PROJECT NAVIGATOR, LTD. Landfill Operations: Designing and Using Digital Data Collection Systems to Predicatively Operate a Landfill as a Large- Scale Bioreactor Presented by Halil Kavak, PhD, Project Navigator, Ltd. Raudel Sanchez, PhD, Project Navigator, Ltd. Ian A. Wester, ScD, Project Navigator, Ltd. Theodore Tsotsis, PhD, University of Southern California Mohammed Shahimi, PhD, University of Southern California SWANA LANDFILL GAS SYMPOSIUM March 11 2009 Atlanta, Georgia www.projectnavigator.com
Project Navigator, Ltd. Project Navigator has six main offices. Brea, CA (714) 388-1800 Pleasant Hill, CA (925) 969-9574 Seattle, WA (206) 390-3948 Houston, TX (713) 468-5004 Malvern, PA (610) 251-6851 Raleigh, NC (919) 539-1928
Compliant Landfill Operation: Standards to be Achieved Landfill Perimeter <10-4 excess cancer risk At Cap Surface < 50 ppm CH 4 HI < 1 GP GWMW <5% CH4 <MCL s Compliance standards envelope the waste prism
Drive and Control Compliance Via LFG Extraction Thermally treat Goal: Set vacuum to match gas extraction rate to gas generation rate Threat:Too much vacuum leads to air intrusion, and can exacerbate e EOZ conditions o
Why should we be concerned about EOZs? Reason Cost Impact 1. Loss of gas collection wells Replacement cost 2. Reduces gas collection efficiency 3. Reduces effectives of gas PLC in the area Potentially no cost impact Additional cost to maintain compliance 4. Increases landfill subsidence Cost for slope & cap repair 5. Increases risk from seismic Repair and replacement cost activity 6. Health, safety, & fire threat Repair and replacement cost
Knowledge / System Size Vs Time Remedy Construction Completed in 2000 Today, 2008 Knowledge Know wledge / LFG Syst em Size Existing Gas Extraction System Capitalize on this Delta to Achieve Cost Reductions Time
Main Components of Digital Site Site Monitoring Display System Decision Making System Remote Data Collection Neural Network Genetic Algorithms ArcGIS EVS Global Mapper WiFi Tl Telemetry OII Northeast enhanced oxidation problem Map temperature profile and vacuum at extraction wells Control pressure to get optimal temperature distribution Finding pathways of air intrusion Visual managing data efficiently Providing tools to operators to keep the site at optimum operational conditions
Remote Data Collection Extraction, Monitoring Wells, and Gas Probe Properties (Location, Depth,..)
Pilot Project Wi-Fi Network Gateway (Base Station) Power grid connection Mounted at the office Connected to Internet Receives sensor network data carried over Wi-Fi network Wi-Fi Extender (Repeater) Solar-powered 7-10 mobile platform Receives data from wireless sensors Wireless Sensors Rapidly deployed Transmit to Wi-Fi network Wireless Sensor Area
Possible Wi-Fi Applications
Evaluation of Data Collection Techniques
Temperature Sensor Installation Wireless Transmitter Probe Head Installation of Sensor in Well Installed Temperature Probe
Wi-Fi Remote Monitoring System Interface Real-Time Online Temperatures (Well IV-5DR) Instrumented Well Locations Temperature (F) 150 145 140 135 130 125 120 115 110 105 100 95 90 85 80 75 70 65 60 55 50 9/2/2007 0:00 9/3/2007 0:00 9/4/2007 0:00 9/5/2007 0:00 9/6/2007 0:00 9/7/2007 0:00 9/8/2007 Date Historical Well Temperatures (09/02/07 to 09/07/07) Extracted from Wi-Fi Sensor Database (Well IV5-DR) Legend 30 ft below surface 55 ft below surface 80 ft below surface 105 ft below surface
Data Analysis s by Utilizing Prediction Tools Prediction Tools Decision Making and Corrective Action
Description of the Problem Landfills are highly heterogeneous porous media Complex phenomena, including flow and transport b of gases and moisture, biodegradation, d and nonlinear reactions Landfills are large-scale bioreactors The landfill is presented by a three-dimensional computational grid The blocks are cubical, but do not have the same size The model contains a number of extraction/monitoring wells Since a landfill is a porous medium, each block has its own permeability tensor, porosity and tortuosity factor Due to compaction, the vertical permeabilities are smaller than the horizontal ones, and increase from the bottom to the top Native Groundwater d Gas probe Landfill Gases Exert a Partial Pressure on the Groundwater Table, Leading to Gas Absorption and GW Impacts e Cap settlement f c Enhanced Oxidation Void Zone, which Translates itself into Settlement at the Landfill Cap CH4, CO2, VOC s Impacted Groundwater (GW) CH4 a Landfill gas extraction CH4
