Alex Shenfield

1
Modelling, Optimisation and Decision Support
Using the Grid
Alex Shenfield a.shenfield_at_sheffield.ac.uk
Rolls-Royce University Technology Centre in
Control Systems Engineering Department of
Automatic Control Systems Engineering The
University of Sheffield, UK.
2
Overview of Presentation

• Introduction
• UK e-Science DAME project
• Motivation for DAME
• DAME Grid-Based Diagnostic System
• Case Based Reasoning
• Model Based Fault Detection and Isolation
  Approaches
• Genetic Algorithms for Many-Objective
  Optimisation
• Use Case
• Conclusions

3
Introduction to DAME

• 3.2M UK e-Science Pilot Project
• Develop, and promote understanding of
• Grid middleware and application/services layer
  integration
• Real-time issues in Grid Computing
• Dependability Issues
• Provide a Proof of Concept demonstrator for the
  Rolls-Royce Engine Diagnostic problem

4
Project Partners

• Four UK Universities
• University of York
• Computer Science Department
• University of Sheffield
• Automatic Control and Systems Engineering
  Department
• University of Leeds
• School of Computing
• School of Mechanical Engineering
• University of Oxford
• Engineering Science Department
• Industrial Partners
• Rolls-Royce Aeroengines
• Data Systems and Solutions
• Cybula Ltd.

5
Motivation for DAME

• Increasing amounts of engine data being collected
• New engine monitoring units record up to 1 Gbyte
  of data per flight
• Rolls-Royce currently has over 50,000 engines in
  service with total operations of around 10M
  flying hours per month
• In the future, terabytes of data will be
  transmitted every day for analysis
• Key Objectives
• Reduce delays
• Reduce cost of ownership for the aircraft

6
Case-Based Reasoning

• CBR is a mature, low-risk subfield of AI
• Primary knowledge source
• A memory of stored cases recording specific prior
  episodes
• Not generalised rules
• New solutions generated by adapting relevant
  cases from memory to suit new situations

Retrieve
Propose Solution
Adapt
Justify
Criticize
Evaluate
Store
7
CBR Maintenance Advisor

• Integrates fault information and knowledge gained
  from the fault diagnosis process
• Emulate the diagnostic skill of an experienced
  maintenance engineer
• Advises maintenance personnel on appropriate
  maintenance action
• Deployed as a Grid Service

8
CBR Engine Architecture
9
CBR Engine Architecture

• Interface between application and data
• Reconfigurable

10
CBR Engine Architecture

• Contains CBR matching and ranking algorithms

11
CBR Engine Architecture

• Processes calls to the CBR service
• Returns results from the CBR service

12
CBR Engine Architecture
13
Model Based FDI

• Data from the real engine is compared against
  data from the ideal model
• The residuals then need to be analysed to work
  out the state of the engine
• This can be used to track changes in engine
  parameters which may indicate impending faults

14
Engine Modelling and Simulation Service

• Based on the Rolls-Royce Trent 500 engine model
• Deployed as a service on the Grid
• Accessible via web browser on the internet
• Grid factories enable parallel execution of
  multiple simulation instances

15
Genetic Algorithms

• Genetic Algorithms (GAs) are global search
  algorithms based on the mechanics of natural
  selection
• GAs are robust search methods
• Can escape local optima
• Can deal with noisy or ill-defined evaluation
  functions
• Some features of GAs are
• GAs search a population of points
• GAs use objective function pay-off information
• GAs are stochastic

16
A Simple Genetic Algorithm
17
Multi-Objective Optimisation

• Many real-world engineering design problems often
  involve solving multiple (often conflicting)
  objectives

• An ideal multi-objective optimisation procedure
  is
• Find multiple Pareto optimal solutions for the
  objectives

18
Multi-Objective Optimisation

• Many real-world engineering design problems often
  involve solving multiple (often conflicting)
  objectives

• An ideal multi-objective optimisation procedure
  is
• Find multiple Pareto optimal solutions for the
  objectives
• Choose one of the trade-off solutions using
  higher level information

19
Integrated Logistic Support Strategy Optimisation

• MEAROS Optimisation
• Removal of aircraft engines is expensive
• By using GAs to optimise soft lives of engine
  components in the MEAROS simulation we can
  develop optimal preventative maintenance
  strategies
• Issues
• MEAROS is a complex stochastic simulation,
  therefore it has to be run multiple times for
  each candidate solution to reduce the effect of
  random variations
• This requires a lot of computing power

THE GRID !
20
MOGA-G Architecture
21
DAME Use Case
22
DAME Use Case

• In the future

Failure Rate Data learnt from DAME
MEAROS MODEL
23
Security

• The Decision Support System will contain
  sensitive data, therefore access must be
  restricted
• i.e. Knowledge Base and Engine Model contain
  information on the design characteristics and
  operating parameters of the engine
• Security implemented using Globus Toolkit to
  provide
• Public Key Encryption
• X509 certificates
• SSL communications

24
Conclusions

• Move from local diagnostic support to
  centralised, distributed diagnostic support
• Integration of Model-Based FDI, CBR and
  Optimisation
• Business Benefits
• Reduction in unscheduled maintenance
• Reduction in aircraft downtime

25
Thanks!

• The authors gratefully acknowledge the
• financial support of the EPSRC and the
• valuable input from engineers at
• Rolls-Royce and Data Systems
• Solutions