Control of Hazardous Releases in Infrastructure Systems (NSF)

The objective of this project is to develop and apply a novel technology for the automatic detection and real-time mitigation of accidental or deliberate hazardous releases in infrastructure systems.  The automatic sensing and containment of a chemical release in these facilities are vital because of the catastrophic potential to human life.  The proposed technology can be used to protect a passenger terminal, transportation tunnel, aircraft fuselage, tall building or conduit carrying water to a municipality in a reliable, cost-effective and socially acceptable way. 

A prototype is being built at laboratory scale to demonstrate the detectability and controllability of hazardous releases in real time.  Microsensor arrays capable of detecting a broad menu of chemical agents will be installed at strategic locations in the prototype.  The sensors detect the instantaneous, spatially distributed concentration of the chemical agent and transmit the associated information to a predictive control model run by real-time hardware.  The model provides optimal operation scenarios for computer controlled bleed valves mounted on the channel walls and connected to a common manifold. Mitigation and final elimination of the chemical cloud is achieved by optimal blowing and suction of ambient fluid or injection of counteracting chemicals.

The predictive control model is based on the dynamics of ambient fluid flow and fate and transport of the hazardous release.  Gradient information is obtained by use of adjoint equations, so optimization of the control actions is achieved with the highest possible efficiency.  The control is optimized over a finite prediction horizon and instructions are sent to the valve manifold.  Next, the sensor arrays detect all changes effected by the control and report them to the control model, which advances the process over the next finite horizon.  Sensing, optimization and feedback are achieved in a time period shorter than the prediction horizon, so the process can proceed in real time. 

The sensors are tested under hostile environmental conditions, gas and liquid flow, different fluid velocities and chemical concentrations.  The optimal location, density and configuration of sensors and actuators are established for a variety of chemical agents.  Computational techniques are also enhanced to improve the speed of determining optimal strategies.  Finally, the proposed sensing and control model will be tested in two large-scale applications including the McNamara terminal of the Detroit Metropolitan airport and the Detroit-Windsor tunnel.  Hypothetical releases will be simulated based on the real geometry and ventilation system of these facilities in order to test the model's ability to work in real time.

The figure shows iso-surfaces of turbulent kinetic energy as a result of fluid suction from lateral ports in a conduit of rectangular cross section. Monitoring turbulent intensity is important because the control action may increase turbulent mixing and spread a hazardous chemical further.

 

 

Impact of Climate Change on Estuarine Morphology and Aquatic Life (GESI)

 

The goal of this research project is to develop the methodology for a quantitative assessment of the impact of climate changes on estuarine and wetland morphology.  By linking some key factors of river bed morphology to biota, a quantitative measure of the sustainability of aquatic life will also be developed.  

Long-term effects of land use practices lead to erosion and sedimentation processes that are responsible for many undesirable changes to aquatic life.  Unfortunately, the temporal and spatial scales of this problem are so disparate that a quantitative assessment of the impact of these practices has been previously impossible.  We propose a novel approach that can overcome these difficulties by nesting a high resolution estuarine morphology model in a large scale watershed model.  Both models are based on sound physical principles and are capable of capturing the true hydrodynamic behavior within their respective domains.  The two models are mathematically compatible and have been independently validated, so there is high confidence that their integrated results will provide a unique method for the investigation of long-term changes in aquatic environments.  The figure shows the collocation of the computational grids of the the hydrologic and hydrodynamic models.  The Voronoi cells of the watershed model are centered on the vertices of the Delaunay triangles of the hydrodynamic model providing a seamless flow of information between the two models.

 The integrated watershed-estuarine model is used to perform a formal sensitivity analysis of estuarine bed morphology to various hydrologic changes at the watershed level.  Variations in watershed headwater land use, as well as climate changes, directly affect the local runoff and erosion processes.  Over a long period of time these changes also affect the width of beaches, the depth and context of sediments, the cohesion of grain particles and the overall strength of the soil matrix.  These hydrodynamic and soil variables are be linked to an aquatic habitat model, resulting in a robust tool for assessing the impact of climatic change and land use practices on specific species.  By running simulation scenarios over long-term periods, we will be able to establish quantitative measures for creating sustainable aquatic systems.  The figure below show the evolution of a typical estuary bed due to sedimentation.  Over time, sediment deposition or erosion changes the vorticity patterns of the flow which in turn affects fish locomotion and plant stability.