Landfill Biodegradation Modeling Classification of Wastes Three classes of wastes : readily Gas generation rate α k (t) of gaseous biodegradable, moderately biodegradable, species k: least biodegradable it Monod equation for biodegradation: k t C k Ai ie, t t 0 t f i 1 d C, k : total gas generation potential of gas k s dt A : fraction of component i in MSW s i ψ - concentration of the substrate λ i : gas generation constant of I Ф concentration of micro-organism t o : time since cover was placed κ - maximum rate of substrate utilization t f : time to fill the landfill κ s - the half velocity coefficient L z : landfill depth 3 Z L z
Governing Equations and Iterative Procedures Governing Equations Four components CH 4, CO 2, O 2 and N 2 Darcy law is assumed Iterative Procedure Finite-volume method is used to solve the equations Governing Equations and Iterative Procedures Darcy law is assumed Convective-diffusion reaction CDR equation governs concentration of gases The CDR equations for the gas q FV allow the use of a non-uniform grid Conjugate-gradient method and forward discretization of time-derivatives are e C equat o s o t e gas component k: z C D z y C D y x C D x t C k k k k k k k used to solve the equations ( z ) C z p k z C y p k y C x p k x k k m z k m y k m x Top view of computational grid CH4 CO2 CH4 CO2 CH4 CO2
Optimization Complexities The permeability, porosity, tortuosity, and gas generation potentials ti vary spatially over several orders of magnitudes. Due to a variety of factors, the amount of experimental data that characterize the properties of a landfill is severely limited. Given a computational grid that represents a landfill, a large number of transport and reaction equations must be solved. The equations are highly nonlinear. Serial computation is not effective, and in most cases impossible. ibl Parallel computations are needed. The Genetic Algorithm is used for the optimization problem.
Optimization Based On A Genetic Algorithm Time-dependent methane concentration profiles at some extraction wells are taken as the data. Synthetic data are generated to validate the algorithm. Massively-parallel computations using 180 processors with message-passing interface are used. The objective function is, F j CH 4,exp j CH 4,mod j 2 GA has four main elements: Selection: for generating a solution Design of the" genome : to constrain the variables, and the generation of the phenotype the model of transport and reaction. Crossover and mutation: for generating new solutions and approaching the optimal one. Eliticism: to select those solutions that eventually lead to the true optimal solution.
Optimization Based On A Genetic Algorithm Random initial guesses for the spatial distributions of the permeability, porosity, tortuosity factor, and total gas generation potential. Solve the governing g equations and compute those properties for which data are available. Evaluate the objective function Check whether convergence has been achieved. If not, use selection, mtation mutation, crossover, and eliticism to update the parameter space. Repeat until the true optimal distributions are obtained. λ=900 λ=700 λ=300 Next Generation GA Flowchart Create Initial Design Populations Evaluate Obj. Function of Designs Select and Reproduce (Create New Designs) Replace Designs of the Old Populations with New Designs Stop? old new
Comparison of Data and Optimal Profiles Over 18,000 parameters are optimized. It took 180 CPU hours to compute the optimal distributions. The processors were Pentium-4. Comparison of data and optimal gas profiles Comparison of data and optimal permeability distribution
Artificial Neural Networks for Landfills? ANN mimics the human brain (neurons are inter-connected to allow the brain make decision on the input). Inputs to ANN are inter-connected to discover relationships between the input variables ANNs are recognized as universal approximators. Able to capture trends in agiven set of data. Capable of forecasting the behavior of a system, given reasonable amount of data. However, the predictions are not based on a specific physical model of the system, and the phenomena that occur there.
