ZWZhi Wu

Zhi Wu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Zhi Wu has a total of 20 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Image-Based Prediction of Lift and Drag Coefficients of a Circular Cylinder Using Convolutional Neural Networks and Flow Visualization

    by Jingwen Hu, Waleed Khan, Zhi Wu

    Force measurement is a fundamental diagnostic technique in experimental fluid mechanics. However, conventional instrumentation, such as mechanical balances and pressure taps, is often intrusive, geometry-dependent, and mechanically constrained. This study proposes a non-contact, image-driven force-estimation framework wherein convolutional neural networks (CNNs) infer the instantaneous lift and drag coefficients of a circular cylinder directly from wake streakline flow-visualization images. A comprehensive hybrid numerical–experimental dataset was established across laminar and turbulent regimes. Navier–Stokes simulations (DNS at Re=150; LES at Re=500 and 3900) were post-processed into synchronized streakline–force pairs, complemented by water-tunnel dye-visualization experiments at Re=500, 8000, and 12,000. Five CNN architectures, including lightweight custom networks and vision backbones (AlexNet, ResNet18), were trained and validated. For numerical datasets, all models show high prediction accuracy in lift and drag coefficients. In simulation-to-experiment transfer at Re=500, the models capture the dominant vortex-shedding Strouhal number within 12.0% (St=0.233 vs. 0.208), while CNNModel_v5 predicts the mean drag coefficient within 0.35%. On experimental datasets, ResNet18 achieved the best predictive performance among the models tested under random splitting. These results indicate that wake flow-visualization images contain rich physical representations for non-intrusive fluid force estimation.

  • Control of flow separation from a curved ramp using multiple unsteady jets

    by Ge Xu, Shengtai He, Zhi Wu, Yu Zhou

    Turbulent boundary layers subjected to an adverse pressure gradient are prone to separation, resulting in increased drag and noise, along with deteriorated aerodynamic efficiency. Active flow control strategies have been shown to effectively suppress flow separation, but often at the cost of significant energy consumption. In this study, an experimental investigation is conducted to evaluate the performance of genetic algorithm-optimized unsteady jets for the control of flow separation from a curved ramp at Reθ = 5700, where Reθ is the momentum thickness-based Reynolds number. Three spanwise slit jets are deployed; a total of nine independent control parameters are optimized, including their driving pressures, duty cycles, excitation frequency, and phase differences. The cost function is designed so that separation suppression and actuation energy are compromised. Particle image velocimetry and a pressure scanner are used to characterize the controlled flows. The optimized control eliminates the recirculation region while reducing energy consumption by 76.9% relative to its steady counterpart. Furthermore, the optimization identifies the key mechanism for enhancing control effectiveness. These results demonstrate that the distributed unsteady jets offer a promising pathway toward highly efficient turbulent flow separation control.

  • Artificial intelligence control of flow separation from a curved ramp

    by Zhi Wu, Ge Xu, Shengtai He, Yu Zhou

    This work aims to control flow separation from a two-dimensional curved ramp. The Reynolds number examined is Reθ = 5700 based on the momentum thickness of the turbulent boundary layer right before the ramp. Three steady jets, blowing tangentially along the ramp from three spanwise slits, are deployed at the most likely flow separation position, upstream and downstream of this position, respectively. Three different control modes are investigated, i.e., a single jet, multiple jets, and genetic algorithm-optimized blowing rates of three jets. The single jet placed at the time-averaged flow separation position is found to be most effective and efficient in eliminating flow separation among the first and second control modes. However, it is the third control mode that may not only eliminate the separation bubble completely but also cut down the energy consumption, by up to 30%, compared to the single jet blowing at the flow separation position. The flow physics underlying the control modes is also discussed.


  • A hybrid artificial intelligence control of a turbulent jet: Reynolds number effect and scaling

    by A.K. Perumal, Z. Wu, D.W. Fan, Y. Zhou

    This work aims to investigate experimentally the effect of the Reynolds numberRe,based on the nozzle diameterD, on jet mixing manipulation using an unsteady radial minijet. A novel artificial intelligence (AI) control system has been developed to manipulate the jet overRe = 5800–40 000. The system may optimize simultaneously the control law and a time-independent parameter, which dictate the actuation ON/OFF states and amplitude, respectively. The control parameters include the mass flow rate, excitation frequency and diameter ratios (Cm,fe/f0andd/D) of the minijet to the main jet as well as the duty cycle (α) of minijet injection. Jet mixing is quantified usingKeandK0, whereKis the decay rate of the jet centreline mean velocity, and subscriptseand 0 denote the manipulated and unforced jets, respectively. It has been found that the maximumKeachievable does not vary withRe.Scaling analysis of the huge volume of experimental data obtained from the AI system reveals that the relationshipKe = g1(Cm,fe/f0,α,d/D, Re, K0) may be reduced toKe/K0 = g2 $(\zeta )$ , whereg1andg2are different functions and the scaling factor $\zeta = ({C_m}/\alpha ){(D/d)^{1 - n}}(1/Re){({f_e}/{f_{e,opt}})^m}$ ,mand 1 − nare the power indices, and subscriptoptdenotes the value at whichKeis maximum. The scaling law is discussed in detail, along with the physical meanings of the dimensionless parametersKe/K0,ζ, $({C_m}/\alpha ){(D/d)^{1 - n}}(1/Re)$ and ${({f_e}/{f_{e,opt}})^m}$ .


