Managing Testing Complexity in Dynamically Adaptive Systems: A Model-Driven Approach

Abstract

Autonomous systems are increasingly conceived as a means to allow operation in changeable or poorly understood environments. However, granting a system autonomy over its operation removes the ability of the developer to be completely sure of the system's behaviour under all operating contexts. This combination of environmental and behavioural uncertainty makes the achievement of assurance through testing very problematic. This paper focuses on a class of system, called an m-DAS, that uses run-time models to drive run-time adaptations in changing environmental conditions. We propose a testing approach which is itself model-driven, using model analysis to significantly reduce the set of test cases needed to test for emergent behaviour. Limited testing resources may therefore be prioritised for the most likely scenarios in which emergent behaviour may be observed.