Dynamic tearing: efficiency enhancement of tearing methods by consideration of solvability criteria during runtime

Abstract

Common tearing methods try to find static tearing variables. This means that selected tearing variables are used for the entire simulation, which also means that all inner equations are used for the entire simulation. Hence, the tearing method sets up the tearing system in a way, that there are no restrictions on the domain of the inner equations. In general, this leads to bigger tearing sets. This paper presents an extension of common tearing methods that generates another tearing set in addition. The additional set has fewer tearing variables, which means that it should be more efficient in general. However, the additional set has some restrictions on its domain of definition. That is why common approaches would not even create it and why it may not be used for the entire simulation. Hence, its domain needs to be analysed during simulation to validate if the smaller set is defined on the current domain. If that is the case the smaller set is used for the calculation, otherwise the original set is used. This paper shows how this additional tearing set can be generated. It is also demonstrated how the domain can be monitored during runtime in order to make the switching process efficient. Results using a prototype implementation in OpenModelica are analysed to show the benefits of this method.