System and methods for assessing risk using hybrid causal logic
원문보기
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06F-017/00
G06N-007/00
G06N-007/08
출원번호
UP-0377313
(2006-03-17)
등록번호
US-7774293
(2010-08-30)
발명자
/ 주소
Mosleh, Ali
Wang, Chengdong
Groen, Franciscus J.
출원인 / 주소
University of Maryland
대리인 / 주소
Rosenberg, Klein & Lee
인용정보
피인용 횟수 :
5인용 특허 :
2
초록▼
A hybrid causal framework applies properties of probabilistic models, such as Bayesian belief networks, to causal logic models, such as fault trees and event sequence diagrams. The probabilistic model establishes a joint probability distribution of causal relationships between events and conditions
A hybrid causal framework applies properties of probabilistic models, such as Bayesian belief networks, to causal logic models, such as fault trees and event sequence diagrams. The probabilistic model establishes a joint probability distribution of causal relationships between events and conditions in the logic models. The probability of the events and conditions are found by propagating probabilities from the probabilistic model through the logic models.
대표청구항▼
What is claimed is: 1. A method for determining via a computing processor a probability associated with a causal scenario including an initiating event, at least one pivotal event and an end state, the method comprising: modeling the causal scenario by a first causal model characterized by a plural
What is claimed is: 1. A method for determining via a computing processor a probability associated with a causal scenario including an initiating event, at least one pivotal event and an end state, the method comprising: modeling the causal scenario by a first causal model characterized by a plurality of first nodes interconnected one with another to define a termination of the causal scenario in the end state via a Boolean state of at least one variable associated with said plurality of first nodes; modeling factors affecting said Boolean state of said at least one variable at a corresponding one of said first nodes of said first causal model by a second causal model characterized by a plurality of second nodes, each of said second nodes representing a corresponding multistate variable indicative of an attribute of said factors, said plurality of second nodes being interconnected one with another in accordance with a joint probability distribution of said multistate variables, wherein said joint probability distribution is defined by a probabilistic network coupled to said second causal model for characterizing said factors, and wherein said second causal model includes at least one node corresponding to said at least one variable, said probabilistic network directly providing to at least one of said plurality of first nodes at least one second factor associated with said at least one pivotal event; constructing from said first causal model and said second causal model a hybrid computational model executable on the processor; and executing said hybrid computational model on the processor to determine a probability of said at least one pivotal event of the causal scenario from a probability of said Boolean state of said at least one variable of said first causal model as determined from a probability of said factors affecting said Boolean state of said at least one variable calculated by said second causal model in accordance with said joint probability distribution of said multistate variables defined by said probabilistic network. 2. The method for determining a probability associated with a causal scenario as recited in claim 1 where said hybrid computational model execution step includes the step of determining a conditional probability of said at least one variable being in a predetermined state given a set of states of said multistate variables. 3. The method for determining a probability associated with a causal scenario as recited in claim 2, where said hybrid computational model execution step includes the step of determining at each of said first nodes a conditional probability given a logical conjunction of states of said at least one variable. 4. The method for determining a probability associated with a causal scenario as recited in claim 3, where said first node conditional probability determining step includes the steps of: providing in a data store accessible to the computational processor a table for storing said conditional probabilities at said first nodes; storing in said table said conditional probability for a first node and a set of states if such is not entered therein; and retrieving a previously determined probability from said table in lieu of said determination thereof if an entry in said table exists for said first node and said set of states. 5. The method for determining a probability associated with a causal scenario as recited in claim 3, where said conditional probability of said event determination step includes the steps of: determining if said first node at which said conditional probability is determined is independent of said states of said at least one variable; and determining a marginal probability at said first node. 6. The method for determining a probability associated with a causal scenario as recited in claim 3, where said hybrid computational model constructing step includes the step of ordering in said first causal model said first nodes so that at least one node corresponding to said at least one variable is independent from another one of said at least one variable. 7. The method for determining a probability associated with a causal scenario as recited in claim 6, where said first node conditional probability determining step includes the steps of: filtering said states of said at least one variable to include only states upon which said variable corresponding to said first node is dependent; and determining said conditional probability at said first node using said filtered states. 8. The method for determining a probability associated with a causal scenario as recited in claim 1 further including the steps of: providing a user interface to the computational processor; manipulating first graphical elements via said user interface to construct a first representation of said first causal model; executing on the processor a procedure to transform said first representation of said first causal model into a second representation thereof, said second representation of said first causal model characterized by said plurality of first nodes; and constructing said hybrid computational model from said second causal model and said second representation of said first causal model. 