At the 2011 AIChE Annual Meeting, Prof. West presented work on multi-scale multi-physics modeling of deposit formation in diesel engine fuel injector nozzles. As well as providing insight into the specific problem of deposit formation in diesel injectors, this work demonstrates an approach suitable for other multi-scale models of chemical engineering systems.
Formation of deposits inside the narrow nozzles of diesel engine fuel injectors disturbs fuel flow and adversely affects engine efficiency and emissions. It is thought that the deposit is formed by oxidation of a thin film of diesel on the nozzle wall. Fuels and additives used to mitigate this problem are currently evaluated using very expensive tests in running engines. Modeling insights into this problem could therefore greatly reduce the cost of fuel and engine development.
The system comprises several different phenomena which span multiple timescales. The free radical chemistry responsible for the deposit precursors occurs on a microsecond timescale while the fuel injections every tens of milliseconds cause washing and replenishment of the fuel film. The deposit itself accumulates over thousands of injection events which occur over hundreds of hours. The challenge lies in addressing the many physical and chemical phenomena in a way that is computationally efficient and gives meaningful results.
We start with a detailed kinetic model for the autoxidation of a diesel fuel surrogate mixture, consisting of hundreds of intermediate species and thousands of elementary reactions, built using Reaction Mechanism Generator (RMG) software. Thermochemical, transport, and solubility parameters are estimated for each of the intermediate species based on their molecular structure. The kinetic model is embedded in a coupled model of evaporation, chemical reaction, phase-separation, and washing, and solved for multiple injection cycles using PyDAS, a new Python wrapper for the stiff DAE solver DASSL.
A model such as this necessarily contains many uncertain and variable parameters. It is useful to identify which of these parameters are most important. To do this we use a modified version of the Morris method to find the global sensitivity of the overall model with respect to each parameter. This has been developed into a Python framework that makes full use of a parallel computing environment and is simple to apply to other models written in the Python programming language.