Jana Flochová and René K. Boel Faculty of Informatics and Information Technology Slovak university of Technology, Bratislava, Slovakia EESA Department,

Slides:



Advertisements
Similar presentations
Impossibility of Distributed Consensus with One Faulty Process
Advertisements

1 Fault Diagnosis for Timed Automata Stavros Tripakis VERIMAG.
Based on: Petri Nets and Industrial Applications: A Tutorial
26 September 2003U. Buy -- SEES 2003 Sidestepping verification complexity with supervisory control Ugo Buy Department of Computer Science Houshang Darabi.
Diagnosability Verification with Parallel LTL-X Model Checking Based on Petri Net Unfoldings Agnes Madalinski 1, and Victor Khomenko 2 1 Faculty of Engineering.
Distribution and Revocation of Cryptographic Keys in Sensor Networks Amrinder Singh Dept. of Computer Science Virginia Tech.
A Novel Method For Fast Model Checking Project Report.
Sheng Xiao, Weibo Gong and Don Towsley,2010 Infocom.
Dynamic Bayesian Networks (DBNs)
TRUST Spring Conference, April 2-3, 2008 Write Markers for Probabilistic Quorum Systems Michael Merideth, Carnegie Mellon University Michael Reiter, University.
Introduction of Probabilistic Reasoning and Bayesian Networks
An Introduction to Markov Decision Processes Sarah Hickmott
Hidden Markov Models in NLP
Service Discrimination and Audit File Reduction for Effective Intrusion Detection by Fernando Godínez (ITESM) In collaboration with Dieter Hutter (DFKI)
Markov Reward Models By H. Momeni Supervisor: Dr. Abdollahi Azgomi.
Planning under Uncertainty
Synthesis of Embedded Software Using Free-Choice Petri Nets.
Petri Nets Overview 1 Definition of Petri Net C = ( P, T, I, O) Places P = { p 1, p 2, p 3, …, p n } Transitions T = { t 1, t 2, t 3, …, t n } Input.
Dynamic Tuning of the IEEE Protocol to Achieve a Theoretical Throughput Limit Frederico Calì, Marco Conti, and Enrico Gregori IEEE/ACM TRANSACTIONS.
Building Low-Diameter P2P Networks Eli Upfal Department of Computer Science Brown University Joint work with Gopal Pandurangan and Prabhakar Raghavan.
Tolerating Faults in Counting Networks Marc D. Riedel Jehoshua Bruck California Institute of Technology Parallel and Distributed.
Design of Fault Tolerant Data Flow in Ptolemy II Mark McKelvin EE290 N, Fall 2004 Final Project.
Chess Review May 11, 2005 Berkeley, CA Closing the loop around Sensor Networks Bruno Sinopoli Shankar Sastry Dept of Electrical Engineering, UC Berkeley.
1 University of Freiburg Computer Networks and Telematics Prof. Christian Schindelhauer Wireless Sensor Networks 13th Lecture Christian Schindelhauer.
Toward Optimal Network Fault Correction via End-to-End Inference Patrick P. C. Lee, Vishal Misra, Dan Rubenstein Distributed Network Analysis (DNA) Lab.
1 Software Testing and Quality Assurance Lecture 5 - Software Testing Techniques.
ESA PetriNet: Petri Net Tool for Reliability Analysis Romaric Guillerm, Nabil Sadou, Hamid Demmou 14 Oct LAAS-CNRS.
Extracting Places and Activities from GPS Traces Using Hierarchical Conditional Random Fields Yong-Joong Kim Dept. of Computer Science Yonsei.
On Probabilistic Snap-Stabilization Karine Altisen Stéphane Devismes University of Grenoble.
Authors: Sheng-Po Kuo, Yu-Chee Tseng, Fang-Jing Wu, and Chun-Yu Lin
Chapter 8 Architecture Analysis. 8 – Architecture Analysis 8.1 Analysis Techniques 8.2 Quantitative Analysis  Performance Views  Performance.
Jorge Muñoz-Gama Universitat Politècnica de Catalunya (Barcelona, Spain) Algorithms for Process Conformance and Process Refinement.
Reinforcement Learning
