A multi-approach algorithm for enabling efficient application of very large, highly detailed reaction mechanisms in multidimensional HCCI engine simulations

摘要

Predictive modeling of Homogeneous Charge Compression Ignition (HCCI) engine performance requires application of detailed reaction mechanisms. However, the computational cost of applying large mechanisms is prohibitive for practical multi-dimensional HCCI engine simulations. In the present paper, we describe a hybrid approach that combines on-the-fly mechanism reduction based on Directed Relation Graph with Error Propagation (DRGEP) method, Graphics Processing Unit (GPU) computing, and Adaptive Multi-grid Chemistry (AMC) model approaches to enable efficient application of highly detailed reaction mechanisms in multi-dimensional HCCI engine simulations. We demonstrate the performance of this hybrid approach using a very large, detailed biodiesel surrogate fuel mechanism which contains 2877 species and 8555 reactions (the methyl decanoate (MD) mechanism of Herbinet, Pitz and Westbrook). Up to 300-fold speed-up was achieved in single-zone HCCI engine simulations using on-the-fly mechanism reduction coupled with GPU computing, relative to the usual approach of using a single full-chemistry model without using a GPU. In addition, the simulation results using the accelerated chemistry solver agree with those of the full chemistry solver excellently. The AMC approach was then incorporated into this hybrid framework, for further acceleration in multi-dimensional simulations by grouping thermodynamically-similar computational cells. Combining these three techniques, we show that a 2-D HCCI engine simulation of the MD mechanism with either homogenous charge or stratified charge can be completed within only thirteen hours on a personal computer (PC) equipped with a Tesla C2050 GPU card - less time than is required to complete a single-zone (0-D) simulation on the same PC using the traditional CPU-only full chemistry approach. These results demonstrate the cumulative benefits attainable by combing multiple available methodologies for enhancing combustion modeling efficiency. The algorithm developed in this work enables incorporation of significantly greater detail (simultaneous application of higher-fidelity transport and chemistry models without prohibitive costs) than was previously possible for modeling HCCI engines.