Aidong Yang, Lars v.Wedel, Wolfgang Marquardt:
Developing advanced computing facilities to improve robustness and efficiency of physical property calculations
ECCE 3, 3rd European Congress of Chemical Engineering, Nürnberg, Germany, 26-28.6.2001
To support various kinds of chemical engineering activities, physical properties of pure components and mixtures are very frequently calculated, and the performance of physical property packages has significant impact on higher level algorithms such as simulation and optimization. Particularly, it has been recognized that the robustness and the efficiency of thermo calculations are just as critical as accuracy. However, compared to the aspect of accuracy for which a lot of work has been done including e.g. proposing better models and producing more accurate data, the former two aspects have gained much less attention. Although a few relevant issues such as effective use of a cubic equation of state and application of local thermodynamic models have been studied in the past, there is still a lack of systematic work towards improving the robustness and efficiency of thermo calculations. To address this problem, in our work a computational framework containing a set of so-called advanced computing facilities (ACFs) is developed. Based on a systematic analysis, the failures in handling invalid physical conditions, the discontinuities caused by switching different thermo models, and the trivial roots and singularities that occurs in the use of cubic EOS have been identified as the most typical issues that reduce the robustness of the thermo calculations. Satisfactory strategies, then, have been proposed to resolve these issues. Regarding the efficiency aspect, both local thermo models and a "cache" mechanism have been considered for speeding up complex thermodynamic calculations. Especially, the techniques for determining local model's formula and for initializing/updating the parameters have been carefully selected in order to meet requirements in different situations. With the above solutions, the calculation of a physical property can always be accomplished with results acceptable by the higher level algorithms and as quickly as possible, no matter what the property is, which region the physical condition is located in and what models are employed. The above individual solutions are developed into software facilities in an object-oriented approach. The solution to each issue has been implemented as a facility class, and the overall control to the processing of a calculation request is handled by a dispatcher, which, after analyzing the calculation request, puts the individual facilities in order to get the request fulfilled with enhanced robustness and efficiency. Thus, this work results in a set of ACF which can either be used individually for addressing individual problems, or be applied all together with the dispatcher to provide a complete solution. Since the selection of problem solving strategies as well as the design of these software components are performed such that they are all independent of any specific thermodynamic packages, each software component and the whole framework can support any package that happens to be available to a user. The ACFs have been developed within a component-based open thermo server. Together with other functional subsystems in the thermo server, they have been tested and applied by several applications.