Development of a Simple Model to Predict Spatial Distribution of Cycle-Averaged Wall Heat Flux Using Artificial Neural Networks

The KIVA 3V code has been applied to predict combustion chamber heat flux in an air-cooled utility engine. The KIVA heat flux predictions were compared with experimentally measured data in the same engine over a wide range of operating conditions. The measured data were found to be approximately two times larger than the predicted results, which is attributed to the omission of chemical heat release in the near-wall region for the heat transfer model applied. Modifying the model with a simple scaling factor provided a good comparison with the measured data for the full range of engine load, heat flux sensor location, air-fuel ratio and spark timings tested. The detailed spatially resolved results of the KIVA predictions were then used to develop a simplified model of the combustion chamber temporally integrated heat flux using an artificial neural network (ANN). The input parameters of the ANN model are engine load, air-to-fuel ratio, spark timing and zonal engine wall temperature distribution; and the outputs of the model are the temporally averaged heat flux to each zone. This simplified model provides a coupling between thermal structural finite element analysis and detailed in-cylinder processes with low computational overhead.