Artificial Neural Networks for Prediction of Knock Events
Downsized turbocharged engines have been increasingly popular in modern light-duty vehicles due to their fuel efficiency benefits. However, high power density in such engines is achieved thanks to high in-cylinder pressure-temperature conditions that increase knock propensity. Control strategies could be used to extend the knock limit if an accurate prediction of knock events were possible. Although knock modeling has been investigated with 3D computational fluid dynamics (CFD) simulations, such models are computationally expensive and cannot be executed in real-time for cycle-to-cycle control purposes. Advances in data analytics and machine learning, however, have enabled the development of real-time executable computer models with different levels of complexity. In this study, artificial neural networks (ANN) were used to develop a predictive model for knock events using the in-cylinder pressure data recorded before knock onset.