Alfredo Soldati

River flood forecasting with a neural network model

A neural network model was developed to analyze and forecast the behavior of the river Tagliamento, in Italy, during heavy rain periods. The model makes use of distributed rainfall information coming from several rain gauges in the mountain district and predicts the water level of the river at the section closing the mountain district. The water level at the closing section in the hours preceding the event was used to characterize the behavior of the river system subject to the rainfall perturbation. Model predictions are very accurate (i.e., mean square error is less than 4%) when the model is used with a 1-hour time horizon. Increasing the time horizon, thus making the model suitable for flood forecasting, decreases the accuracy of the model. A limiting time horizon is found corresponding to the minimum time lag between the water level at the closing section and the rainfall, which is characteristic of each flooding event and depends on the rainfall and on the state of saturation of the basin. Performance of the model remains satisfactory up to 5 hours. A model of this type using just rainfall and water level information does not appear to be capable of predicting beyond this time limit.

Mechanisms for particle transfer and segregation in a turbulent boundary layer

Particle transfer in the wall region of turbulent boundary layers is dominated by the coherent structures which control the turbulence regeneration cycle. Coherent structures bring particles toward and away from the wall and favour particle segregation in the viscous region, giving rise to non-uniform particle distribution proles which peak close to the wall. The object of this work is to understand the reasons for higher particle concentration in the wall region by examining turbulent transfer of heavy particles to and away from the wall in connection with the coherent structures of the boundary layer. We will examine the behaviour of a dilute dispersion of heavy particles { flyashes in air { in a vertical channel flow, using pseudo-spectral direct numerical simulation to calculate the turbulent flow eld at a shear Reynolds number Re = 150, and Lagrangian tracking to describe the dynamics of particles. Drag force, gravity and Saman lift are used in the equation of motion for the particles, which are assumed to have no influence on the flow eld. Particle interaction with the wall is fully elastic. As reported in several previous investigations, we found that particles are transferred by sweeps { Q2 type events { in the wall region, where they preferentially accumulate in the low-speed streak environments, whereas ejections { Q4 type events { transfer particles from the wall region to the outer flow. We quantify the eciency of the instantaneous realizations of the Reynolds stresses events in transferring different size particles to the wall and away from the wall, respectively. Our ndings conrm that sweeps and ejections are ecient transfer mechanisms for particles. In particular, we nd that only those sweep and ejection events with substantial spatial coherence are eective in transferring particles. However, the eciency of the transfer mechanisms is conditioned by the presence of particles to be transferred. In the case of ejections, particles are more rarely available since, when in the viscous wall layer, they are concentrated under the low-speed streaks. Even though the low-speed streaks are ejection-like environments, particles remain trapped for a long time. This phenomenon, which causes accumulation of particles in the near-wall region, can be interpreted in terms of overall fluxes toward and away from the wall by the theory of turbophoresis. This theory, proposed initially by Caporaloni et al. (1975) and re-examined later by Reeks (1983), can help to explain the existence of net particle fluxes toward the wall as a manifestation of the skewness in the velocity distribution of the particles (Reeks 1983). To understand the local and instantaneous mechanisms which give rise to the phenomenon of turbophoresis, we focus on the near-wall region of the turbulent boundary layer. We examine the role of the rear-end of a quasistreamwise vortex very near to the wall in preventing particles in the proximity of the wall from being re-entrained by the pumping action of the large, farther from the wall, forward-end of a following quasi-streamwise vortex. We examine several mechanisms

Artificial neural network approach to flood forecasting in the River Arno

Abstract The basin of the River Arno is a flood-prone area where flooding events have caused damage valued at more than 100 billion euro in the last 40 years. At present, the occurrence of an event similar to the 1966 flood of Firenze (Florence) would result in damage costing over 15.5 billion euro. Therefore, the use of flood forecasting and early warning systems is mandatory to reduce the economic losses and the risk for people. In this work, a flood forecasting model is presented that exploits the real-time information available for the basin (rainfall data, hydrometric data and information on dam operation) to predict the water-level evolution. The model is based on artificial neural networks, which were successfully used in previous works to predict floods in an unregulated basin and to predict water-level evolution in the Arno basin under low flow conditions. Accurate predictions are obtained using a two-year data set and a special treatment of input data; which allows a balance to be found between the spatial and temporal resolution of rainfall information and the model complexity. The prediction of water-level evolution remains accurate within a forecast time ahead of 6 h, which is the minimum time lag for the river to respond to dam releases under saturated conditions of the basin. The predicted flow rate percentage error ranges from 7 to 15% from the 1-h ahead to 6-h ahead predictions, and the accuracy of prediction increases for each time ahead of prediction, as the flow rate increases, suggesting that the model is particularly suited for flood forecasting purposes.