  • Modification of flow behind a circular cylinder by steady and time-periodic blowing

    by Zhi Wu, Haecheon Choi

    We apply steady and time-periodic blowing, respectively, to flow over a circular cylinder at a Reynolds number of 21 000 for the modification of flow in the wake. The blowing is applied near the separation point on the upper and lower surfaces from either a slit (uniform blowing) or eight holes (discrete blowing) along the spanwise direction at each surface. For steady blowing, the blowing momentum coefficient ranges from 0.008 to 0.269, and, for time-periodic blowing, it does from 0.013 to 0.162. The changes in the drag are estimated by a momentum theorem with the velocity measurements in the wake. The steady discrete blowing is more effective and efficient for drag reduction than the steady uniform blowing, providing a maximum drag reduction of 38%. Counter-rotating vortices induced by the optimal discrete blowing change the Kármán vortex cores into three-dimensional vortices and weaken their strength. On the other hand, steady uniform blowing significantly increases the wake width and leads to drag increase. With time-periodic blowing, the maximum drag reductions by uniform and discrete blowing are 68% and 28%, respectively, indicating that the time-periodic uniform blowing is more effective than the steady discrete blowing. For the time-periodic blowing, two different vortex lock-on phenomena (symmetric and asymmetric vortex shedding in the wake) are found, and they lead to a significant drag decrease and increase, respectively.


  • A predictive model of the drag coefficient of a circular cylinder

    by Dohyun Jin, Zhi Wu, Haecheon Choi

    We suggest a predictive model of the drag coefficient of a circular cylinder using the information of the mean streamwise velocity and Reynolds normal stresses at a streamwise location in the wake. The boundary layer approximation in the wake and Bernoulli equation in the inviscid region are used to model the contribution from the pressure distribution at the streamwise measurement location. To examine the accuracy of the present model, unsteady three-dimensional numerical simulations are conducted at various Reynolds numbers from laminar to critical regimes. The drag coefficients predicted from the present model agree very well with those from the present numerical simulation.


  • Turbulent boundary layer manipulation under a proportional-derivative closed-loop scheme

    by Z. X. Qiao, Z. Wu, Y. Zhou

    This work aims to experimentally investigate the manipulation of a turbulent boundary layer over a flat plate using a proportional-derivative (PD) controller. The control action is generated by an array of two flush-mounted piezo-ceramic actuators. Two different schemes are examined, i.e., feed-forward and feedback PD controls, with a view to suppressing the viscous-scaled near-wall cycle of high-speed events in the near-wall region and hence reducing skin friction drag. It has been found that the use of the feed-forward PD scheme may reduce the local maximum drag reduction by up to 33% at 14 wall units downstream of the actuator array, exceeding the open-loop control result (30%) as well as our previously reported combined feed-forward and feedback scheme (28%) [Z. X. Qiao, Y. Zhou, and Z. Wu, “Turbulent boundary layer under the control of different schemes,” Proc. R. Soc. A 473, 20170038 (2017)], and furthermore, this significantly cuts down the required input energy by 27%, compared to the open-loop control. On the other hand, the feedback PD scheme achieves the same control performance as the open-loop control, that is, producing a local maximum drag reduction of 30% without any saving in the input energy. The underlying control mechanism behind these control schemes is proposed based on the analyses of the hot-wire data measured with and without control.


  • Jet mixing optimization using machine learning control

    by Zhi Wu, Dewei Fan, Yu Zhou, Ruiying Li, Bernd R. Noack


  • A low-cost reusable micro-newton scale micro-thruster

    by Junhui Law, Winston Sun, Zhi Wu, Ho-Yin Chan, Yu Zhou, Wen Jung Li

    This paper presents a low-cost reusable Atomizing Liquid Micro-thruster (ALM) that can produce intermittent streams of water sprays based on thermal bubble-jet effect using batch-fabricated PCB micro-heaters. The propulsion performance of the micro-thruster has been studied in two methods: (a) by measuring its average weight loss during each spray and the velocity profile of the ejected droplets; (b) by measuring the deflection of micro-heater in pendulum motion. We have demonstrated that the current design of the micro-thruster has propulsive force in the micro-Newton range per thrust at atmospheric pressure.