After demonstrated some cases,
he mentioned traditional digital twin depended on simulations to approximate
the system, but it is not possible to simulate complex system. So he proposed
the Deep Digital Twins (DDT) and gave new definitions below.
1.
An
implicit physics model of an asset learned from healthy asset data, requiring
no explicit physics knowledge.
2.
A
digital representation from which sensor values can be sampled under both
stationary and non-stationary operational settings such as rotational speed,
throttle and load.
3.
A
data driven model which does not require any asset specific feature engineering
4.
A
probabilistic model which is able to automatically produce a health indicator
which is a metric of the deviation from the healthy asset data.
Finally, he showed some application of DDT such as gearbox and Commercial Modular Aero-Propulsion System Simulation (CMAPSS). Lastly, Mr. Khemani concluded that digital twins demonstrated for the prediction of remaining useful life, reliability and functional safety for different products. Concept of deep digital twins (DDT) worked with scenarios where digital twin modeling was infeasible. Reinforcement learning was introduced to digital twin evolving within the system lifetime.
Reference:
CAiRS
- https://www.cairs.hk/view/index.php
沒有留言:
發佈留言