9. A method for analyzing risk associated with a causal scenario comprising the steps of: manipulating via a user interface first graphical elements to model the causal scenario by a first causal model defined by sequential logic, said first causal model determining an occurrence or nonoccurrence of at least one pivotal event in the causal scenario in accordance with states of a plurality of first variables respectively associated with said first graphical elements; manipulating via said user interface second graphical elements to model factors associated with said at least one pivotal event by a second causal model defined by fault logic, said second causal model computing a probability of states of a plurality of second variables in accordance with a joint probability distribution of said plurality of second variables, each of said second variables respectively associated with said second graphical elements, said first causal model and said second causal model having at least one variable common therebetween; executing on a computational processor a first procedure to transform said first causal model into a nodal graph representation thereof, said nodal graph representation of said first causal model characterized by a plurality of first nodes respectively associated with said plurality of first variables; constructing a computational model from said second causal model and said nodal graph representation of said first causal model; computing on said computational processor a probability of a state of said at least one variable common between said first and second causal models in accordance with said joint probability distribution, wherein said joint probability distribution is determined by a probabilistic network coupled to said second causal model to characterize said factors, said probabilistic network directly providing at least one second factor associated with said at least one pivotal event to said first causal model; and computing on said computational processor a probability of said event in accordance with said probability of said state of said at least one variable common between said first and second causal models. 10. The method for analyzing risk associated with a causal scenario as recited in claim 9 further including the steps of: executing on said computational processor a second procedure to transform said second causal model into a second nodal graph representation thereof, said second nodal graph representation of said second causal model characterized by a plurality of second nodes respectively associated with said plurality of second variables, said second nodes interconnected one with another in accordance with said joint probability distribution of said second variables; and constructing said computational model from said second nodal graph representation of said second causal model and said nodal graph representation of said first causal model. 11. The method for analyzing risk associated with a causal scenario as recited in claim 10, where said computational model execution step includes the step of determining at each of said first nodes a conditional probability given a logical conjunction of states of said at least one variable common between said first and second causal models. 12. The method for analyzing risk associated with a causal scenario as recited in claim 11 further including the step of ordering said first nodes so that a first node corresponding to said at least one common variable is independent from another first node corresponding to another said at least one variable common between said first and second causal models. 13. The method for analyzing risk associated with a causal scenario as recited in claim 12, where said first node conditional probability determining step includes the steps of: filtering said states of said at least one variable common between said first and second causal models to include exclusively states upon which said variable corresponding to said first node is dependent; and determining said conditional probability at said first node using said filtered states. 14. An apparatus for evaluating risk in a system, comprising: a computational processor including a sequential logic unit characterizing anticipated risk scenarios of the system, said sequential logic unit including a plurality of decision units each corresponding to a pivotal event in said risk scenario, each of said decision units having an occurrence state or a nonoccurrence state of said pivotal event in accordance with a corresponding condition provided thereto; a fault logic unit associated with said computational processor and coupled to said sequential logic unit for providing said condition to each of said decision units, said fault logic unit including a plurality of combinatorial elements for determining each said condition from a corresponding set of causal factors; a probabilistic network unit associated with said computational processor and coupled to said fault logic unit for characterizing said causal factors of said system, said probabilistic network unit including a plurality of nodes interconnected one with another to define a joint probability distribution between variables representing said causal factors, said probabilistic network unit directly determining at least one of said conditions and being coupled to said sequential logic unit to provide said at least one condition thereto; and a hybrid causal model associated with said computational processor and determining a probability of said risk scenarios from a probability of each said pivotal event of said sequential logic as determined from a probability of each said corresponding condition provided thereto by said fault logic unit as determined from said joint probability distribution between said variables representing said corresponding causal factors of said probabilistic network unit. 15. The apparatus for evaluating risk factors associated with a system as recited in claim 14, wherein said probabilistic network unit determines directly at least one of said conditions and is coupled to said sequential logic unit so as to provide said at least one condition thereto. 16. The apparatus for evaluating risk factors associated with a system as recited in claim 14, wherein said hybrid causal model is characterized by a plurality of nodes arranged in a topological hierarchy corresponding to interconnections of said combinatorial logic and said probabilistic network unit, each of said nodes corresponding to a Boolean variable of said combinatorial logic or a variable of said probabilistic network unit. 17. The apparatus for evaluating risk factors associated with a system as recited in claim 14, wherein said sequential logic unit is an event sequence diagram. 18. The apparatus for evaluating risk factors associated with a system as recited in claim 14, wherein said fault logic unit is a fault tree. 19. The apparatus for evaluating risk factors associated with a system as recited in claim 14, wherein said probabilistic network unit is a Bayesian belief network.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (2)
Przytula,Krzysztof W.; Milford,Richard I., Automatic generation of baysian diagnostics from fault trees.
Hammet, Claudia P.; Ulmer, David H.; Cowan, John; Nemecek, Rachel; Dunlap, Jr., Edward M.; Williams, Thomas R., Customer impact predictive model and combinatorial analysis.
Cheriere, Vincent; Roger, Jeremy; Vilalta Estrada, Laia; Geanta, Ioana, Method, devices and program for computer-aided analysis of the failure tolerance of an aircraft system, using critical event charts.
Loganathan, Karthikeyan; Srivastava, Ayush; Venditti, Paul; Brahmavar, Ramesh Pai; Banerjee, Goutam; McCarthy, Kevin Thomas, Systems and methods for rule-based control system reliability.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.