DNV GL proposes the âProbabilistic Digital Twinâ? (PDT) to close the gap between digital twins used increasingly by operators to manage the performance of their assets and risk analysis still largely conducted manually before assets enter service.
A digital twin is a digital âmirrorâ? of a physical asset, including models of its structure and dynamics which are updated through a combination of multiple data sources. They bring significant benefits for data management and decision making, providing a consistent, accurate single source of information.
Risk models are rarely brought forward into operations they typically exist separately within engineering, operations, and health and safety disciplines and are mostly used in desk studies, based on analyzing historical data and offering only a static picture of potential risks.
In reality, the risk is dynamic, varying in time with operational conditions and the condition of the asset, but this is not captured by current risk models which are seldom updated and lack real-time and prediction capabilities.
A PDT may include reliability and degradation models to predict the remaining lifetime of mechanical components. However, it is more than a predictive maintenance tool. Risk is not only about component failures, but also about exposure to hazards and how the asset is operated.
The main elements which distinguish a probabilistic digital twin from traditional digital twins are Probabilistic degradation and failure models, reflecting uncertainty and variability of conditions and processes that affect performance and lead to failures. Logic and relational models, relating performance variables to failures and loss events. Surrogate models, approximating heavier simulation models, allowing fast queries and enabling propagation of uncertainty and model coupling.
