MultipleFault Health Detection Algorithms

1
Multiple-Fault Health Detection Algorithms

• John Cook
• 14 April 2005

2
Problem

• When multiple sensors in an ISHM system indicate
  an error, there may be a root cause for this
  problem.
• This root cause can be isolated by constructing a
  rule set or database to examine the sensor data
  at a systemic level.

3
Solution Requirements

• Algorithm needs to be located above the sensor
  level
• Algorithm should be modified whenever a new
  problem is identified
• Algorithm should have access to good sensor data

4
Sensor Issues

• number
• location/placement
• dependencies
• derived parameters

5
Types of Errors

• Localized
• Upstream
• Downstream
• Unknown

6
Localized Errors

• Occur when a specific component of the
  application breaks
• All sensors on that part of the application give
  erroneous readings
• Usually can be resolved by fixing one component

7
Upstream Errors

• Caused when an error in one component affects
  other components in sequence behind it
• These errors can be detected by knowing the
  dependencies of system components

8
Downstream Errors

• Caused when an error in one component affects
  other components in sequence after it
• Detection scheme similar to upstream errors
• A fault can cause an upstream and downstream
  error to happen simultaneously

9
Unknown Errors

• A specific set of conditions may not be found in
  either algorithm.
• An error message will be returned stating that an
  unknown error has occurred among certain related
  sensors.

10
Fixing Unknown Errors

• Human intelligence is typically needed to find
  the real cause of such a problem.
• When the root cause is isolated, the algorithm
  can be updated to reflect this event

11
Communication with G2

• G2 sends the sensor data to MATLAB
• MATLAB receives the data as a vector
• After processing, MATLAB returns a text message
  to G2 stating the algorithm output

12
Approaches

• 2 types of algorithms
• Rule-based
• Rules made to relate sensor errors to a root
  cause
• Database
• Correlation of sensors to error messages and
  dependencies
• Algorithm will run each time a new set of sensor
  data is received

13
Rule-Based Algorithm

• Rules will be implemented in MATLAB
• Rules will be a series of if-then or case
  statements inside a while loop

14
Rule Advantages

• Code executes quickly
• Easy-to-read code
• Can be edited quickly

15
Database-based Algorithm

• Database is a look-up table
• Given a certain combination of inputs, the
  database returns an error message if applicable

16
Database Advantages

• Database approach is more organized
• Database size grows more slowly than rule-based
  structure

17
Implementation Issues

• Algorithm is system-dependent
• Algorithm is only as good as the database
• Sensor placement affects algorithm performance

18
Your Turn

• Below is the diagram of a Rankine steam cycle
• Devise a sensor placement scheme and use a method
  to isolate 3 multi-sensor faults

19
The End

• Questions?