ABSTRACT
We present a framework for the automated generation of fault trees from models of real-world process workflows, expressed in a formalised subset of the popular Business Process Modelling and Notation (BPMN) language. To capture uncertainty and unreliability in workflows, we extend this formalism with probabilistic non-deterministic branching. We present an algorithm that allows for exhaustive generation of possible error states that could arise in execution of the model, where the generated error states allow for both fail-stop behaviour and continued system execution. We employ stochastic model checking to calculate the probabilities of reaching each non-error system state. Each generated error state is assigned a variable indicating its individual probability of occurrence. Our method can determine the probability of combined faults occurring, while accounting for the basic probabilistic structure of the system being modelled. From these calculations, a comprehensive fault tree is generated. Further, we show that annotating the model with rewards (data) allows the expected mean values of reward structures to be calculated at points of failure.
