1. Dispersion of microbes in water distribution systems Chris Choi Department of Agricultural and Biosystems Engineering The University of Arizona

2. Michigan State University Drexel University The University of Michigan Carnegie Mellon University The University of Arizona The University of California, Berkeley Northern Arizona University Main Research Focus of Choi’s Group: Water Distribution Systems CAMRA – National Homeland Security Center

3. UA’s Research Responsibilities Exposure, Detection, Fate and Transport of Agents UA’s Research Responsibilities Exposure, Detection, Fate and Transport of Agents - The goal is to improve our ability to quantify exposure to biological agents of concern (Category A and B agents) in drinking water systems and indoor air environments.

4. EPANET-based Simulation -HD Model - WQ Model Experimental Validation using Water-Distribution Networks at the Water Village ANN-based Prediction Models RISK ASSESSMENT Indicator Microorganisms (Gerba et al.) Proposed Research Plan

5. Collaboration Plan Water Distribution Laboratory Microbiology (Dr. Gerba’s Laboratory) Quantitative Microbial Risk Assessment Utilities (Tucson Water) National Laboratories (Sandia National Laboratories) EPA Bio-sensor Researchers Private Companies (Hach Event Monitor and Bio-Sentry) Accurate data sets are essential!

6. What is EPANET? EPANET models the hydraulic and water quality behavior of water distribution piping systems. EPANET is a ‘free & open source’ Windows program written in C & Delphi programming languages that performs extended period simulation of hydraulic and water-quality behavior within pressurized pipe networks. A network can consist of pipes, nodes (pipe junctions), pumps, valves and storage tanks or reservoirs. A “Node-to-Node” macroscopic Approach – Remember the cube Prof. Hass Introduced. 3D Control Volume 1D Control Volume Node i Node j 2D Control Volume

7. EPANET is one of many WDS tools Ref @ Angel Website:

8. Vulnerability of Water Distribution Systems What if… $ 100 pump contaminants Fire Hydrant without backflow prevention devices

9. Serious Engineering and Sensor Research Efforts by Various Organizations: Example at an EPA Lab in Cincinnati Lab visits

10. Topics at 2006 WDSA Engineering Conference

12. Field Trips

14. City of Tucson Water Distribution Network UA & Downtown Area

15. Water Distribution Network near the University of Arizona Campus Univ. of Arizona Exemplary Case Water In Intrusion

16. Sample Water Distribution Network Water In Biological Agent Intrusion Point Perfect Mixing Assumed at the Cross Joint for Modeling Tools to Subdivision A to Subdivision B to Subdivision C 50 %

17. Real World Systems Models

18. Contaminated Water (C = 1) Un-contaminated Water (C = 0) C = 0.5? Perfect Mixing Assumption

19. Corresponding Risk Microbial Risk Assessment & Consequences

20. Contaminated Water (C = 1) Un-contaminated Water (C = 0) C = 0.85 C = 0.15 Perfect Mixing Assumption Courtesy of Sandia National Laboratories

21. Corresponding Risk Microbial Risk Assessment & Consequences

22. How to correct this problem? Computational Approach – CFD Experimental Approach Understanding of Fluid Mechanics (turbulent flow, in particular) and Transport Phenomena

23. Experiment conducted by Osborne Reynolds (1842 - 1912) Laminar & Turbulent Flow in Pipes Re<2100, laminar flow Re>2100, turbulent flow

24. Governing Equations for Laminar Flow

25. Key Parameters

26. 2D Control Volume Each box a infinitesimal control volume Discretized Governing Equation Discretization CFD Procedure

27. Turbulent Flow Navier-Stokes equations should apply, but this is not usually solvable for random and inherently unsteady (in a small scale) turbulent flows. Suggested approach: to time-average the N-S equations and look at the effect of the unsteady turbulent motions: Reynolds’ Equation Introduce & Ref @ Angel Website:

