Bi-Fidelity Neural Network Model for Multi-fuel Capable Internal Combustion Engines

Energy Assisted Compression Ignition (EACI) has garnered significant attention in recent years due to its potential applications in aviation propulsion. To enhance combustion and facilitate ignition with low cetane number (CN) jet fuels, a ceramic-based glow plug is utilized as an ignition assistant (IA). Control systems are essential for internal combustion (IC) engines to optimize performance, and ensure durability. They manage various parameters like fuel injection (timing, duration, and amount), ignition timing, and after-treatment systems, making the engine efficient, responsive, and environmentally friendly. A reliable engine control system requires a substantial amount of pressure-crank-angle signals spread across a wide range of operating conditions to effectively train the controller software. For engines operating under a wide range of conditions and control parameters, traditional data collection through experimental methods is both time-consuming and expensive, especially under challenging conditions, such as low CNs. Computational Fluid Dynamics (CFD) can be supplemented to the experimental data. However, the CFD simulation time to generate the wide range of data points for the control systems requires a huge number of CPU hours. In this study, we propose a bi-fidelity regression problem, where we integrate the high-fidelity, expensive experimental data with “low-fidelity”, cost-effective CFD data. We employed an artificial neural network (ANN) based methodology for multi-fidelity regression, utilizing two separate ANNs for each level of fidelity. This approach demonstrated effectiveness by achieving physically acceptable error levels in pressure prediction across various operating conditions and engine control parameters, highlighting its potential for efficient and accurate engine control system training in complex IC engine operations.