CBR Decision Support - Department of Computer Science

advertisement
CBR for Fault Analysis
in DAME
Max Ong
University of Sheffield
Overview
Overview of presentation
• Introduction to Case-Based Reasoning (CBR) technology
• CBR for Fault Analysis
– Maintenance Advisor
– Grid Service Implementation
– User interface to the Grid Service
• Related Work
– MEAROS
– CBR Workflow Advisor
Distributed Aircraft Maintenance Environment - DAME
Case-Based Reasoning
Retrieve
Propose Solution
Adapt
• CBR is a mature, low-risk subfield of A.I.
• Primary knowledge source
– a memory of stored cases recording
specific prior episodes
– not generalised rules
Justify
Criticize
Evaluate
Store
• New solutions generated by adapting
relevant cases from memory to suit new
situations
“Reasoning by remembering,
reasoning is remembered.”
Distributed Aircraft Maintenance Environment - DAME
Case-Based Reasoning
“CASE”
• Description of situation
• Description of problem
in that situation
• Description of how
problem was addressed
• Results or outcome of
addressing the problem
in that way
What is a case?
“A case is a contextualised piece of
knowledge representing an
experience that teaches a lesson
fundamental to achieving the goals
of the reasoner.” (Kolodner, 1993)
•
Cases link together knowledge that
belongs together:
– Information concerning the fault
– Response to the fault
– Effects of those responses
Distributed Aircraft Maintenance Environment - DAME
CBR Knowledge
Casebase
• Repository of fault cases
• Cases represent individual engine fault events and maintenance actions
Knowledge Model
• Contains general knowledge about a specific field of application
• Comprises the data structure (data model), the relationship between values
within the data (valuation models) and information for defining applicationspecific vocabulary in the data
Valuation Model
• Defines the similarities and evaluations of individual pieces of data in relation to
each other
• Facilitates knowledge-based, intelligent searches (directed search)
Distributed Aircraft Maintenance Environment - DAME
CBR Search
CBR Tool Implements
Nearest Neighbour (k-NN) algorithm
Bad
Condition
Engine
Case 'b'
X
New Case
New Case
b
Y
N1
[= Index 2]
•
b
Close - Up
Y
a
X
Case 'a'
a
Good Condition Engine
EPR
[= Index 1]
Degree Of Similarity NewCase , Case ' a'  f  Weighted Distance   WX  X  X a  WY  Y  Ya

 Wi  X iNewCase  X iCase'a '

or Wi  X
NewCase
i
X

1
Case'a '
i

2
i  1 to n
n  Number Of Attributes ( Indexes)
Distributed Aircraft Maintenance Environment - DAME
CBR Induction
Inductive Retrieval
– Induction - ID3 Algorithm
– Knowledge-Guided Induction
1. Discriminate by EPR
EPR > 1.1
Example:
Yes
No
Yes
No
TG
Case 4
T
R
N1
EP
e
gin
En tus
S ta
Case 1
Case 1
Very Good
1.6
100%
500 oK
Case 2
Good
1.5
105%
550 oK
Case 3
Very Bad
1.0
106%
600 oK
Case 4
Bad
1.1
100%
500 oK
Case 2
Good Condition
Engines
Case 3
Bad Condition
Engines
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
CBR Maintenance Advisor:
 Emulates the diagnostic skill of an experienced maintenance
engineer
 Provides the user with ‘best practice’ advice when confronted
with a set of fault symptoms
 Provides a confidence measure for each suggested solution
 Facilitates the processing of logistic data, current and
historical fault data to intelligently isolate an engine problem
Some common misconceptions:
x CBR is a relational database of engine faults
x CBR is only a search engine
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
• Why the Grid?
 Collaborative working environment (virtual organisation)
 Distributed data
 Interaction with other services
 High performance computing (large casebase computation)
 Scalability
 Delivery via Grid Services with security
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
Implementation




Import fault and maintenance
data from SQL database
Creation of datatypes within
CBR
Map fault information to
Analysis and Valuation Model
Pre-indexed casebase,
knowledge and valuation
models are stored in XML
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
Implementation
CBR Maintenance Advisor - Grid Service
• Deployed as a service on the
Grid with Globus 3 Toolkit
• Accessible by user via a web
browser across the Internet
• Secure access with user and
host authentication, SSL
encryption.
• Access to fault information
and knowledge gained from
fault diagnosis processes
• Obtain maintenance advice
in form of appropriate
maintenance action
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
Implementation
`
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
Implementation
Distributed Aircraft Maintenance Environment - DAME
CBR Maintenance Advisor
Implementation
Distributed Aircraft Maintenance Environment - DAME
Quick summary
• Case-Based Reasoning (CBR) technology
• CBR for Fault Analysis
– “Maintenance Advisor” Service in DAME
– Globus 3 Grid Service Implementation
• Related work
– MEAROS
– CBR Workflow Advisor
Distributed Aircraft Maintenance Environment - DAME
MEAROS
Modular Engine Arisings & Overhaul Simulation
• Removal of aircraft engines is expensive, significant fraction of operating costs
• MEAROS enables ‘optimal’ preventative maintenance strategies to be determined
• MEAROS within virtual maintenance environment can enhance fleet management of aircraft and engines
• MEAROS simulation is a very compute-intensive process
• The Grid offers high-performance computing resources, enabling faster results and quicker decisions to
facilitate maintenance planning
Module Failure
Rate Data
The module failure rate data is used
to improve the failure distributions for
the components inside the model.
MOGA
Send Population of
Solutions
MEAROS
Client
Receive Evaluation
Results
Multi-objective
Optimisation with
Genetic Algorithms :
• selection
• crossover
• mutation
• re-insertion
• Interprets the results from MEAROS
using a costing model
• Provides a way of distributing solutions
to MEAROS and collecting the results
Distribute
Solutions
MEAROS
Collect
Results
Stochastic model for the
simulation of operation,
maintenance and
supply of gas turbine
engines to fleets of
aircraft
Distributed Aircraft Maintenance Environment - DAME
Download