CSOnet A Metropolitan Scale Wireless Sensor-Actuator Network M

CSOnet: A Metropolitan Scale Wireless Sensor-Actuator Network M. D. Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC WIDE kickoff - Siena Italy - Sept 26, 2008. September 26, 2008 Michael Lemmon University of Notre Dame Luis A. Montestruque Em. Net, LLC

Outline n Combined Sewer Overflow Problem q n n In-line Storage using CSOnet System Hardware CSOnet Middleware Real-time Control Strategy Future Directions September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

CSOnet Project Background n n n CSOnet concept: use of wireless sensor-actuator networks for distributed monitoring and control of combined sewer overflow (CSO) events. Concept originally created at Notre Dame (Lemmon/Talley) q Funded through the State of Indiana’s 21 st Century Technology Fund 1 million USD (2004 -2006) - 1 million USD (2007 -2009) Academic, private sector, and public sector partners Purdue (Bagchi/Chappell) City of South Bend Em. Net LLC (L. Montestruque) Greeley & Hansen Notre Dame (Lemmon/Talley) n n 150+ Sensor network monitoring 13, 000 acres (summer 2008) Actuation component scheduled for completion in summer 2009 September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 3

Combined Sewer Overflow Events Dry Weather n Wet Weather n Overflow Combined Sewer Overflow During Wet Weather Interceptor Sewer Combined sewer overflow (CSO) events occur when a municipality dumps untreated water from combined storm and sanitary sewer flows into a river/stream. “Such ‘exceedances’ can pose risk to human health, threaten aquatic life and its habitat, and impair the use and enjoyment of the Nation’s waterways. ” EPA, “Combined Sewer Overflow Control Policy, ” April 19, 1994. (www. epa. gov) n St. Joseph River September 26, 2008 EPA fines for CSO events - 1994 CSO Control Act - Fines are Significant n Problem is Large-scale Over 772 cities nationwide South Bend Wastewater Treatment Plant Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Combined Sewer Overflow Events September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Combined Sewer Overflow Events September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Solution Strategies Sewer Separation Off-line Storage Tunnels Chicago’s TARP project Expansion of WWTP These strategies all require significant investment in new infrastructure September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

In-line Storage n n n Store excess water in unused parts of the system Reduces need for infrastructure enhancement Requires real-time monitoring and control of flows Reduced inflow to Manhole B allows us to increase inflow through Manhole A Inflow of Storm Water conduit Manhole A September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame conduit Manhole B Luis A. Montestruque Em. Net, LLC

Centralized Model Predictive Control n n n Pre-1974 combined sewer trunklines Interceptor Sewer to Wastewater Treatment Plant Monitoring and Control over SCADA network Retention Basin Combined Sewer Trunk Lines Retention Basin City Engineer Retention Basin CSO Diversion Structure Interceptor Sewer Line Outfall to River September 26, 2008 Outfall to River Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Outfall to River Luis A. Montestruque Em. Net, LLC Outfall to River

SCADA Systems n n September 26, 2008 Expensive Limited implementation Limited feedback data High reliance on hydraulic models Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Distributed In-line Storage n n Distributed Feedback Control of In-line Storage sets up interacting “control zones” Gateways connect to Internet to provide global monitoring of system INTERNET City Engineer September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Distributed In-line Storage September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Ireland/Miami Network (Summer 2005) I CSO 22 Diversion Point mb Tru ined nk Se Lin wer e R R R Co I GR VI R R R Prototype system controls retention basin based on flow measurements at CSO 22 diversion structure R n I n n G n V n 7 Relay Rnodes (radios) 3 Instrumentation Inodes (sensors) 1 Gateway Gnode connect to the internet Automated valve Network is fully deployed and operational Retention Basin First month of service the system prevented 2 million gallon CSO discharge CSO 22 Area September 26, 2008 Continuous operation since summer 2005. Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

South Bend CSOnet PHASE 1: (summer 2008) n 800 km of sewers n 50 km 2 of CSO area n 110 sensors n Monitor 36 outfalls, 27 interceptor locations, 42 trunkline locations, and 5 basins PHASE 2: (summer 2009) n Control at least 18 outfalls n Control storage in retention basins September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 14

CSOnet Website Overflow Event Sensor Location September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Outline n Combined Sewer Overflow Problem q n n In-line Storage using CSOnet System Hardware CSOnet Middleware Real-time Control Strategy Future Directions September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

CSOnet Architecture n Hierarchical network q q Unicasts in low level subnets, Multicasts over higher level 3 types of Nodes - Gnode, Rnode, Inode September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

CSOnet Inode Emplacement Rnode To next Rnode Stoplight Pole Manhole Cover Antenna Inode microprocessor Stilling Well Sensor September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Slide provided by courtesy of Em. Net LLC

Actuators Actuated Valve GNode Cabinet or Traffic Signal Manhole Cover Antenna INode Stilling Well Conduit Actuated Valve Larger Parallel Throttle Line Stilling Well Weir Pneumatic Bladder Overflow Line Combined Sewer Trunkline Sensor September 26, 2008 Interceptor Line Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Slide provided by courtesy of Em. Net LLC