Soft Sensor for Faulty Measurements Detection and Reconstruction in Urban Traffic Department of Adaptive systems, Institute of Information Theory and Automation,
On Probabilistic Snap-Stabilization Karine Altisen Stéphane Devismes University of Grenoble.
1 Distributed Fault Detection for untimed and for timed Petri nets René Boel, SYSTeMS Group, Ghent University with thanks to: G. Jiroveanu, G. Stremersch,
Integrating UML and Petri Nets Problem with Current Software Engineering Methodology Stochastic Petri nets and their useful properties Translating UML.
LANGUAGE MODELS FOR RELEVANCE FEEDBACK Lee Won Hee.
Estimating Component Availability by Dempster-Shafer Belief Networks Estimating Component Availability by Dempster-Shafer Belief Networks Lan Guo Lane.
Agenda Fail Stop Processors –Problem Definition –Implementation with reliable stable storage –Implementation without reliable stable storage Failure Detection.
A Passive Approach to Sensor Network Localization Rahul Biswas and Sebastian Thrun International Conference on Intelligent Robots and Systems 2004 Presented.
April 28, 2003 Early Fault Detection and Failure Prediction in Large Software Systems Felix Salfner and Miroslaw Malek Department of Computer Science Humboldt.
Petri Nets Lecturer: Roohollah Abdipour. Agenda Introduction Petri Net Modelling with Petri Net Analysis of Petri net 2.
Modelling by Petri nets
Modeling Mobile-Agent-based Collaborative Processing in Sensor Networks Using Generalized Stochastic Petri Nets Hongtao Du, Hairong Qi, Gregory Peterson.
Behavioral Comparison of Process Models Based on Canonically Reduced Event Structures Paolo Baldan Marlon Dumas Luciano García Abel Armas.
Software Engineering1  Verification: The software should conform to its specification  Validation: The software should do what the user really requires.
School of Computer Science & Software Engineering
CAP 4800/CAP 5805: Computer Simulation Concepts
Probabilistic Automaton Ashish Srivastava Harshil Pathak.
1 Hidden Markov Models Hsin-min Wang References: 1.L. R. Rabiner and B. H. Juang, (1993) Fundamentals of Speech Recognition, Chapter.
Distributed cooperation and coordination using the Max-Sum algorithm
IEEE AI - BASED POWER SYSTEM TRANSIENT SECURITY ASSESSMENT Dr. Hossam Talaat Dept. of Electrical Power & Machines Faculty of Engineering - Ain Shams.
PDEVS Protocol Performance Prediction using Activity Patterns with Finite Probabilistic DEVS DEMO L. Capocchi, J.F. Santucci, B.P. Zeigler University of.
A fault tree – Based Bayesian network construction for the failure rate assessment of a complex system 46th ESReDA Seminar May 29-30, 2014, Politecnico.
Mean Field Methods for Computer and Communication Systems Jean-Yves Le Boudec EPFL Network Science Workshop Hong Kong July
Chapter 8 Fault Tolerance. Outline Introductions –Concepts –Failure models –Redundancy Process resilience –Groups and failure masking –Distributed agreement.
Adding Dynamic Nodes to Reliability Graph with General Gates using Discrete-Time Method Lab Seminar Mar. 12th, 2007 Seung Ki, Shin.
第1部: 自己安定の緩和 すてふぁん どぅゔぃむ ポスドク パリ第11大学 LRI CNRS あどばいざ: せばすちゃ てぃくそい
Clockless Computing COMP
Risk-informed Decision Making under Incomplete Information
COT 5611 Operating Systems Design Principles Spring 2012
CAP 4800/CAP 5805: Computer Simulation Concepts
CAP 4800/CAP 5805: Computer Simulation Concepts
A. Mancusoa,b, M. Compareb, A. Saloa, E. Ziob,c
Hierarchical Search on DisCSPs
COT 5611 Operating Systems Design Principles Spring 2012
CAP 4800/CAP 5805: Computer Simulation Concepts
Predictability Verification with Petri Net Unfoldings
COT 5611 Operating Systems Design Principles Spring 2014
Presentation transcript:

