Engine Health Management System Diagnostics and Prognostics

1
Engine Health Management SystemDiagnostics and
Prognostics

• IBM Academy Conference Apr 26-27, 2005

Dan Cleary LiJie Yu
General Electric Global Research
Center Niskayuna, NY
2
Contents

• Aircraft engine monitoring overview
• Present Capability
• Problem statement and quality goals
• EHM technology map
• Alert integration
• Learning and adaptation
• Information fusion
• Summary

3
Robust Data Infrastructure for Engine Monitoring
Data Management Diagnosis
Data Acquisition
GE Engineering

• In-depth analysis
• Model updates

• Email
• Page
• Cell

Root Cause Analyzer
Internet

• Manage Alert Levels
• Access Alert Details
• Manage Watch Lists
• Export Plots Reports
• Monitor In-Flight Data
• Reviewing Diagnostic results and Recommendations

4
RCA - Sensor Fusion Technology for Aircraft
Engine Diagnostics

• Measure and evaluate performance shifts
• Combine with engineering expertise
• Provide accurate and consistent diagnostics
  recommendations
• Service offered directly to customers

5
Problem Statement and Quality Metrics

• Problem Statements
• Engine Health Management is a difficult problem
  to solve
• Need Intelligent Alerts. False alerts misdirect
  resources.
• Existing diagnostic models are labor intensive
• Multiple touch point diagnostics process

Quality Metrics

• Accuracy and precision
• Failure Coverage
• Productivity
• Reduced modeling efforts
• Incorporate customer Logic

Integrated EHM
6
Technical Challenges

• Researching developing the Best decision
  algorithms for detection, diagnostics and
  optimization for EHM.
• Integrating highly diverse information sources
• Discovering new fault modes and signatures
• Integrating with existing IT infrastructure,
  adopting effective and efficient software
  architecture
• Collaborating among multi-functional teams
• Getting the most from historical data
• Obtaining reliable customer feedback

7
Technology Roadmap

• Alert Integration
• On demand
• Alerted engine

Engine Aircraft Data
Start
Set A
Set B
Set
Feature Extraction Characterization
Data Cleaning
Engine Feature Sets
Hints

• Experience
• Diagnostic Models
• Statistics
• Engineering
• Business Logic

Capability
Data Level Fusion Techniques
Technology Thrusts
Diagnostics/Prognostics

• Alerting integration and decision thresholds
• Decision level fusion framework
• Automated diagnostic model learning and
  adaptation
• Improved failure discovery and examination

Diagnostic Decision Fusion
Recommendation
Customer Feedback
Learning Adaptation
8
Alerting Integration
Low
High
False alerts
Phase 1
Phase 2
Phase 3
Ongoing
High
Low
Increased Diagnosis Confidence
Prediction
Productivity
9
Automatic Learning and Adaptation
Objective Develop an adaptive system to
automatically learn failure signature patterns
and to provide probabilistic prediction for
engine diagnostics and health monitoring.
Various engine features are extracted and
examined for this activity.

• Business Benefits
• Automate diagnostic model creation and tuning
  process thus reducing maintenance cost
• Enable model optimization
• Automatic model performance tracking for belief
  worthiness assessment
• Automatically learn new fault signatures

• Technical Challenges
• Integrate physical and empirical information
• Improve model coverage, accuracy and precision
  with limited training data
• Integrate with feedback system for fully
  automated model adaptation and tuning
• Adaptive to new unknown failure mode

10
Learning-Adaptation Process
Model learns and adapts from both experience and
physical model investigations
11
Example Learning Adaptation
Parameter Readings
Feedback adaptation
P1
Feature Extraction
Clustering
Regression
P1 -22.35 P2 0.34 P3
-0.08
Model
P2
Neural Network
Neural Fuzzy
Genetic Algorithm
P3
Bayesian Learning
Learn
New case
Run
Case Classification
P1
0.01
No fault
0.95, accuracy 85, 5 similar cases
Fault 1
P1 25.85 P2 3.18 P3 -0.58
P2
0.15
Fault 2
0.3
Fault 3
0.05
P3
Fault 4
12
Information Fusion
Fusion
FUS
Data Module
DM
Decision Module
FUS
DM
Expert Assessment
Information Module
Data Module
FUS
DM
Decision Module
Information Module
Sensor Data
FUS
DM
Decision Module
Competing Decisions
13
Summary

• Automated diagnostic process supports business
  success and growth
• Physical model and data driven approach is
  combined for diagnostic model optimization
• Automated, integrated, and adapted information
  management strategy is adopted
• Technology map is driven by total engine health
  management goal