Model-based ai approach for knock control in spark ignition engine by vibration block modeling

Abstract

Engine Knock poses a signifcant threat to the integrity and performance of spark ignition engines, leading to decreased efciency, power, and overall performance. Traditional control strategies utilize knock sensors to identify and mitigate these adverse efects, although these may not always provide accurate or timely responses. This presents an innovative model-based approach to knock control, utilizing Artifcial Intelligence (AI) techniques. Specifcally, we propose a time-series Artifcial Neural Network (ANN) model designed to simulate SI Engine Cylinder block vibrations. This study investigates transformer models, Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) models. These models are trained using vibrational data from engine blocks, thereby developing a comprehensive understanding of the unique behaviors of the engine. Among the models, the Attention LSTM model showed slightly superior performance, outperforming the other two models across the full range of engine vibration conditions. Despite promising results, further refnement and tuning of these models are required for real-world implementation. The proposed AI model ofers a promising direction for advancements in knock control technology, with potential implications for enhancing engine durability, performance, and fuel efciency.