Jana Flochová and René K. Boel Faculty of Informatics and Information Technology Slovak university of Technology, Bratislava, Slovakia EESA Department, Ghent University, Belgium On fault diagnosis of random free-choice Petri nets

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Outline of the presentation  Models, diagnosis of DES based on Petri net models  Minimal context and explanations – (Jiroveanu, Boel, Bordbar 2008)  Probabilistic (random) free choice Petri nets  Calculation of likelihood values for minimal explanations; probabilities of failures  Deterministic analysis of the past, probabilistic analysis of the future  Examples  Conclusions

Models – Petri Nets 4)M 0 : P  N is the initial marking <, #,  denote precedence, conflict, concurrency relations of nodes A free-choice Petri net is a restricted class where every arc from a place to a transition is either the unique output arc from that place, or a unique input arc to the transition.

Models – Petri Nets An occurrence net O is a net O= (B, E,  ), with the elements of B called conditions, those of E called events, satisfying following properties   x  B  E   [x  x] (no node is in self conflict)  x  B  E   [x < x] (is a partial order, acyclic)  x  B  E   {y: y < x}  <  (is well-formed)  b  B:  b  1 (  b denotes the set of input elements of b => each place has at most one input transition, no backward conflict). A configuration C=(B c, E c,) is a subset of O, which is: conflict free (no two nodes are in conflict), causally upward-closed (if x´< 1 x, and x  C, then x´  C), and min(C)  min (O).

Models – Petri Nets

We consider the following structural and functional assumptions:  The overall plant model is bounded (possibly well formed free-choice)  The initial marking M 0 is precisely known, the set of transitions T = T o  T uo  The plant observation is represented by a subset of observable transitions  The occurrence of an observable transition T o is always reported correctly and without delays  No design-error assumptions Diagnosis based on PN – problem statement

We consider the following structural and functional assumptions:  The overall plant model is bounded (possibly well formed free-choice)  The initial marking M 0 is precisely known, the set of transitions T = T o  T uo  The plant observation is represented by a subset of observable transitions  The occurrence of an observable transition T o is always reported correctly and without delays  No design-error assumptions Diagnosis based on PN – problem statement

We consider the following structural and functional assumptions:  The overall plant model is bounded (possibly well formed free-choice)  The initial marking M 0 is precisely known, the set of transitions T = T o  T uo  The plant observation is represented by a subset of observable transitions  The occurrence of an observable transition T o is always reported correctly and without delays  No design-error assumptions Diagnosis based on PN – problem statement

We consider the following structural and functional assumptions:  The overall plant model is bounded (possibly well formed free-choice)  The initial marking M 0 is precisely known, the set of transitions T = T o  T uo  The plant observation is represented by a subset of observable transitions  The occurrence of an observable transition T o is always reported correctly and without delays  No design-error assumptions Diagnosis based on PN – problem statement

We consider the following structural and functional assumptions:  The overall plant model is bounded (possibly well formed free-choice)  The initial marking M 0 is precisely known, the set of transitions T = T o  T uo  The plant observation is represented by a subset of observable transitions  The occurrence of an observable transition T o is always reported correctly and without delays  No design-error assumptions Diagnosis based on PN – problem statement

We consider the following structural and functional assumptions:  The overall plant model is bounded (possibly well formed free-choice)  The initial marking M 0 is precisely known, the set of transitions T = T o  T uo  The plant observation is represented by a subset of observable transitions  The occurrence of an observable transition T o is always reported correctly and without delays  No design-error assumptions Diagnosis based on PN – problem statement

