Desenvolvimento de uma ferramenta para modelagem virtual de um motor de combustão interna ciclo Otto usando redes neurais

Abstract

The design, build and test of spark ignition engine has an important stage to be carry out on the test stands which allows to test performance, efficiency, components and calibration in the workflow of the engine development cycle. There are a certain level of variables to be measured and calibrated that increase the complexity of tasks to do, demanding for a high time and costs on bench testing. The virtual modeling of this complexity thermal machine might be done using the artificial neural network through instrumentation for acquire several physical variables and then evaluation several possible architectures. Basically, we capture a dataset that we presented to our network. Therefore, we developed a process that allows we identify and model the engine and put on in a virtual model based on ANN. However, there are a bunch of possible architectures available and through a genetic algorithm for optimization we can set up the best ANN architecture for that specific engine to be modeled. Our method can reduce the time and costs in bench testing taken to virtual environment and consequently reducing the time of development and time to market. Finally, we proof the efficiency of our approach in the experimental results of a specific project.