 

 Source Identification by Adjoint Sensitivity Analysis (NSF)

 

This project is attempting to identify the location of a contaminant source and the time history of its intensity from measurements of concentration by field sensors located at some arbitrary locations.  Once a nonzero value of a chemical is detected by the sensors, each node in a discrete model of the environment is a potential source.  In addition, each discrete time instant has the potential to correspond to a different value of the intensity of every source. In multi-dimensional, time-dependent problems, a formidable number of parameters needs to be identified when attempting an inverse solution of contaminant fate and transport.  A direct solution would require numerous simulations, each corresponding to a perturbation of a single time instant and a single potential source in order to assess the sensitivity of the sensor measurement to the change of the intensity at each time and node of the model. 

In this project, we employ the adjoint sensitivity method for fast source identification. Adjoint "matter" sources are placed at the sensor location and the adjoint equations are solved numerically on the same computational grid with the direct problem. The Figure shows the regularized sensor error corresponding to the source inversion problem of an instantaneous source using two wall-mounted sensors. Flow is from left to right and the error results from guessing the location of the source after performing a single simulation. The maxima occur at the sensor locations while a well-defined minimum is visible at the location and time of the actual source. 

 The simulations involve low to medium resolution, which is adequate for the chemical release detection problem. The objective of this work is to capture time-averaged flow conditions over the time scales associated with sensor sampling and signal response.  As a result, turbulent flow scales that require higher resolution are not considered.  This has no consequences on the inversion model performance, and in addition, by reducing the computational time, makes possible the detection of chemical releases in real time.

 

 

 

 

 

 

 

 

 

 

 

 

Mitigation of Chemical Spills in Surface Waters (D'Appolonia S.P.A.)

We are developing an interactive tool for the mitigation of planned or accidental chemical spills in rivers and estuaries.  The source may be stationary or moving, known or unknown and the chemical may be inert or reacting, immiscible or completely mixed.

Chemical sensors are used to both detect and classify the spill.  The location of the sensors depends on local geometry and bathymetry.  The density of the sensor network is subject to optimal design considerations and is related on the need of redundancy and the model's resolution.  Mitigation of the spill is based on a system of pre-installed countermeasures, again on optimally design locations.  These measures or constructs depend on the type of the spill and may be either physical or chemical in nature.  For example, oil spills are contained by automatic deployment of floating dams.  Similarly, non buoyant spills may be eliminated by injection of counter acting chemicals. 

The model  is equipped with a graphical user interface that allows the user to execute alternative scenarios of mitigation in real time and to alter the basic parameters of the model in order to assess the uncertainty of the various measure that are examined.

 

 

Multi-Physics Model for Clutch Engagement (Ford Motor Co.)

We are developing a 3-dimensional, multi-physics model that can capture the dynamics of the multi-phase flow in an open clutch, squeeze flow during engagement, flow in the porous friction material, asperity contact, heat transfer and plate deformation following contact. A multiple disc system with realistic friction materials, waviness and grooves is considered, and the results are validated with full-scale experimental measurements. 


The sketch shows a typical layout of the system with the fluid contained between a stationary and a rotating disc. The small rectangular channels correspond to the grooves on the porous friction material, which is modeled separately.

We use the Volume-of-Fluid method to capture the multi-phase flow (oil and air) in the clearance, and an iterative algorithm that matches the externally applied force to the fluid pressure in the gap during engagement.  A moving reference frame is employed to simulate the rotating discs and several novel boundary conditions are introduced to account for the lack of accurate inflow/outflow boundary data. The figure shows the results of multi-phase flow computations in the vicinity of the grooves.

 

Grid Nesting of Hydrodynamic Models (CILER)

The hydrodynamics of the ocean encompasses processes whose spatial and temporal scales differ by several orders of magnitude. Local interaction of waves and currents, contaminant loading at river estuaries and non-point sources, fine sediments from bluff erosion and many similar phenomena affect the state of the entire system. However, it is often impossible to detect and monitor such events in the ocean because of the formidable computational effort required. It is possible to do so in specific nearshore regions, but this cannot be done without first determining the hydrodynamics of the whole ocean. The transfer of energy from large to small scales is the predominant factor driving the small scale effects, whose local features require that the nearshore processes be resolved as accurately as possible. Furthermore, these features often modify the large scale hydrodynamic effects, and, in the case of mass transport, almost always determine the characteristics of the source itself.

Despite advances in computer technology, nesting a high-resolution, limited-area nearshore model within a coarse-grid model for the ocean is the only viable approach to capturing the details of hydrodynamic phenomena that originate in the nearshore region, but may be of great importance to the coarsely resolved ocean scale. In a nested grid model, the nearshore component is hydrodynamically driven by the ocean model, so two-way passage of information is required between the two models, which must be enforced by appropriate boundary conditions. Furthermore, eddy energetics must converge at the nest boundaries to satisfy the basic conservation laws. Recent developments in grid-nesting indicate that the most difficult step of the process is the establishment of proper boundary conditions at the nest interface and the prevention of spurious-wave generation. The problem is further complicated in non-hydrostatic flows, where internal waves may create a pattern of flow that is not easily resolved by characteristic-type, non-reflecting boundary conditions.

