Subscribe to receive more articles like this: Print/digital

Software for predictive modeling of toxin migration and lethality within building structures to optimize the placement of CB sensors in buildings, transportation hubs, and other public venues

A software tool for designing and implementing an optimized sensor network is needed to monitor and respond to unsafe environmental conditions within buildings. The Sensor Network Design Tool (SNDT), developed by Applied Nanotech, Inc., incorporates building-air transport models and selectable probability distribution models integrated with databases of gas sensors and their properties, as well as detects a broad spectrum of hazardous contaminants including toxic chemicals and chemical and biological (CB) attack agents.

This SNDT was originally developed with funding from the Department of Homeland Security (DHS), which was seeking a system to designate where to install CB sensors in buildings, transportation hubs, and public venues with the goal of obtaining the shortest time to detection and the most comprehensive coverage for a given number of sensors. Deciding where to place a network of sensors in a sizable structure is a daunting task compounded by a current lack of standardization on methodology for evaluating system performance. Despite the DHS requirement that the SNDT provide for rapid and verifiable deployment in a wide spectrum of CB threat scenarios, it has to be accessible to non-expert users.

These inherent capabilities of SNDT make it directly applicable to the design of sensor networks for industrial and commercial settings, where toxic chemicals are handled and leaks or spills can occur, such as: semiconductor fabrication, pharmaceutical, chemical, petroleum, building construction, nuclear, or defense facilities. In additional to accidental release scenarios, SNDT can be applied in indoor air quality, contaminant migration, or ventilation system performance assessment to provide automated capability for the quantitative evaluation of airflow and contaminant movement in complex situations.

Figure 1 shows the operational structure of the SNDT. Through a series of design steps and menus, a system operator is guided through the design and verification protocol. Beneath its graphical interface the SNDT contains a processing algorithm combining a multi-zonal air transport engine with constrained non-linear optimization. This software generates and analyzes a multitude of possible sensor networks and toxin release scenarios and searches for optimal sensor placement. The output is a visual representation where the sensors should be located, along with numerical data comparing the networks and scenarios analyzed.

The major software components comprise:

1. Front end interface where the operator is interrogated and inputs desired agents for the network to sense along with a building model.
2. Multi-zone building airflow and contaminant migration engine to calculate agent transportation yielding concentration densities as a function of time and space.
3. Sensor network design module that generates candidate sensor networks by selecting appropriate sensors from the database and distributes them about the building, then evaluates possible release scenarios using the multi-zone building airflow and contaminant migration engine to determine time to fi rst detection, or if detection occurs at all.
4. Sensor database is a database of sensors for various threat agents or toxic industrial compounds and includes agent properties (molecular weight, spore size, lethal dose, incapacitating dose, etc.), and sensor sensitivity, response time, cost and associated engineering data.

Multi-Zone Building Airflow and Contaminant Migration Engine
ANI collaborated with Lagus Applied Technology, Inc. (LAT, http://www.tracergas.com) on the predictive modeling of toxin migration and lethality effects within building structures. Prior to SNDT development the company had extensive expertise in CB sensor technologies; LAT performs modeling and measurement of contaminant migration and building air transport, and had already developed CB-Protect, a software tool that became the foundation of the SNDT.

CB-Protect is a multi-zone building airflow and contaminant migration engine. Zonal models treat the building as a set of volume zones, typically being rooms, hallways, stairwells and HVAC ducting, and employ coefficients linking each zone to all others, which physically represent air flow rates. The flow rate matrix has discreet variants representing for example, HVAC blowers being on or off , doors open or closed, etc., and can have continuously variable time dependent matrix coefficients. Each unique matrix defines a building state. The coupled rate equations are then solved using standard numerical integration and matrix techniques to provide the temporal and zonal evolution of the agent concentration throughout the building.

Zonal models require an initial building description to be input. Though the actual flow rate matrix can be established experimentally by tracer gas studies, the SNDT employs CAD style building models to minimize equipment requirements, using tabulated ASHRE data of leakage rates through the various construction materials (sheet-rock, cinder-block, concrete, etc.) and HVAC blower/ducting throughputs. The flow rate matrix values and volume zone description completely specifies the building and its air flow properties. Then, with the building and fl ow rate values established, the airflow and contaminant migration engine is used to analyze simulated toxin release within the building.

Sensor Network Design Module
The SNDT and CB sensor database software modules are integrated with CB-Protect. The Sensor Network Design Module (SNDM) is the core of the SNDT. It generates candidate networks by selecting sensors from the database responding to the desired agents. The permutations of m unique sensors distributed throughout the n building zones are then successively examined for detection performance. The number of sensors m ranges from 1 to a maximum generally determined by budget constraints. The definition of optimal network is objectively cast in terms of a time score. Initial development has focused on the averaged least time to detection T_d squared as the criteria to minimize,

where the index i runs over all possible unique sets of conditions and P_i is the probability that those specific set of conditions will exist. Specific facts must be speculated regarding an agent release: the type of agent employed, the quantity and duration of release and the release zone. The probability P_i can be considered as the product of the probabilities for each variable:

With this the full expression for the value of the minimization functional is:

The value of the functional f using equation (3) is now evaluated for all the candidate sensor networks. The sensor network with the minimum value of f is then considered the optimal network.

Choices for the probability distributions for the elements comprising equation (2) must be made, and each has unique considerations. In practical implementations, identifying all building states is infeasible and fortunately unnecessary; using a small number of representative building states yield results comparable to very detailed methods. The distribution for agent type should include all known agents in order to be comprehensive, but is simplifi ed because all gases are transported equivalently in the zone model. Therefore the distribution need only incorporate a single gas calculation weighted over the agent type and sensor properties.