Faults T f are represented by a subset T f  T uo of unobservable (silent transitions – ( due e.g. limited sensor information ) A fault or an unreliable sensor (when some messages may become lost) can be modelled provided that another unobservable transition is included in the model "in parallel" to the observable transition Normal behaviour Faulty behaviour Diagnosis based on PN – problem statement

G. Jiroveanu, R.K. Boel, and B. Bordbar. On- Line Monitoring of Large Petri Net Models Under Partial Observation. Journal Discrete Event Dynamic Systems, 2008 Minimal context, minimal explanation, minimal marking. Diagnosis based on PN – problem statement

Centralized diagnosis of DES based on minimal explanations

Probabilistic settings  The probability of firing a transition should not depend on what concurrent transitions do, and the order on which concurrent transitions fire should not be randomized  Firing should not necessarily be reduced to one transition at a time.  The probability of firing a given transition depends only on its own recourses.

Probabilistic settings 0,7 0,25 0,05

Probabilistic settings The probability function on the set of configurations is defined as follows

Probabilistic settings  A stochastic analysis of faults that either occurred in the past or that may occur in the future prior to the next observed event occurrence (Flochová et al. 2007); so that the explanation only includes unobservable future events not belonging to the minimal explanations.  A deterministic analysis of faults that must have occurred in the past (Jiroveanu, Boel, Berdbar 2008) and a probabilistic analysis of faults that may occur in the future prior to the next observed event occurrence.

Probabilistic settings Having the set of minimal configurations C(On), respectively the set of minimal explanations of the received observations L N (O n ) is defined

Probabilistic settings The plant diagnosis after observing On based on the set of minimal explanations - obtained by projecting the set of minimal explanations onto the set of fault events Having the set of minimal configurations C(On), respectively the set of minimal explanations of the received observations L N (O n ) is defined

Probabilistic settings The plant diagnosis after observing On based on the set of minimal explanations - obtained by projecting the set of minimal explanations onto the set of fault events Having the set of minimal configurations C(On), respectively the set of minimal explanations of the received observations L N (O n ) is defined

Probabilistic settings The plant diagnosis after observing On based on the set of minimal explanations - obtained by projecting the set of minimal explanations onto the set of fault events Having the set of minimal configurations C(On), respectively the set of minimal explanations of the received observations L N (O n ) is defined

Probabilistic settings All explanations - similar expressions after removing all underscores.

Probabilistic settings

Steps needed in order to derive fault probabilities: (1)Compute the set of minimal explanations of the most recent observed event. Derive minimal explanations of the last observed event t 0 and minimal explanations of a sequence of observed events. (2) Compute the unnormalized probability of all minimal explanations (3) Sort explanations in descending order starting from the most probable ones. Shellsort can be used, branch and bound like improvements can be useful in order to avoid enumerating very unlikely explanations. (4) Accept top x % (0-100 %) of explanations according to the input requirements. (5) Compute the set of maximal explanations of the most recent observed event, if required.

Probabilistic settings (6) Compute the unobservable continuations, which follow after the next observable transitions and partition the continuations into the following sets: the set of configurations, which contain at least a faulty event; a set of configurations, which contain at least a faulty event of the fault of the type i; and the set of configurations, which don’t contain any faulty event. A modification of classical AI depth search, which evaluates at first the node that has the most nodes between itself and the last observed transition, can be used for computing the set of continuations equipped with probabilities.

Probabilistic settings ( 7) Compute the unnormalized probabilities of the faults (faults of the type i) of all continuations (of unobservable reaches after the last observation). (8) Compute the unnormalized probabilities of the faults (faults of the type i) based on the sets of all explanations. (9) Normalize the probabilities

Example

Laboratory example- older Fischertechnik-model old unreliable sensors and all parts, AB PLC control

!!!!Possibly a model, shortly

Minimal explanations of the last event

Conclusions Two methods of probabilistic diagnosis were presented, both methods use minimal explanations and contexts concept, probabilities assigned to conflicting transitions and, reverse Petri nets. They both are based on [George and you] or better [George, you and Bordbar], and [Benveniste et al.] approaches. 1. the method uses the probabilistic analysis of the plant evolution before the last observed event and the probabilistic estimation of the future evolution of the plant after the last observed event [NYC]. 2. The second method (novel approach) is based on the deterministic analysis of the plant evolution before the last observed event and the probabilistic estimation of the possible future failure evolution of the plant.