Turbulence modification by large-scale organized electrohydrodynamic flows

The interactions of flows generated by ionic discharges with wall turbulence are not only of interest for turbulence control, but also for devices of industrial importance, such as wire-plate electrostatic precipitators (ESPs). Under conditions of uniform discharge, in wire-plate ESPs, arrays of regular, spanwise vortices are found in the absence of a through-flow. These arise from ionic discharges from the spanwise wires placed between the grounded plates on each side. The interactions of such electrohydrodynamic (EHD) flows with a turbulent through-flow are still poorly understood. Direct numerical simulation (DNS) is an attractive method for investigating such problems since the details of the interactions can be unraveled, and the results are directly applicable to industrial-scale systems because their Reynolds numbers are typically quite low. In this study, pseudospectral channel flow simulations were performed with the electrohydrodynamic effects being modeled by a spatially varying body-force term in the equations of fluid motion. The interactions between EHD flows and wall structures were elucidated by examining the instantaneous structure of the flow field. Results indicate that the mean flow, the EHD flows, and the turbulence field undergo significant modifications caused by mutual interaction. First, it is found that EHD flows reduce drag, allowing larger flow rates for a given pressure drop. Second, the EHD flows themselves appear weakened by the presence of the through-flow, particularly in the central region of the channel. The EHD flows affect the turbulence field by both increasing dissipation and turbulence production, the overall turbulence level being determined by the balance between the increased dissipation and production. Even though high EHD flow intensities may increase streamwise and wall-normal turbulence intensities, the Reynolds stress is reduced, consistent with the observed reduction in drag. From a mechanistic viewpoint, there are indications that EHD flows of the type investigated here reduce drag by decreasing the relative importance of the positive Reynolds stress contributions, i.e., second (ejections) and fourth (sweeps) quadrant events, compared to the negative Reynolds stress contributions, i.e., first and third quadrant events.

Direct numerical simulation of particle wall transfer and deposition in upward turbulent pipe flow

Abstract Transfer and deposition of inertial particles or droplets in turbulent pipe flow are crucial processes in a number of industrial and environmental applications. In this work, we use direct numerical simulation (DNS) and Lagrangian tracking to study turbulent transfer and deposition of inertial particles in vertical upward circular pipe flow. Our objects are: (i) to quantify turbulent transfer of heavy particles to the wall and away from the wall; (ii) to examine the connection between particle transfer mechanisms and turbulence structure in the boundary layer. We use a finite difference DNS to compute the three-dimensional time dependent turbulent flow field (Reτ=337) and Lagrangian tracking of a dilute dispersion of heavy particles––flyashes in air––to simulate the dynamics of particles. Drag, lift and gravity are used in the equation of motion for the particles, which are assumed to have no influence on the flow field. Particle interaction with the wall is fully elastic. Results on preferential distribution of particles in the boundary layer, particle fluxes to and off the wall and particle deposition mechanisms are shown. Our findings confirm: (i) the specific tendency of particles to segregate in the near-wall region; (ii) the crucial role of the instantaneous realizations of the Reynolds stresses in determining particle fluxes toward and away from the wall; (iii) the relative importance of free-flight and diffusion deposition mechanisms.

Orientation, distribution, and deposition of elongated, inertial fibers in turbulent channel flow

In this paper, the dispersion of rigid, highly elongated fibers in a turbulent channel flow is investigated. Fibers are treated as prolate ellipsoidal particles which move according to their inertia and to hydrodynamic drag and rotate according to hydrodynamic torques. The orientational behavior of fibers is examined together with their preferential distribution, near-wall accumulation, and wall deposition: all these phenomena are interpreted in connection with turbulence dynamics near the wall. In this work a wide range of fiber classes, characterized by different elongation (quantified by the fiber aspect ratio, λ) and different inertia (quantified by a suitably defined fiber response time, τp) is considered. A parametric study in the (λ,τp)-space confirms that, in the vicinity of the wall, fibers tend to align with the mean streamwise flow direction. However, this aligned configuration is unstable, particularly for higher inertia of the fiber, and can be maintained for rather short times before fibers ...