Composite Manhole Cover n n n Integration of processor, radio transceiver, and antenna into manhole cover q William Chappell - Purdue September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC Manhole is extremely corrosive environment Initial prototypes “rusted” away with a few months

Intelligent Radio Antenna n n Municipal deployments must cope with fading effects due to multipath interference. Intelligent switching between multiple antennae single directional antenna Intelligent switching antenna Ideal Antenna with angular diversity September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC Slide provided by courtesy of Bill Chappell (Purdue)

Chasqui Module n Inode and Rnode are based on the Chasqui Module Chasqui 2. 0 with antenna Chasqui 1. 0 Chasqui 1. 1 Chasqui 2. 0 n n n Based on UCB Mica 2 Module Max. Stream Radio (115 kbps/900 MHz) Rugged Sensor-Actuator I/F Precision Real-time Clock (2 ppm drift) Tiny. OS Compatible September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

GNode n Gateway Node consists of q q q Single Board Computer (SBC) xx 86 - compact Linux Chasqui Node (radio - actuator interface) Cellular connectivity to Internet. Single Board Computer September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Outline n Combined Sewer Overflow Problem q n n In-line Storage using CSOnet System Hardware CSOnet Middleware Real-time Control Strategy Future Directions September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Middleware for Mesh Radio Networks n Middleware maintains an network abstraction that can be easily used by application software q n Middleware services q q q n Time-slotted publish-subscribe network abstraction Clock Synchronization Networking Service Routing Service Power Management Service Reprogramming Service Data Sources Reply to Request Data Sink Network Programmed using Tiny. OS Request for Data Type September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Networking Service n Network service uses directed flooding to create a gradient table 22 22 3 3 22 11 1 3 11 1 22 3 SINK 22 11 1 22 September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame 11 1 Luis A. Montestruque Em. Net, LLC 26

Stateless Gradient-based Routing n Broadcast to all upgradient nodes 2 2 3 3 2 2 2 2 1 1 1 2 3 3 1 3 3 SINK 2 2 1 1 1 2 September 26, 2008 3 3 1 1 2 2 3 1 1 1 2 3 1 1 2 2 3 3 2 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 27

Congestion Issues Subnet 23 Throughput Results n CSOnet has two types of subnets - large diameter with few sensors - small diameter with many sensors Small diameter networks can have congestion problems unless the data received at the gateway is buffered September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Throughput no buffering Throughput with buffering G. 23. 1 100% I. 23. 1 46% 89% I. 23. 2 56% 91% I. 23. 3 77% 93% I. 23. 4 91% 100% I. 23. 5 n Node 75% 85% Luis A. Montestruque Em. Net, LLC

Power Management n Management of system duty cycle q q n ON ON ASLEEP Requires tight “clock” synchronization q q n 2 percent duty cycle - 5 minute period During sleep cycle, microprocessor put into deep sleep mode. External timer is used to wake the system back up Chasqui uses Dallas DS 3231 RTC with 2 ppm drift. Resync network clocks every six hours Chasqui service lifetime q q q 2 years between service 3. 6 volt - 19 amp-hour lithium source Currently more cost effective to replace batteries than to use renewable power systems such as solar. September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 29

Wireless Reprogramming n n “Stream” Reprogramming protocol developed by Dr Saurabh Bagchi (Purdue) Less overhead than Deluge Stream segments the program image into Stream-RS (Stream Reprogramming Support) and Stream-AS (Stream Application Support) Stream-RS q q n Program memory Core reprogramming component Preinstalled, before deployment, in all nodes Stream-AS q q Currently executing program A small subset of reprogramming component that is attached to the user application Instead of wirelessly transferring through the network user application plus the entire reprogramming component, Stream transfers Stream-AS plus the user application September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC Flash Stream-RS (Image-0) Stream-AS + User application (Image-1) Unused portion Slide provided by ourtesy of Saurabh Bagchi (Purdue)

Outline n Combined Sewer Overflow Problem q n n In-line Storage using CSOnet System Hardware CSOnet Middleware Real-time Control Strategy Future Directions September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Complete Dynamic Wave Model n Momentum Equation n Continuity Equation n Q = flow rate (m 3/s) A = cross section area of flow (m 2) Manning’s equation H = head level (m) ground level H=0 September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Simplified Wave Model n n Flow Resistance Equation (momentum equation) Simplified Continuity Equation Qu Hu Where ai = water surface area manhole wu Hu Q Hd pipe L Ground H=0 September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC wd Flow, Q Hd

Distributed Real-time Control n Model Variables q q q wi = storm inflow Oi = overflow ui = diverted flow Hi = water height (head) Q I = pipe flow rate Optimal Control Problem n n Maximize “diverted flow” Subject to: q q q September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Conservation of Mass Conservation of Momentum Admissible control No flooding WWTP capacity limit Luis A. Montestruque Em. Net, LLC