Conclusions Two methods of probabilistic diagnosis were presented, both methods use minimal explanations and contexts concept, probabilities assigned to conflicting transitions and, reverse Petri nets. They both are based on [George and you] or better [George, you and Bordbar], and [Benveniste et al.] approaches. 1 st method uses the probabilistic analysis of the plant evolution before the last observed event and the probabilistic estimation of the future evolution of the plant after the last observed event [NYC]. 2. The second method (novel approach) is based on the deterministic analysis of the plant evolution before the last observed event and the probabilistic estimation of the possible future failure evolution of the plant.

Conclusions Two methods of probabilistic diagnosis were presented, both methods use minimal explanations and contexts concept, probabilities assigned to conflicting transitions and, reverse Petri nets. They both are based on [George and you] or better [George, you and Bordbar], and [Benveniste et al.] approaches. 1 st method uses the probabilistic analysis of the plant evolution before the last observed event and the probabilistic estimation of the future evolution of the plant after the last observed event [NYC]. 2 nd method (a novel approach) is based on the deterministic analysis of the plant evolution before the last observed event and the probabilistic estimation of the possible future failure evolution of the plant.

The probabilistic setting allows us to incorporate statistical knowledge: on the production of faults: some event may be more likely than the others depending on reliability tests on devices, on the previous experience on monitoring the plant or the network (relative frequencies of spontaneous faults), on the loss of information on faults (e.g. masking of an alarm, temporally unavailable links, faults of protocols). Methods allow some smoothness of observation, i.e. including of misleading observations and not observing of a normally observable events in the model. Randomization of the model also provides a convenient way of introducing robustness of the model against modeling errors on faults propagation. Advantages of the approach

Problems and open questions The process of randomization has to be done very carefully and one has to tackle several problems in assigning probabilities. Decentralized diagnosis algorithms and distributing setting are needed to allow fault detection in large plants possible solution - several communicating probabilistic Petri nets components computing local probability assignment for all locally possible traces explaining observations. -components can interact by exchanging tokens via boundary places (or boundary synchronizing transitions), common normalization for both interacting component; Relaxing the assumption of well formed free choice Petri nets following [Haar 2003]

Benveniste, A. et al.: “Fault detection and diagnosis in distributed systems: an approach by partially stochastic Petri nets.” Discrete Event Dynamic Systems: Theory and Applications, vol. 8, pp , June A. Benvensite, E. Fabre, and S. Haar. Markov nets: Probabilistic models for distributed and concurrent systems. IEEE Transactions on Automatic Control, 48(11):1936–1950, Benveniste, A. et al.: “Diagnosis of asynchronous discrete event systems, a net unfolding approach.” IEEE Transactions on Automatic Control, 48(5), pp , May S. Haar, ”Probabilistic cluster unfoldings for Petri nets”,Technical report 1517, IRISA, Rennes, France, J. Esparza. S. Romer and W. Vogler. An improvement of McMillan’s unfolding algorithm. Lect. Notes in Computer Science 1055, 87–106, Springer-Verlag, J. Flochova, R. K. Boel, and G. Jiroveanu. On Probabilistic Diagnosis for Free-Choice Petri Nets. Proceeding of ACC, NYC, US, 5655–5656, G. Jiroveanu, R.K. Boel, and B. Bordbar. On-Line Monitoring of Large Petri Net Models Under Partial Observation. Journal Discrete Event Dynamic Systems, 18:323–354, M. Nielsen, G. Plotkin, and G. Winskel. Petri nets, event structures and domains, part I. Theoret. Computer Science, 13:85–108, 1981.

??? Thank you for your attention