Search Results

Accurate modeling of the internal flow and spray characteristics in fuel injectors is a critical aspect of direct injection engine design. However, such high-fidelity computational fluid dynamics (CFD) models are often computationally expensive due to the requirement of resolving fine temporal and spatial scales. This paper addresses the computational bottleneck issue by proposing a machine learning-based emulator framework, which learns efficient surrogate models for spatiotemporal flow distributions relevant for static coupling injection maps, namely total void fraction, velocity, and mass, within a design space of interest. Different design points involving variations of needle lift, fuel viscosity, and level of non-condensable gas in the fuel were explored in this study. An interpretable Bayesian learning strategy was employed to understand the effect of the design parameters on the void fraction fields at the exit of the injector orifice.

Heterogeneous physical systems can often be considered as highly complex, consisting of a large number of subsystems and components, along with the associated interactions and hierarchies amongst them. The simulation of a large-scale, complex system can be computationally expensive and the dynamic interactions may be highly nonlinear. One approach to address these challenges is to increase the computing power or resort to a distributed computing environment. An alternative to improve the simulation computational performance and efficiency is to reduce CPU required time through the application of surrogate models. Surrogate modeling techniques for dynamic simulation models can be developed based on Recurrent Neural Networks (RNN).This study will present a method to improve the overall speed of a multi-physics time-domain simulation of a complex naval system using a surrogate modeling technique.

This paper critically evaluates the use of Neural Networks (NNs) as metamodels for design applications. The specifics of implementing a NN approach are researched and discussed, including the type and architecture appropriate for design-related tasks, the processes of collecting training and validation data, and training the network, resulting in a sound process, which is described. This approach is then contrasted to the Response Surface Methodology (RSM). As illustrative problems, two equations to be approximated and a real-world problem from a Stability and Controls scenario, where it is desirable to predict the static longitudinal stability for a High Speed Civil Transport (HSCT) at takeoff, are presented. This research examines Response Surface Equations (RSEs) as Taylor series approximations, and explains their high performance as a proven approach to approximate functions that are known to be quadratic or near quadratic in nature.

Neural network augmented model inversion control is used to provide a civilian tilt-rotor aircraft with consistent response characteristics throughout its operating envelope, including conversion flight. The implemented response types are Attitude Command Attitude Hold in the longitudinal channel, and Rate Command Attitude Hold about the roll and yaw axes. This article describes the augmentation in the roll channel and the augmentation for the yaw motion including Heading Hold at low airspeeds and automatic Turn Coordination at cruise flight. Conventional methods require extensive gain scheduling with tilt-rotor nacelle angle and airspeed. A control architecture is developed that can alleviate this requirement and thus has the potential to reduce development time. It also facilitates the implementation of desired handling qualities, and permits compensation for partial failures.