This project tries to develop a nested scheme by developing a subspace projection method for time integration and non-reflective boundary conditions that allow information to travel freely through the computational boundaries of the nested grids. Although the proposed method is designed as a general grid-nesting technique, this project focuses exclusively on sub-domain communication issues, so nesting ratio issues are not discussed. This allows the presentation to shed light on the transfer of information between sub-domains without having to introduce any interpolation or other approximation errors associated with other nesting ratios.

 

 

Multi-Domain Computations (EPA)

When a numerical model is arbitrarily terminated by an open boundary, special boundary conditions need to be applied at that boundary to ensure that waves travel freely through this open boundary. Furthermore, when the open boundary represents an interface between two separate computational grids, waves on both sides must pass freely as well.

The technique proposed in this project assumes that the computation can be carried independently in two adjacent computational sub-domains. On common sub-domain boundaries, we allow the two computational grids to overlap, as is customarily done in nesting-grid methods for interpolation purposes. In addition, the receding boundary method assumes a derivative boundary condition for the pressure. Specifically, the derivative of the pressure is assumed to be zero.

This is of course a crucial assumption. The advantage is that it allows us to independently solve the pressure Poisson equation in each sub-domain. Hence, communications between sub-domains are decreased. The disadvantage is that on the sub-domain boundaries there is no incompressible correction for the velocities normal to the boundary. This in turn results in inaccurate corrections for the velocities, and furthermore, affects the corrections in all parts of each sub-domain, since the errors of the velocities on the boundaries are carried to the interior through advection and diffusion.

 

To minimize the errors caused by the derivative boundary condition for the pressure, the proposed receding boundary method makes the sub-domain boundaries in the overlapping regions non-stationary. The overlapping parts of the two grids are forced to move away from the boundary for a few time steps and then are reset to their original location. When the boundary is receding, the derivative boundary condition is applied at the last overlapping grid point for that time step. This prevents the repeated use of the assumed derivative boundary condition at the same location, and prevents the accumulation of errors on the sub-domain boundaries. In addition, the inward transport of errors is also decreased because the information outside the receding boundaries is discarded. However, if one allows the sub-domain boundaries to continuously recede, the sub-domain size will be eventually affected and force the overlapping regions of the sub-domains to vanish. To prevent this from happening, the sub-domain boundaries are reset to their original locations every few time steps. The missing information on the reset area is duplicated from the overlapping part of the other sub-domain.

 

 

Conjunctive Modeling of Surface and Groundwater Flow (Marmon Corp.)

A non-hydrostatic model for overland flow is developed for the purpose of providing the framework for predicting the fate and transport of chemicals from surface to groundwater and vice versa. The technique is based on the turbulent Navier-Stokes equations for the surface wave and Richards' equation for the movement of moisture in the underlying porous media.


It is widely accepted that a significant but undetermined portion of surface and sub-surface water contamination does not originate from any specific point sources. It is often suggested that chemicals that have been introduced in surface water or deposited on the soil's surface find their way to natural streams and aquifers.  Our model has
evolved from a hydrologic overland flow model by introducing a distributed source along the bed of the stream and allow contaminants to move by longitudinal advection and dispersion.
The hydrodynamic description has been usually based on kinematic wave assumptions and flow into the soil has been handled by empirical infiltration expressions. Although it has been long understood that turbulent diffusion alone is not responsible for the movement of contaminants, the dominant mechanism for the transport of contaminants by overland flow has not been rigorously established. Contaminants transported by sediments have been found to be a major contributor to non-point source pollution, but the hydrologic models are not capable of differentiating the conditions necessary for suspension and deposition of bed materials.

The basic hypothesis of this project lies in the vertical structure of both the overland and soil-water flow. Previous mathematical models have unveiled a mechanism for the uptake of contaminants by overland flow, which explains all available observations. It is believed that during surface irrigation a surge front develops which is characterized by a turbulent, rolling structure in the vertical
plane. Bed shear and suction reach very high values near the front, which becomes highly erosive. Soil particles are suspended and entrain the main body of overland flow. Desorption and dissolution of chemicals is then possible, followed by transport of contaminants both in solute and particulate form. Solute matter is therefore free to infiltrate into the soil where it re-adsorbs and perhaps reaches the ground water. It is clear that to model the aforementioned processes we need a two-dimensional model that can account for vertically varying velocity components and shear stresses, allows for the turbulent diffusion of solute matter, sediment suspension and deposition, and simultaneously couples the surface flow to the subsurface movement of moisture.


