Probability Analysis

Probability Types

Various probability types and distributions are accepted as described in the Open-PSA Model Exchange Format [MEF], for example, constant values, exponential with two or four parameters, and uniform, normal, log-normal distributions.

Bellow is a brief description. For more information, please take a look at the [MEF] format documentation.

P-model
The probability of an event occurring when the time to failure is unknown or unpredictable.
Lambda-model or Exponential with Two Parameters

The probability that a primary event will occur within given time (\(t\)). Appropriate for events within systems that are continuously operating and have a known probability of failure during a unit time period (\(\lambda\)).

\[P = 1-e^{-\lambda \cdot t}\]

For small lambda and time, the approximation is

\[P \approx \lambda \cdot t\]

Probability Calculations

Probability calculation algorithms assume independence of basic events in the fault tree. The dependence can be communicated with common cause groups.

The Exact Probability Calculation

Since the resultant sets may neither be mutually exclusive nor independent, direct use of the sets’ total probabilities may be inaccurate [WakXX]. The exact probability calculation is achieved with Binary Decision Diagram ([BDD]) based algorithms [DR01]. This approach does not require calculation of products. As long as a fault tree ([PDAG]) can be converted into BDD, the calculation of its probability is linear in the size of BDD.

The Approximate Probability Calculation

Approximate calculations are implemented to reduce the calculation time. However, the users must be aware of the limitations and inaccuracies of approximations [WakXX]. If approximate calculations yield probability values above one, the result is adjusted to one. In this special case, appropriate warnings are given in the final report.

The Rare-Event Approximation

Given that the probabilities of events are very small value less than 0.1, only the first series in the Sylvester-Poincaré formula may be used as a conservative (upper-bound) approximation; that is, the total probability is the sum of all probabilities of minimal cut sets. Ideally, this approximation gives good results for independent minimal cut sets with very low probabilities. However, if the cut set probabilities are high, the total probability may exceed 1.

This is the default quantitative approximation for analysis algorithms that inherently run with approximations (MOCUS and ZBDD).

The Min-Cut-Upper-Bound (MCUB) Approximation

This method calculates the total probability by subtracting the probability of all minimal cut sets’ being successful from 1; thus, the total probability never exceeds 1. Non-independence of the minimal cut sets introduce the major discrepancy for this technique. Moreover, the MCUB approximation provides non-conservative estimation for non-coherent trees containing NOT logic [WakXX].

Importance Analysis

Importance analysis is performed for basic events in a fault tree [DR01]. The same configurations are used as for probability analysis. The analysis is performed by request with probability data. The following importance factors are calculated:

  • Fussel-Vesely Diagnosis Importance Factor (DIF)
  • Birnbaum Marginal Importance Factor (MIF)
  • Critical Importance Factor (CIF)
  • Risk Reduction Worth (RRW)
  • Risk Achievement Worth (RAW)

The short description and interpretation of the factors can be found in Theory.

Alongside the importance factors, the analysis provides the probabilities of events and their number of occurrences in products.

Safety Integrity Levels

[IEC_61508] standard metrics and Safety Integrity Levels [SIL] are approximated with quantitative analysis on fault trees [DR05]. Time fractions spent in every SIL bucket for PFD and PFH is reported with a histogram, as suggested by [DRS08]. Note that these computations require probability analysis over a period of time.

Warning

The current implementation for the PFH calculation is simplistic, resulting in potentially less accurate values than the more rigorous approaches suggested in the above papers. The PFH results may become more inaccurate with testable, repairable, and/or non-continuously-operated components. At best, the approximate value is expected to be of the same magnitude as the real value, which puts the approximation into the same Safety Integrity Level.