Supervisory Control Strategy Control Selection Algorithm n Assume costs are ordered as n If node ij is flooded n If node ij is not flooded then : active flood constraint September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Supervisory Control Results n“Optimal” q q Supervisory Control Strategy Open valves until “flooding” constraint is active Then reduce diverted inflow to prevent violation of flooding constraint Storm Existing System Overflow (ft 3 x 106) Controlled System Overflow (ft 3 x 106) Overflow Volume Decrease (ft 3 x 106) Overflow Decrease (%) Storm 1 1. 50 1. 10 0. 40 27% Storm 2 3. 46 2. 68 0. 78 23% Storm 3 13. 6 9. 45 4. 15 31% Table A. Scenario A—Moving Uniform Rainfall September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Pressure-based Feedback n n Actual system only has “pressure” (head) measurements Decentralized Pressure-based Controller used to enforce flooding constraint. Model of Head Level Dynamics Limitation: the diverted flow must be positive STORM FLOW VALVE FLOODING CONSTRAINT ACTIVE HEIGHT OVERFLOW INODE Pressure Sensor (water height) TIME September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Input-Output Behavior of Node n Testbed Experiments showed that “head” level dynamics had at least 3 state variables q Head level, downstream flow rate, water stored in upstream link Ellipsoidal shape of response implies additional energy storage September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Head Model Identification n n Drive interceptor line with a “persistently exciting” input signal Gather input/output data for a “design” set and a “test” set. Use “design” data set and Matlab’s Sys. ID toolbox to identify a statebased model Test that model against the “test” data set. SB Interceptor Line Node 8 Model September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Pressure-based Controller Design n n Disturbance rejection problem Loopshaping Design q n Controller Often yields PID-type control State-based controller PLANT Reference pressure Loopshaping Design Plot September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame NODE 8”s RESPONSE Luis A. Montestruque Em. Net, LLC

Flooding under Supervisory Control n n September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC Supervisory strategy is only “necessary” for optimality This strategy can lead to localized flooding in a flooded node loses “control authority”

Flood Prevention n n Flooding may occur if node loses “control authority” Flood Prevention Protocol q If node i is about to flood and has no remaining control authority THEN request upstream node to “hold” at its current head level. Hu pipe Hd September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Simulation Results - Head Levels September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Simulation Results - Total Overflows (ft 3) Passive Threshold Supervisory Control Decentralized Control Percent Change STORM C 405980 123750 152490 60% STORM D 1206900 770430 883560 26% STORM E 2682800 2050200 2141200 20% STORM G 9280600 8068800 8413400 9% • 10 -60 percent reduction in total overflow September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Outline n Combined Sewer Overflow Problem q n n In-line Storage using CSOnet System Hardware CSOnet Middleware Real-time Control Strategy Future Directions September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Moving toward Distributed Local Control n CSOnet’s controller q q q High-level supervisor to enforce optimality Low-level decentralized controllers to enforce safety (no flooding) We could do better with “distributed” local controls STORM FLOW OVERFLOW VALVE INODE Head level information from upstream and downstream nodes Pressure Sensor (water height) September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Need for Real-time Middleware n n n Distributed control requires hard/firm real-time message delivery It may be possible to develop real-time middleware services in isolation, but real-time guarantees are quickly lost as additional services are added. The lack of composable middleware services capable of providing end-to-end hard/firm real-time guarantees limits is an obstacle to the use of low-level distributed control. Future work is moving in this direction Additional CSOnet Developments q q n Deployment of Actuation in South Bend System (summer 2009) Two additional Indiana cities are installing CSOnet Monitoring and Control of Civil Infrastructure q q Bridge monitoring Leak detection in water distribution networks September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

CSOnet Additional Projects Fort Wayne, IN n n CSO real time monitoring CSO control In-line storage Multiple Interceptors Greenfield, IN n n CSO real time monitoring Infiltration localization Hoboken, NJ n n Feasibility study Flood prevention (reversed flows) Omaha, NE n n Feasibility study Forced main September 26, 2008 Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC

Homeland Security: Air • Protection against release of chembio agents in urban environments • Goal: determine release point and plume expansion using embedded sensor network SF 6 wireless sensor Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 49

Homeland Security: Water Turbidity Sensor • Rapid detection of variations of water quality in water distribution systems • Standard off-the-self sensors: • Temperature • p. H • Turbidity • Conductivity • Dissolved Oxygen • Pressure Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 50

Structural Monitoring • Multi sensor • Vibration • Strain • System identification problem • RFID tag in load triggers network • Decentralized processing Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 51

Under development systems for water security • Multiple pathogen concentration system • Centrifuge technology can concentrate arbitrarily large volumes of water in 100 m. L format • Protozoa, bacteria, and virus • Virus capture using positively-charged filter • Trigger mechanism using wireless sensors • Internet access Michael Lemmon Dept. of Electrical Engineering University of Notre Dame Luis A. Montestruque Em. Net, LLC 52