Pulsed jet phase-averaged flow field estimation based on neural network approach

Abstract

Single hot-wire velocity measurements have been conducted along a three-dimensional measurement grid to capture the flow-field induced by a 45°inclined slotted pulsed jet. Based on the periodic behavior of the flow, two different estimation methods have been implemented. The first one, considered as the reference base-line, is the conditional approach which consists in the redistribution of the experimental data into spaceand time-resolved three-dimensional velocity fields. The second one uses a neural network to estimate 3D velocity fields given spatial coordinates and time. This paper compares the two methods for a complete flow-field estimation based on hot-wire measurements. Results suggest that the neural network is tailored to capture the phase-averaged dynamic response of the jet induced by the actuator, and identify the coherent structures in the flow field. Interesting performances are also observed when degrading the learning database, meaning that neural networks can be used to drastically improve the temporal or spatial resolution of a flow field estimation compared to the experimental data resolution.

Review of Visual Measurement Methods for Metal Vaporization Processes in Laser Powder Bed Fusion

“…Therefore, high-speed schlieren imaging is introduced to observe the process of evaporating metal. Deep learning has also been applied to image analysis of schlieren imaging systems, where neural networks can effectively capture flow structure features, such as excitation and vortices [ 148 , 149 , 150 ], and extract data information about the flow that can also be used for prediction [ 151 ] and reconstruction [ 152 , 153 ]. To understand how the melt pool and vapor plume interact during the laser and powder interaction, I. Bithara et al [ 76 ] coupled the melt pool and plume dynamics by combining the high-speed schlieren imaging technique and in-situ X-ray method to correlate the vapor plume generated by the interaction of the laser and metal powder with the keyhole it creates in the melt pool, and judged the stability of the melt pool by the morphology of the vapor plume.…”

Methods for Panoramic Visualization and Digital Analysis of Thermophysical Flow Fields. A Review.

“…For example, the authors of [56] have successfully applied a neural network to classify wake vortexes behind a wing profile. Neural networks are beginning to be used to predict [57] or reconstruct [58] flow dynamics. New physics-based methods for calculating the loss function [58].…”

“…Neural networks are beginning to be used to predict [57] or reconstruct [58] flow dynamics. New physics-based methods for calculating the loss function [58]. Deep machine learning allows simulation of turbulence and other largedimensional gas-dynamic systems [59].…”

Animations Analysis in Experimental Fluid Dynamics

“…So far, very few papers have been devoted to this problem, but their number is growing rapidly. Neural networks can effectively capture gas flow structures on large datasets [8], predict [9] and reconstruct [10] flow development using Image Retrieval, Template Matching, Parameters Regression, Spatiotemporal Prediction and other techniques. Deep learning may be used to model high-dimensional gas-dynamic systems such as turbulence [11].…”