 

 

Multiphase Flow, Agitation and Mixing (Lafarge NA)

Several manufacturing processes depend on the efficient mixing of water, powder, foam and small amounts of certain additives.  Mixing implies that initially separated phases eventually become randomly distributed into each other.  This is typically achieved by agitation, i.e. induced motion of the materials in a specific way, usually in a circulatory pattern inside some sort of container.  

Large surface areas and reduction in thickness of material elements are very important for mixing, as the separate phases come in contact and allow molecular diffusion to occur.  Similarly, in liquid-gas mixing, shear stresses lead to a reduction in bubble size and promote mixing.  In an orbital mixer, the velocity gradient in the radial direction causes shearing of the fluid elements, so after each revolution they experience a reduction in thickness, thus achieving mixing in a compact container.

      Agitation is the cumulative effect of shear stresses over time. In calculating the shear forces and time a material spends under shear forces, we must take into account the significant variability of shear stresses across the entire volume of the device.  Therefore, to quantify the agitation associated with an entire device, we must compute the integral of the shear stresses.  Unlike the normal stress which identifies with a scalar variable in a static fluid, the shear stress is a tensor, or a two-dimensional array.  In three space dimensions, this tensor has nine components.  Furthermore, in turbulent flow the dominant part of the shear stress comes from the Reynolds stresses, which are generated by turbulent fluctuations of fluid velocity.  The model used in this work calculates the local values of all the components of both viscous and Reynolds stresses and integrates them over the entire volume of a mixing device.   We also compute the integral of the local enstrophy, which measures the fluid's ability to stretch and fold.  

 

 

Nonlinear Interactions of Stratified Flow in Lakes and Estuaries (EPA)

In practical applications of three-dimensional hydrodynamic models, it is often argued that high-order accuracy of computation and fine resolution of density fronts is not necessarily a point of primary concern because a plethora of other factors dominate the performance of the model.  This is often true and especially in cases where there is great uncertainty associated with the input data, it is hard to argue the value of a 5% improvement in the location or sharpness of a salinity or thermal front at the mouth of an estuary is worthy of serious consideration.

On the other hand, there exist practical applications where such an improvement in accuracy can result in dramatic changes in a model's predictive ability of certain phenomena.  In particular, a variety of hydrodynamic phenomena depend entirely on exceeding certain thresholds that lead to solution bifurcations and often a chaotic behavior, depending on the branch that is followed.  For example, it makes a great difference if a salinity front is arrested before or after a sewage outfall in an estuary or if excess numerical dissipation induces cohesive sediment flocculation where it is really absent. 

 The figure shows the simulation of a thermal current from Lake Michigan entering Green Bay.  Our study shows a high sensitivity of the thermal front on the model's ability to resolve discontinuities.  We investigate the impact of high-resolution schemes in a practical application involving a complex environment.  The model has the ability to employ several numerical formulations and we plan to show that in certain cases what appears to be a marginal improvement in accuracy can lead to major differences in the predicted flow conditions.

 

 

 

 

 

Active Mitigation of Flood Waves (NSF)

We develop a mathematical technique for active flood hazard mitigation. The procedure is based on the hypothesis that passive measures are not available or are inadequate, and a hazardous flood wave is already propagating in the river system. For example, a levee protecting a municipality has already been breached or its failure is imminent and real-time responses are needed to mitigate the impact of the flooding. The proposed method is also based on the assumption that there exist one or more alternative locations for emergency flow diversion, which can diminish the damages from the collapse of the levee, and even prevent its failure altogether.

The proposed method will scan the entire river basin in the vicinity of the endangered levee, identify the optimum location for a flow diversion, and provide the optimum time schedule for intervention. The decision will be based on information derived from a two-dimensional, river-floodplain simulation model with a dynamic lateral flow withdrawal component. The optimum location and timing of this flow diversion will be determined in real time by mapping the problem on the adjoint space, which determines the associated sensitivities and optimization gradients with just two runs of the model.

The actual flood mitigation is accomplished by an optimum spatial and temporal combination of on-demand storage relief and the focusing of depression waves on the endangered levee location. The first reduces the severity of the approaching flood wave by providing off-line storage on adjacent flood plains. The second directs a family of depression waves towards the endangered levee, thus negating directly the impact of the flood wave. The optimum combination of these measures in real time is achieved by use of the adjoint equation method for shallow-water flow.