ANN for Forecasting the Behavior of the Landfill Historical landfill gas data (T, P, Gas Consents) A section of the landfill was used in the study. 32 Wells are located in this zone. 60% of dt data was used to train ANN Forecasted T, P, CH 4,CO 2 and O 2 distribution The predictions help: Operators to make quick and effective decision for the short term. Operating the wells at the optimal vacuum conditions to avoid potential fires and optimal gas quality. Calibrating the site condition for long term landfill stability.
How Does an ANN Work? CH4, CO2, O2, P, Well 1 CH4, CO2, O2, P, Well 2 CH4, CO2, O2, P, Well 3 1 n 1 Σ F1 Σ F2 b11 1 n 1 b 2 n 1 Σ F1 Σ F2 b12 1 n 2 b21 b22 2 n 2 Σ F1 Σ F2 b13 1 n 3 b23 2 n 3 T @ well 1 T @ well 2 T @ well 3 CH4, CO2, O2, 1 2 P, Well 4 Σ n 4 F1 Σ n 4 F2 T @ well 4 b14 b24 CH4, CO2, O2, 1 2 n P, Well 5 Σ 5 n F1 Σ 5 F2 T @ well 1 b15 b25
Artificial Neural Network 20 hidden layers with a Tansig transfer function are used. Data are separated into training, validated, and testing. The output layer is evaluated. The performance function P is minimized. Weight and biases are updated using a back propagation algorithm All the calculations require less than 100 iterations. W b n a P m i m i m ij new m i new N m 1 1 N W b m ij i 1 F m ij old old m N i 1 W n m ij m i 2 a 2 a m 1 j b m i 2 M M m F n t a i T Actual T Cal m 1 j i M M m F n t a i i i i n M i n M i
CH 4 Forecasting
CO 2 Forecasting
Temperature Forecasting
Conclusion for ANN and GE Modeling Genetic algorithm can be used to correctly predict the spatial distributions of the morphology of a landfill. However, GA requires a significant amount of CPU and can only be run on highh performance computers As an alternative, artificial neural networks can be used to get quick estimates and forecast for the future, short-term, behavior of a landfill. One should be able to combine the two to develop a predictive tool for the short-term, as well as long-term behavior of a landfill.
Data Visualization at Automation o Visualization Tools Visualization 250F 180F 130F 100F 75F
Enhanced Oxidation Zones 250F 180F 130F 100F 75F T>135 F T>155 F
Gas Well Vacuum Distribution Vs. Gas Well Temperatures Legend Temperature ( o F) 135 257 (max T measured) Well Head Vacuum (inch Water) Enhanced Oxidation Zones
3D View of Possible Hot Zone Recent Cracks Possible Hot Zone Extension
35 Extraction Wells Legend Sampled Extraction Wells Non-sampled Extraction Wells
36 Methane Concentration Distribution Legend CH4 Concentration, % CH4 Concentration at Gas Probes, % Well Location Probe Location
37 Vacuum Distribution Legend Well Head Vacuum (Inches of Water Column) CH4 Concentration at Gas Probes, % Well Location Probe Location
Settlement Forecasting Note: Change in elevation between March and September 2004 Legend Control Points Elevation Change in ft -0.820 --0.720-0.710 - -0.620-0.610 - -0.520-0.510 - -0.410-0.400 - -0.310-0.300 - -0.210-0.200 - -0.110-0.100 - -0.005005-0.005-0.010 0.100-0.200