Simulation tool Virtual Test Bed (VTB)
• The model of the system within the simulation
  environment is a composition of validated VTB
  native models.
• The set of integrated data were used for the
  training of the diagnostic system
• The fault identification capabilities of the
  trained system were tested on real data

22
Diagnostic Index

• The wavelet-based index has been applied with
  success for diagnostic purpose
• The wavelet-based index values resulting from the
  training can identify the faulty operating
  conditions

23
Test of Index Validity

• Real data are used for index validation
• Data are used for index validation that where not
  used for training
• Faulty or non faulty conditions are identified in
  three different cases
• Non-faulty (operating condition 1)
• Non-faulty (operating condition 2)
• Faulty (operating condition 3)

24
Distributed Diagnostics a USC experience
25
Agent Location
PoliMi Milan-Italy
USC Columbia SC-USA
System Manager Agent PC 131.175.14.8
Measurement section and drive control
Wavelet Unit Agent PC 129.252.22.202
Measurement section Data Acquisition and
Monitoring Agent
Fuzzy Unit Agent PC 129.252.22.215
Internet/Intranet (TCP/IP protocol)
Simulation section VTB Agent
26
Roles and Interactions
27
Screenshot of System Manager Agent
28
Future Directions next steps

• Identification of incipient faults
• Implementation of VTB models for devices with
  incipient faults
• Implementation of algorithms for the recognition
  of incipient faults
• Training of algorithms for the recognition of
  incipient faults with simulated data
• Testing of the diagnostic system on real data

29
Future Directions

• Exploitation of remote and distributed monitoring
  and diagnostic systems features
• On-line remote validation of the faulty model
• On-line remote testing and validation of a remote
  monitoring and diagnostic system on simulated
  faulty systems

30
Conclusions

• The importance of virtual environments as support
  to monitoring and diagnostic system design has
  been introduced
• A specific solution based on the Virtual Test Bed
  has been presented
• Examples of application have been discussed