White-box/Grey-box/Black-box models for Digital Twins

There are techniques in AI/ML to solve all of the above challenges. For example Recurrent Neural Network (RNN) can be used to process the time evolution data to build a classification model. As the system's behavior under each type of fault can be different , now with the Fault classification model using RNN whenever a fault occurs and the subsequent pattern of data helps in determining the type of fault. This is also know as Fault Diagnosis.

When twin models are developed using underlying physics, such models are known as white-box model. For example when computational fluid dynamics (CFD) is used to model the flow behavior , it uses fundamental governing equations of fluid-flow . Hence CFD modeling is a white-box model. If the models are developed only by processing the data then the model is know as black-box model. Many system identification models using principles of statistical methods & data analysis. While white-box models tends to more versatile & accurate, but they are relatively difficult to develop and solve. At the same time, for a particular use case and limited range of use, a black-box can be developed with considerable lesser effort.

Grey box model a hybrid approach is used by combining the governing equations with data. For example the special forms of physics informed Artificial Neural Network development, the NN are developed and configured considering their ability to solve particular type of partial differential equations (PDEs) and system of equations. Such grey box models provide good combination of versatility and accuracy.