Towards Diagnosis Modulo Theories

We introduce a new flexible approach to modelbased diagnosis based on satisfiability modulo theories (SMT) solving. The idea is to translate a diagnosis problem into a SAT problem with additional theory constraints. An entirely discrete model can be translated directly into a SAT representation, but for a hybrid (continuous-discrete) model, additional numerical equalities between terms will be created which would be handled by an additional theory solver for arithmetic. The advantages of the approach are in ease of modelling through support for continuous values; reduced need for monitors to discretise inputs; and flexibility, in that other theories, such as arrays or bit vectors, can also be incorporated into models. In this paper we describe how a diagnosis engine based on this idea can be constructed using an existing SMT solver and demonstrate the effectiveness of the approach, which we call diagnosis modulo theories (DMT), on the ADAPTLITE domain from the DXC competition.