| A fast and reliable prediction of combustion timing is a prerequisite for developing control strategy in combustion devices such as internal combustion engines, boilers, and gas turbines. For Homogeneous Charge Compression Ignition (HCCI) engine, which is a desirable compromise between spark ignition engine and compression ignition engine, controlling the initiation of combustion remains as a challenge for practical applications. A fast and accurate model for the Start-Of-Combustion (SOC) is required for developing control strategies of HCCI to maximize its many advantageous features. One way of determining SOC is to use Well Mixed Reactor (WMR) model. The main disadvantage of this method is that it is time consuming to compute SOC since the calculation includes hundreds of reaction steps in the chemical mechanism. One alternative is to use a semi-empirical ignition delay integral relation that requires ignition delays as function of engine conditions during the compression stroke. In this study, the idea of training Artificial Neural Networks (ANN) for ignition delay and coupling ANN with a semi-empirical model are explored to provide a fast and reliable model, ANN Combined Ignition Delay (ACID) model, for SOC. Through extensive comparisons, this model predicts SOC in good agreement with those obtained from a well-mixed reactor model using detailed mechanisms. The CPU time for each run takes about 20--30 ms on a PC. The proposed model is potentially promising for use in real-time dynamic control of HCCI engine combustion. As a possible application, the ACID model is integrated with KIVA-3V for efficient calculation. The results from the integrated model agree well with those from the experiments while not compromising its speedy calculation capability. |
