4:10 pm Presentation
“On the Mechanism that Sustains Intermittent Attached Cavitation Inception in a Laminar Boundary Layers”
Presented by OMRI RAM (Adviser: Prof. Katz)
Cavitation inception occurs in locations of the flow where the pressure drops just below the vapor pressure, happens on extremely fast times scales. Even though it has been extensively studied over the last five decades, both fundamental mechanisms of inception and its dependence on the flow characteristics remain unclear. In this study, water tunnel experiments involving high-speed microscopic imaging focus on the processes of cavitation inception in a flow with an attached boundary layer downstream of the minimum pressure point. A cavitation patch is formed as free stream nuclei approach the low-pressure region, expand rapidly and attach to the surface. If the minimum pressure is kept close to the vapor pressure these cavities collapse in milliseconds, leaving a cloud of microbubbles small than 50 micrometers downstream of the minimum pressure point. Some of these bubbles migrate randomly near the surface, presumably under the balancing influence of drag that pushes them downstream and local adverse pressure gradient that pulls them upstream. Hence, the bubbles are maintained near the surface, even without flow separation. Of those that migrate upstream, a fraction grows to form another intermittent cavity that generates new microbubbles. Hence, once the initial attachment occurs, subsequent attached cavities appear at a frequency that is much higher than that associated with the freestream nuclei. High-resolution PIV measurements confirm that these bubbles move in a low-velocity region that forms close to the minimum pressure point due to the adverse pressure gradients. It appears that these bubbles size is upper bounded by the height of this region and lower bounded by the volume required to generate significant pressure force upstream.
4:35 pm Presentation
“A Data-Based Wall Model for Large Eddy Simulations”
Presented by YUE HAO (Advisers: Profs. Meneveau & Zaki)
An all connected neural network was adopted to develop a wall model for large eddy simulations (LES). Spatially filtered data from the JHTDB channel flow at Re_tau=1000 were used for training and testing. The network was trained to predict wall shear stress from LES information at the first grid point above the wall. A priori tests were performed to determine the accuracy of the wall model by comparing the predicted wall shear stress to its filtered value from the database. Input combinations of u and y were tested and yielded comparable performance to the equilibrium wall model. The modified input variables u/y and log(y/y_0)/u that were proposed by Yang et al. (2019) were effective for extrapolation, which is critical for wall modeling. Multiple terms in the integrated momentum equation were tested as additional inputs, but no obvious improvement was achieved because the terms are poorly correlated with wall shear stress. Performance of the wall model was, however, improved by expanding the spatial support of the inputs. When non-local flow information was included, the predicted wall shear stress was better correlated with DNS than the equilibrium wall model. The physical reason is the inclination of the elongated structures in the near-wall region.