Mechanisms of particle deposition in a fully developed turbulent open channel flow

Particle dispersion and deposition in the region near the wall of a turbulent open channel is studied using direct numerical simulation of the flow, combined with Lagrangian particle tracking under conditions of one-way coupling. Particles with response times of 5 and 15, normalized using the wall friction velocity and the fluid kinematic viscosity, are considered. The simulations were performed until the particle phase reached a statistically stationary state before calculating relevant statistics. For both response times, particles are seen to accumulate strongly very close to the wall in the form of streamwise oriented streaks. Deposited particles were divided into two distinct populations; those with large wall-normal deposition velocities and small near-wall residence times referred to as the free-flight population, and particles depositing with negligible wall-normal velocities and large near-wall residence times (more than 1000 wall time units), referred to as the diffusional deposition population....

FORECASTING RIVER FLOW RATE DURING LOW-FLOW PERIODS USING NEURAL NETWORKS

The pollution in the river Arno downstream of the city of Florence is a severe environmental problem during low-flow periods when the river flow rate is insufficient to support the natural waste assimilation mechanisms which include degradation, transport, and mixing. Forecasting the river flow rate during these low-flow periods is crucial for water quality management. In this paper a neural network model is presented for forecasting river flow for up to 6 days. The model uses basin-averaged rainfall measurements, water level, and hydropower production data. It is necessary to use hydropower production data since during low-flow periods the water discharged into the river from reservoirs can be a major fraction of total flow rate. Model predictions were found to be accurate with root-mean-square error on the predicted river flow rate less then 8% over the entire time horizon of prediction. This model will be useful for managing the water quality in the river when employed with river quality models.

ON THE EFFECTS OF ELECTROHYDRODYNAMIC FLOWS AND TURBULENCE ON AEROSOL TRANSPORT AND COLLECTION IN WIRE-PLATE ELECTROSTATIC PRECIPITATORS

Abstract Predicting transport of aerosols or particles in wire-plate electrostatic precipitators is complicated by the influence of EHD flows and turbulent flow field. In this work we use the direct numerical simulation by Soldati and Banerjee (1998) to analyze the effects of the EHD flows and of turbulence on particle transport and collection efficiency. Particles of different size and charge were tracked in two different flow fields corresponding to different potential applied to the wires of the precipitator. Results were compared against simulations in which the electrostatic field acted only on particles, − i.e. no EHD flows. It is apparent that EHD flows are large advective motions with scale of the wire-to-wall distance and have a strong effect on the local behavior of particles, sweeping them into different regions of the channel. However, it was found that the overall collection efficiency of the precipitator is not significantly affected by the presence of EHD flows. Even in the vicinity of the wall, EHD flows appear to have negligible influence on particle deposition.

Some issues concerning large-eddy simulation of inertial particle dispersion in turbulent bounded flows

The problem of accurate Eulerian-Lagrangian modeling of inertial particle dispersion in large-eddy simulation (LES) of turbulent wall-bounded flows is addressed. We run direct numerical simulation (DNS) of turbulent channel flow at shear Reynolds number Re-tau=150 and corresponding a priori and a posteriori LES on two coarser grids. For each flow field, we tracked swarms of particles with different inertia to examine the behavior of particle statistics, specifically focusing on particle preferential segregation and accumulation at the wall. Our object is to discuss the necessity of a closure model for the particle equations when using LES and we verify if the influence of the subgrid turbulence filtered by LES is an important effect on particle motion according to particle size. The results show that well-resolved LES gives particle velocity statistics in satisfactory agreement with DNS. However, independent of the grid, quantitatively inaccurate predictions are obtained for local particle preferential segregation, particularly in the near-wall region. Inaccuracies are observed for the entire range of particle size considered in this study, even when the particle response time is much larger than the flow time scales not resolved in LES. The satisfactory behavior of LES in reproducing particle velocity statistics is thus counterbalanced by the inaccurate representation of local segregation phenomena, indicating that closure models supplying the particle motion equation with an adequate rendering of the flow field might be needed. Finally, we remark that recovering the level of fluid and particle velocity fluctuations in the particle equations does not ensure a quantitative replica of the subgrid turbulence effects, thus implying that accurate subgrid closure models for particles may require information also proportional to the higher-order moments of the velocity fluctuations. (c) 2008 American Institute of Physics.