LIFT: Learning fault trees from observational data

Abstract

Industries with safety-critical systems increasingly collect data on events occurring at the level of system components, thus capturing instances of system failure or malfunction. With data availability, it becomes possible to automatically learn a model describing the failure modes of the system, i.e., how the states of individual components combine to cause a system failure. We present LIFT, a machine learning method for static fault trees directly out of observational datasets. The fault trees model probabilistic causal chains of events ending in a global system failure. Our method makes use of the Mantel-Haenszel statistical test to narrow down possible causal relationships between events. We evaluate LIFT with synthetic case studies, show how its performance varies with the quality of the data, and discuss practical variants of LIFT.