structural abstraction for strong fault models diagnosis (dx 2014 bisfai 2015) roni sternmeir...
TRANSCRIPT
AI@BGU
Structural Abstraction for Strong Fault Models Diagnosis (DX 2014 • BISFAI 2015)
Roni Stern Meir Kalech Orel Elimelech
Ben Gurion University of the Negev, IsraelDepartment of Information Systems Engineering
2
Outline Introduction to Diagnosis Model-Based Diagnosis
Definition & Motivation Abstraction Literature Review Research Goal
Methodology Evaluation
Results Conclusions Future Work
3
Outline Introduction to Diagnosis Model-Based Diagnosis
Definition & Motivation Abstraction Literature Review Research Goal
Methodology Evaluation
Results Conclusions Future Work
4
What is a Diagnosis?• Identifying the reason for a problem by examining observed symptoms.
• Determining which part of the system is failing.
6
Diagnosis Approaches• Expert Systems.
• Case-Based Reasoning.
• Probabilistic Reasoning.
• Model-Based Diagnosis.
• And more…
7
Outline Introduction to Diagnosis Model-Based Diagnosis
Definition & Motivation Abstraction Literature Review Research Goal
Methodology Evaluation
Results Conclusions Future Work
8
Model-Based DiagnosisCar System Model A Real Car
[Raymod Reiter. A theory of diagnosis from first principles. 1987].[Johan de Kleer and Brian C. Williams. Diagnosing multiple faults. 1987].
13
Grounding
Faulty
Input 1 = No Fluids
Input 2 = No Fluids
Expected Output = No FluidsObserved Output = Fluids
Inputs 3 & 4 = No Fluids
Pipe 9 is
Faulty
Grounding
14
Architecture & Terminology
Original System
Abstract System
Diagnosis Engine
Abstract DiagnosesDiagnoses for the Original System
Find Abstraction Finds
Grounding
16
Ungroundable Abstract Diagnosis
Input = No Fluids
Expected Output = No FluidsObserved Output = Fluids
Pipe 2 Mode: Healthy \ Blocked
Healthy No Fluids
Blocked No Fluids
Pipe 1
Pipe 2
An abstract component of 2 pipes
The grounding process fails
Using abstraction here is not easy
Can’t explain the observed output Faulty
17
Literature Review
[Metodi, Stern, Kalech and Codish. Compiling Model-Based Diagnosis to Boolean Satisfaction. 2012]
Past work assumed 2 behavior modes: Healthy \ Faulty
Faulty Any desired behaviorPipe 1
Pipe 2
Abstract diagnoses will always be groundable
18
Literature ReviewSome have already tried to diagnose systems with multiple fault modes.
Conflict-Directed with Abstraction• [Feldman, Provan and van Gemund. 2010]
Compilation-Based with Abstraction• [Torta and Torasso. 2013]
20
Outline Introduction to Diagnosis Model-Based Diagnosis
Definition & Motivation Abstraction Literature Review Research Goal
Methodology Evaluation
Results Conclusions Future Work
21
Discard abstract components
that may not be groundable
1. Pessimistic Approach
More components
Harder to diagnose
The grounding process is easier
22
Keep all abstract components and take the risk of failing
during the grounding process.
Less components easier to diagnose
The grounding process is harder
2. Optimistic Approaches
23
2.1. Weak-Optimistic Approach
2.2. Strong-Optimistic Approach
Grounding
Fail
Succ
ess
Grounding
Are there more abstract diagnoses?
Fail
Succ
ess
No
Yes
25
Outline Introduction to Diagnosis Model-Based Diagnosis
Definition & Motivation Abstraction Literature Review Research Goal
Methodology Evaluation
Results Conclusions Future Work
26
Evaluation
• Empirical evaluation.
• The approaches were implemented in Prolog.
• External SAT Solver.
• Modified version of the ISCAS-85 benchmark.
27
Results• Three systems:
• Diagnosis algorithm: SATbD
• Timeout: 20 seconds
System ID |Comps.| Before
Abstraction
|Comps.| After
Abstraction
c880 383 77
c1355 546 58
c2670 1193 167
X Axis – The ApproachesY Axis - Success Rate of Solved Instances
28
Conclusions & Summary
General approach for abstraction in strong-fault models.
Evaluation on a modified version of the ISCAS-85 benchmark.
Abstraction speeds up the diagnosis process.
Mostly the (weak) optimistic approach is the best.
29
Schedule & Future Work Find an efficient method to use abstraction in strong-fault models.
Model the approaches to SAT.
Evaluate the approaches (ISCAS-85 benchmark).
Compare our method to Torta and Torasso. 2013, Feldman et al. 2010.
Optimize the approaches.
Find an hybrid approach based on systems pre-process.
30
www.orel-e.com
+972 52 4370054
https://il.linkedin.com/in/
orelelimelech
Questions
?