Probabilistic Risk Assessment and Risk Management

What is Probabilistic Risk Assessment?

Probabilistic Risk Assessment is a method of systematically examining complex technical systems to measure both the likelihood that an accident will occur (probability) and the level of damage or loss that will result (consequences).
A PRA examines the reactions of a system to variations in its normal operations or environment.
Both the spectrum of potential damage states and the frequency with which each state occurs are examined by PRA and both are treated as uncertain variables.
PRA can basically be thought of as a thought process which develops integrated, systematic, and quantitative information suitable for aiding risk management decision-making in the face of uncertainties.

Basic PRA Characteristics

PRA is hierarchical. Events of a problem or situation are separated into groups based on common properties. Using even higher level common properties the events can be re-grouped.
PRA is scenario based. It uses a string of events to diagram occurrences from an initial problem to an end result. Scenarios are diagrammed using master logic diagrams (MLD's), event sequence diagrams, event trees, and fault trees.
Uncertainties and variabilities associated with the modeling of the physical and chemical aspects of events, the parameters of the models, and the frequency of events are explicitly identified by PRA. Bayesian analysis is used to combine information from analysis, databases, testing, and judgment.
Monte Carlo simulation methods are used to propagate uncertainties and variabilities in the models, event trees, and fault trees.
A PRA defines the damage levels and the frequency of obtaining each state using a system of algebraic equations. These equations are produced using various diagrammatic and logical tools such as event trees and fault trees.

Scenarios in Risk Assessment

Scenarios are generally strings of events that lead to some kind of conclusion. The starting point for a scenario is called the initiating event. An initiating event is a problem in the system that causes an alteration in the normal operation. A scenario finishes with an end state or a damage state. An example of a damage state would be the loss of the Shuttle Orbiter. A damage state is defined by the decision maker. Between the initiating event and the damage state are pivotal events which determine whether the given damage state is reached as a result of the initiating event. Pivotal events may be protective, mitigative, aggravative, or benign.
Scenarios may be documented by a variety of different diagrams. In safety and reliability risk assessments the most common diagrams used are event trees, fault trees, and functional event sequence diagrams.

Master Logic Diagram

A master logic diagram is used to depict an arrangement of initiating events that is reasonably complete. It would be quite impractical to try to completely predict the occurrence of system perturbations in every detail. For this reason, analysts who wish to predict the relevant events use a functional categorization of perturbations to the system which lead to a component characterization of each function. The top event in a master logic diagram is the damage state such as failure of an entire system. The lower levels of the diagram represent subsystem or component failures that lead to failure of the system.

FESD's, Event Trees, and Fault Trees

Functional event sequence diagrams (FESD) are often used to present an outline of the system response to subsystem or component failures. An FESD is made up of an initiating event, pivotal events, and damage states. The pivotal events depict all the possible occurrences that could arise from the initiating event. An FESD is made using inductive reasoning that means that consecutive events are developed by thinking of the next possible outcome. Each FESD presents a different scenario that is usually converted to an event tree. An event tree, like an FESD is made up of binary outcomes for each event. Event trees are used because it is easier to obtain the needed algebraic equations from an event trees than a FESD. Event trees require probability of occurrence of each event. These probabilities may be developed using a fault tree. In this way event trees and fault trees compliment each other. Together they depict the necessary and sufficient conditions for the occurrence of each damage state. As mentioned above they are used to find the needed algebraic equations. A fault tree uses deductive reasoning, which means that the lower events are found by thinking of all possible ways in which the top event could have occurred. Using fault trees and event trees together is a more complete way of documenting scenarios than using either one individually.

Uncertainties and Variabilities

Because probabilistic risk assessment is made up of very complex scenarios it is necessary to account for variations in physical processes and uncertainties in knowledge. Variability refers to changes in the physical process over the period of many similar trials. Uncertainty refers to knowledge of the parameter or variable. Many variables and parameters could be found without any uncertainty if sufficient experimentation could be performed. Unfortunately such experimentation is often unavailable thus the uncertainty of a variable is represented by a probability distribution. Uncertainty will decrease as more knowledge of the parameters is made available. Uncertainties are developed at the lowest level of a risk model. PRA frameworks allow for appropriate treatment of variabilities and uncertainties. Quantification of uncertainties and variabilities of a scenario based risk model allows for identification of the problems most important to risk.