Sensors constitute information providers which are subject to imperfections and assessing the quality of their outputs, in particular the trust that can be put in them, is a crucial task. Indeed, timely recognising a low-trust sensor output can greatly improve the decision making process at the fusion level, help solving safety issues and avoiding expensive operations such as either unnecessary or delayed maintenance. In this framework, this paper considers the question of trust dynamics, i.e. its temporal evolution with respect to the information flow. The goal is to increase the user understanding of the trust computation model, as well as to give hints about how to refine the model and set its parameters according to specific needs. Considering a trust computation model based on three dimensions, namely reliability, likelihood and credibility, the paper proposes a protocol for the evaluation of the scoring method, in the case when no ground truth is available, using realistic simulated data to analyse the trust evolution at the local level of a single sensor. After a visual and formal analysis, the scoring method is applied to real data at a global level to observe interactions and dependencies among multiple sensors.