OVERVIEW

This laboratory test bed project was completed this year and focused on the concept of creating real-time decision-making algorithms in support of automating embedded networked sensing (ENS) design in an environment-specific context.  The design algorithm used to demonstrate the concept combines a GA to solve the combinatorial optimization problem associated with identifying the best combination of limited number of sensor locations for identifying a heat source location, and a descent-based inverse modeling algorithm for updating the location estimate. While such algorithms have been used in computational experiments and for post priori analyses, their integration into a real-time ENS design algorithm based on the environmental medium in question is new.  The algorithm was validated for the case of a single heat source.  The algorithm failed to correctly identify the presence of dual sources, pointing to the need for incorporation of additional intelligence in the ENS design algorithm.  Overall, the results of this demonstration point to the need for collaborative research between artificial intelligence, distributed networking, and environmental systems investigators to better couple learning and decision-making algorithms with realistic distributed parameter environmental simulators in the context of creating robust ENS designs for a variety of applications.

APPROACH

This project was carried out in a controlled, model soil system developed previously. Intermediate-scale models are laboratory-fabricated so that the medium’s geometry, however complex, is well-known.  Thus, these systems provide us with a realistic yet definable test bed in which to test (a) real-time environmental simulator calibration (parameter identification) and (b) dynamic sensor network design algorithms aimed at achieving optimal coverage for a given environmental objective.

SYSTEMS / EXPERIMENTS

The experiments are being executed in a physical aquifer model fabricated to allow controlled release and monitoring of pollutants in flowing groundwater. The aquifer test bed consists of a 1 x 0.5 x 0.4 m tank containing a water-saturated sandy porous medium.  The DAQ system has been for temperature sensors for the most part, although ore recently we have begun to test COTS and UCLA-fabricated nitrate sensors as well (see below).

• The experimental sequence proposed for the first project is as follows:
• Heat transfer experiments with distributed, networked thermistors.
• Hydrodynamic tracer studies using distributed, networked nitrate microsensors being developed in CENS.
• Dense nonaqueous phase liquid (DNAPL) emplaced source dissolution studies using distributed, networked halogenated organic solute microsensors.

(i) Parameter Identification Algorithms.  Off-the-shelf thermocouples were deployed in a 3D grid in the physical aquifer model (Figure 1) to sense temperature changes in a porous media produced by a continuous, point heat source under steady and unidirectional flow conditions.  Numerous data sets describing T(x, y, z, t) were acquired varying flow rate in the system.

Figure 1.  Plan view of the experimental layout and potential sensor locations for exercising the monitoring network design algorithm.

 (ii) Sensor Network Design Algorithm.   Earlier this year we reported on a real-time estimator for the transport parameter and source identification for the heat transport experiments was developed and integrated into the LabVIEW-based DAQ system using Matlab.  The Levenberg-Marquardt optimization algorithm was used to determine the unknown transport parameters or determine the source location.  The estimator was validated through the heat transfer experiments described above.  Typical results are shown in Figure 2 below, which demonstrates the convergence of the parameter estimation over time (as more data are acquired).  
 
Figure 2.  Results from the real-time source estimation algorithm executed 120 minutes into a heat transfer experiment.  The question of sensor network coverage is beginning to be addressed here:  symbols depict estimated source location for the given number of sensors employed in the estimate.
The model parameter identification problem was adapted to address the more relevant sensor network design problem.  This problem poses the question of where sensors should be located to minimize model prediction uncertainty for a given budget and collected data.  This combinatorial optimization problem is approached using a GA (Genetic Algorithm).  A GA-based heuristic evolutionary algorithm is used to find the optimum set of sampling points (number and location), to estimate the parameter values or parameter structure (in case of heterogeneous medium), and to identify the location and dimension of source.

Figure 3.  Optimal sensor network design for the purpose of identifying the source location; design proceeded sequentially (2 sensors at a time) toward a total of 8 sensors.

Figure 4.  Layout for the physical aquifer model test of the dynamic response of carbon-fiber based nitrate sensors (sensor size here roughly 0.1 mm diam. by 0.5 cm length). Red point represents source and locations 1 – 4 sensor deployments downstream of the source.  
 
 (iii) Evolution toward contaminant transport experiments.  Progress toward this objective has been carried out in concert with the CENS Sensor group.  As noted in the discussion of problems above, we identified an incompatibility with the nitrate sensors and the NI DAQ system in December 2003 and have since modified the NI system.  Preliminary data obtained from a dynamic testing of a single nitrate sensor is shown in the figure below.  The work on nitrate sensors themselves is discussed in more detail in the context of project Error Resilient Sensor Array below.

Figure 5.  The observed sensor responses at locations 1 – 3 showing sharp, non-attenuated signals near the source and diffuse fronts further downstream (sensor #4 failed to yield a reasonable response).

MAJOR ACCOMPLISHMENTS AND RESEARCH PARTNERSHIPS

The following milestones were achieved in support of the completed project Contaminant Source Assessment:

• Created real-time simulator parameter estimation algorithms and tested them in a laboratory test bed; this work was published .
• Created real-time decision-making algorithms in support of automating embedded networked sensing (ENS) design in an environment-specific context;
• A laboratory-scale investigation is being undertaken in collaboration with Prof. Steve Bryant (U. Texas - Austin) who is will use CENS sensors to study hydrodynamic dispersion in porous media.  Bryant, along with U. Texas colleagues Larry Lake, and Russell Johns have acquired project support from the Department of Energy.

PEOPLE

Faculty:

Prof. Thomas Harmon, School of Engineering, UC Merced
Prof. Deborah Estrin, Computer Science, UCLA
Prof. Miodrag Potkonjak, Computer Science, UCLA
Prof. Jose Saez, Civil & Environmental Engineering, LMU
Prof. Jenny Jay, Civil & Environmental Engineering, UCLA
Prof. Steve Margulis, Civil & Environmental Engineering, LMU

Staff:

Dr. Juyoul Kim, Civil & Environmental Engineering, UCLA
Dr. J. Eric Haux, School of Engineering, UC Merced
Mr. Mohammed Rahimi, CENS – UCLA

Graduate Students:

Ms. Yeonjeong Park, Civil & Environmental Engineering, UCLA
Mr. John Ewart, Environmental Systems, UC Merced
Mr. Tom Schoellhammer, Computer Science, UCLA
Mr. Naim Busek, Computer Science, UCLA

Undergraduate:

Mr. Juan Soriano,  Computer Science, Merced College
Mr. Obimdinachi Iroezi, Computer Science, UCLA
Ms. Mallory Davidson, Chemical Engineering, U Washington (2004 REU)
Ms. Nicole Jurisch, Chemical Engineering, U Washington (2004 REU)
Mr. Ryan Tamashiro, Civil & Environmental Engineering, LMU
Mr. Daris Cook, Civil & Environmental Engineering, LMU
Mr. Wudineh Woldtensay, Civil & Environmental Engineering, LMU