Probabilistic Risk Assessment of Multi-State
Systems Based on Bayesian Networks
Jie Cao *, Baoqun Yin *, Xiaonong Lu**
*Department of Automation, University of Science and Technology of China, Hefei, China
** School of Management, Hefei University of Technology, Hefei, China
caojie92@mail.ustc.edu.cn, bqyin@ustc.edu.cn, lxn520@mail.ustc.edu.cn
Abstract—Probabilistic approaches are common in the risk
assessment of complex engineering systems. Although
conventional methods such as fault tree (FT) have been used
effectively in probabilistic risk assessment (PRA), they are only
suitable to binary-state systems. As an extension of FT, multi-
state fault tree (MSFT) is a good way in the modelling of multi-
state systems, but it suffers severely limitation of efficient
analysis and assess for systems risk, which is of great significance
in PRA. Due to the difficulty of risk assessment in the multi-state
systems, a new method based on Bayesian networks (BNs) is
proposed. The BN model is constructed by converting MSFT and
logic operators of the system through a mapping algorithm. Then
the calculations of consequence probability and importance
degree for each component are proposed based on Bayesian
inference. Also, diagnose of failure system states can be achieved
by posterior inference of BN. Finally, an example is illustrated to
verify the effectiveness and feasibility of the proposed method.
Keywords—risk assessment, multi-state systems, MSFT, BNs,
consequence probability, importance degree
I. INTRODUCTION
Engineering decisions often involve probabilistic risk
assessment of the system’s working status under evolving and
uncertain information. PRA is widely used as it can not only
estimates the likelihood and consequences of accidents, but
also categorizes their potential scenarios [1]. Traditional PRA
analysis methods, such as Fault Tree Analysis (FTA) and
Event Tree Analysis (ETA), are based on the assumption that
the events involved are binary events (Normal or Fault).
However, in many real situations, both the system and its
components are capable of being in different failure modes [2].
Furthermore, any system and its component with one failure
mode can be divided into lots of failure states according to the
degree of the failure, and binary-state system can be treated as
a special case of the multi-state system.
Recently, significant amounts of research efforts have been
devoted to deal with the PRA analysis of the multi-state
systems. In [3], a multi-state fault tree analysis (MFTA) model
is proposed to assess the reliability of the complex system, and
the MFTA model is constructed based on multi-state block
diagrams using binary variables. Xue [4] introduced the
discrete function theory into multi-state system analysis, and
adapted two methods (inclusion-exclusion and enumeration)
to calculate the probability of each state. However, these
above methods are only suitable within the scope of the
coherent systems. Huang [5] proposed a generic method to
construct the multi-state fault tree through reliability block
diagrams and decision tables. However, with the growth of the
system scale and types of states, analysing the MSFT based
risk assessment becomes a rather complex problem. Therefore,
an efficient general method is need to efficiently assess the
risk of the multi-state systems.
Currently, Bayesian networks have been proposed in many
researches as an ideal method for risk assessment and failure
analysis in complex systems [6]. The BN is a graphical model
consisting of nodes and directed links, which respectively
represent random variables and their probabilistic
dependencies. The variables may represent the states of the
components of a system, or their capacities and demands. Lee
et al. [7] presented a large engineering project risk
management procedure by using BN, and then identified the
major risk items which affected system performance. The
main feature of BN is that, by entering evidence on one or
more variables, e.g., the observed states, capacities or
demands of a subset of the components, the information
propagates throughout the network and updates distributions
of other random variables, e.g., the states of other components
or the system state, in accordance with the Bayes’ rule, which
provides an effective way of risk analysis in PRA. Several
research have aimed at exploring the capabilities of BN in the
risk assessment of binary-state systems [8, 9], which
considered mapping fault tree into BN to assess the binary-
state systems. However, these methods have not considered
the situation of multi-state systems.
In this paper, we try to use the Bayesian network for
probability risk assessment of the multi-state systems. We first
introduce a brief description of the model of BN. Then,
according to the MSFT structure and logical operators of the
multi-state system, we proposed a mapping algorithm to
convert MSFT to BN model. We use the BN inference
technique to calculate the probability distribution of the
systems state, and the performance of each component’s
contribution to a failure system statue is presented by the
importance degree. Also, with the posterior inference of BN,
diagnostic analysis is proposed to find out the suspected cause
for a failure system. At last, through an application of the risk
Jan. 31 ~ Feb. 3, 2016 ICACT2016