28. Basic Equations for Turbulent Flow

29. Governing Equations for Turbulent Flow (based on  model) where

30. Boundary Conditions

31. Key Parameters Re = f(Flow Speed) 0 < Re < 60,000+

32. Turbulent Schmidt Number (Sc t ) Key Parameters In a k-ε turbulent model, total diffusivity is composed of the molecular and eddy diffusivities. The eddy diffusivity is calculated through the turbulent Schmidt number. Therefore, eddy diffusivity is directly proportional to the eddy viscosity computed at each node and inversely proportional to Sc t. (Laminar Flow)

33. Preparation of Sensors, Pumps, and Dataloggers for Water Distribution Network Laboratory

34. Role of Experimental Data

35. Detailed Computational and Experimental Results

36. Complex Oscillatory Flow Patterns Depending Upon Flow Speed Concentration Profiles Velocity Vectors

37. Detailed Computational and Experimental Results

38. Mixing patterns along the interface Lagrangian Particle Trajectories

39. Mixing patterns along the interface

40. Water Village

41. The Water Village Oasis Garden Data & Education Building Water Quality Laboratory Main Office Water Quality Center Environmental Research Laboratory Rainwater Catchment Xeriscape Gardens Project Office Edible Garden Visitor Parking Mesquite Bosque Tucson International International Airport Airport Unit A: Microbiology Lab Unit B: Water Distribution Network Lab

42. Unit A: Microbiology & POU Lab

43. Unit B: Water Distribution Network Laboratory

45. A schematic of the experimental setup (A)Saltwater Tank (B)Pump (C)Gate Valve (D)Electrical Conductivity Sensor (E)Flow Sensor (F)Freshwater Tank.

46. Scenario 1: Equal inflows and outflows (Re S = Re W = Re N = Re E ) Scenario 2: Equal outflows, varying inflows (Re S ≠ Re W, Re N = Re E ) Scenario 3: Equal inflows, varying outflows (Re S = Re W, Re N ≠ Re E ) Complexities & Three Scenarios

47. Comparison – An Example NaCl mass rate splits from the experimental, numerical and water quality model outcomes at different Re E/N (East Outlet), when Re S = Re W and Re E ≠ Re N.

48. Numerical Results based on Revised Sc t Dimensionless concentration of the experimental and numerical results with corrected Turbulent Schmidt Number (Sc t ) for the East Outlet when Re S ≠ Re W and Re E = Re N. Correction & Generalization

49. Revision of Water Distribution Model CFD simulations based on four Reynolds numbers at each node with Sc t = 0.135 Revise EPANET using C programming language based on CFD results

50. Comparison Current WDS Model Improved WDS Model

52. Additional Reading Material Ref @ Angel Website:

53. Artificial Neural Network 5x5 Sensor location Region 1 2 3 4 5 A 5 x 5 network simulation data using EPANET 24 nodes with water demand 100 ft between pipes Two pumps, 1-point curve, 100 GPM – 40 ft 5 sensor locations 4 potential release REGIONS

54. Artificial Neural Network – 10x10 IP-1 S-1 S-2 S-3 S-4 IP-2 IP-3 IP-4 Implemented with EPANET, 10 x 10 nodes, 4 likely injection points ( ) 4 sensors ( ) All points have water demand (except from injection points and sensors) Three pumps supply water demands (1-point curve, 80 GPM @ 50 ft) Pipe diameter = 2 in, 100 ft between nodes, H-W Coef = 100 No elevation change,

55. Injection of Surrogates into the Network

56. EPANET-based Simulation - HD Model - ImprovedWQ Model, if possible - Improved WQ Model, if possible Simulation Data Validation using Water-Distribution Network at the Water Village ANN-based Prediction Models CFD-based EPANET WQ Re-evaluation Validation of Improved EPANET WQ Model GROUP 3: RISK ASSESSMENT Summary

57. Research Sponsors and Major Collaborators