Bayesian networks in reliability

Information on the reliability of engineering systems under changing environment is necessary for management and operation. The Bayesian network (BN) can be used as a tool for updating the reliability in near real-time. We work towards the application of BN to complex engineering systems, by addressing the following challenges.

Efficient BN structures:
It is straightforward to represent a reliability problem in a BN structure. However, for systems with many components or basic random variables, computational issues make more efficient representations necessary.

Inference algorithms:
Exact inference algorithms exist for general discrete BNs, which however are not (directly) applicable for reliability BNs with many components. For hybrid BNs, consisting of both discrete and continuous random variables, exact inference is only possible in a number of special cases. We aim at developing (approximate) sampling based inference algorithms that are capable of dealing with both hybrid BNs and complex reliability problems.

Researchers

Collaborations

Structural/System Reliability Group (https://systemreliability.wordpress.com/), Seoul National University, South Korea

Funding

Deutsche Forschungsgemeinschaft (DFG)

Selected Publications

Zwirglmaier K., Straub D., Groth K. (2016) Capturing cognitive causal paths in human reliability analysis with Bayesian network models. Submitted to Reliability Engineering and System Safety
Zwirglmaier K., Straub, D. (2015) A discretization procedure for rare events in Bayesian networks. Submitted to Reliability Engineering and System Safety
Špačková O., Straub D. (2013). Dynamic Bayesian networks for probabilistic modeling of tunnel excavation processes. Computer-Aided Civil and Infrastructure Engineering, 28(1): 1-21.
Bensi M.T., Der Kiureghian A., Straub D. (2013). Efficient Bayesian network modeling of systems. Reliability Engineering & System Safety, 112: 200-213.