The distribution for agent quantity is unknown but reasonable maximum quantities for terrorist attacks are what an individual or a vehicle, depending on release zone, can carry. In practice, an optimal sensor network for one specific quantity of an agent release will also be the optimal sensor network for a larger quantity of the same agent because the spatial and temporal transport profiles scale with the quantity, so the particular sensor that fi rst detects in one situation will also be the first to detect in the second situation, only with a shorter time to detection. What is required to be confirmed by the optimization algorithm is what the minimum release quantity that can be detected by the network under evaluation is, and does that minimum detectable quantity allow concentration levels above the lethal or incapacitating concentration.

The probability distribution for agent release duration is also unknown, but reasonable values are in the minutes to tens of- minutes time scales. The issue for this particular probability distribution is that a very short release duration can lead to lethal concentration levels unless the first to-detect sensor is in the release zone. On the other hand, a very long release duration will keep concentration levels low, but they could still be above lethal thresholds unless the detection sensor is in the release zone and has a sensitivity threshold above the ambient concentration level. The release zone distribution has several models to choose from. Candidates considered and evaluated include:

• Flat Distribution: where each zone is equally likely for a release with probability inversely proportional to the number of zones, or n-¹.
• Area Weighted Distribution: where the probability for release in a zone is equal to that zone’s area divided by the sum of all zone areas.
• Security Weighted Distribution: where zones that are secure or have limited accessibility can be assigned a small to zero probability for a release.
• Casualty Weighted Distribution: where the probability for release in a zone is proportional to the likely resulting casualties.

Probability distributions that incorporate compound strategies can also be considered; an easily accessible zone and likely large amount of casualties is a more desirable target.

Chemical and Biological Sensor Database
Reliably sensing the presence of a CB or toxic industrial agent with unattended sensors requires a diverse array of devices, as no single sensor can detect all possible chemical and biological entities. Sensors range from the simple, inexpensive metal-oxide devices used for gas detection to expensive analytical instruments such as gas chromatographs and mass spectrometers. Inexpensive sensors tend to have poor selectivity, low sensitivity and correspondingly high limits-of-detection. Instruments that will unequivocally identify the substance present will be sophisticated and expensive, and may require technicians to operate, monitor and interpret the data.

The Chemical and Biological Sensor Database (CBSD) database contains existing and available sensors and their properties. The database holds fields for the sensor type, manufacturer, detectable analytes, limits-of detection, sensitivity to cross-contaminants, maintenance requirements/lifetimes, and cost per unit. The CBSD is accessible from the SNDM software module and completes the input information required to allow design of the permanent sensor network.

Figure 3. The probability-weighted time
score value for networks of 1 to 5 sensors.

Typical Program Output and Sensor Network Evaluation
Figure 2 shows example results from the SNDT. The icons indicate where the network of four sensors should be placed for optimal detection of a release. Figure 3 shows the probability weighted time score value for networks of 1 to 5 sensors. Increasing the number of sensors decreases the average detection time until a point of diminishing returns is reached. Ancillary data produced by the SNDT include the time score, the number of scenarios evaluated where no detection occurs, the total probability of no detection occurring, and maximum time to detection.

The SNDT has been evaluated in experimental trials. Gas concentration levels throughout each zone of a building were accurately calculated as a function of time. Probability distributions were devised for which zone the release would occur in, which CB agent was released, and its quantity. The algorithm then generated a subset of all possible sensor networks, ran the building modeling program to calculate concentrations for CB releases based on the probability distributions, then calculated the time to first detection. The network configuration with the probability-weighted least time to first detection was selected as the optimal network. The candidate networks were generated from sensors chosen from a database containing the sensor parameters (analyte sensitivity, time response, cost, etc.).

Conclusion
A new software tool is being developed to assist in the design of optimized sensor networks. ANI’s SNDT can greatly facilitate optimal sensor network design implementation via open and verifiable optimization algorithms, and is useful for commercial facilities where toxic gases and chemicals are handled.

Fitted with appropriate sensors, “smart buildings” have the ability to detect a release, determine where it originated and predict where and how it will travel, taking into account HVAC and building status. “Smart buildings” can even be fitted with actuators to close off ventilation or direct personnel to the safest escape routes. With a priori planning and the required infrastructure in place, SNDT offers the potential for reduced exposure hazard.

DONALD R. SCHROPP, JR. IS A SENIOR SCIENTIST AT APPLIED NANOTECH, INC., 3006 LONGHORN BLVD., SUITE 107, AUSTIN, TX 78758. HE PERFORMS PHYSICAL MODELING, AND DEVELOPS MEASUREMENT METHODS AND APPLICATION SPECIFIC DATA ANALYSIS ALGORITHMS ESPECIALLY TARGETED TOWARD THE RESEARCH AND DEVELOPMENT ENVIRONMENT OF CUSTOMIZED EXPERIMENTAL SETUPS AND INSTRUMENTATION. HE HAS WORKED IN THE FIELDS OF ATOMIC PHYSICS, SPACE SCIENCE, AND MOST RECENTLY AS SENIOR SCIENTIST AT CANDESCENT TECHNOLOGIES, INC., A LARGE SILICON VALLEY VENTURE TO PRODUCE FIELD EMISSION DISPLAYS. HE RECEIVED HIS PH.D. IN PHYSICS FROM YALE UNIVERSITY. HE CAN BE CONTACTED AT 512-339-5020 X129 OR DSCHROPP@ APPLIEDNANOTECH.NET.

To read the entire article, sign up for a free subscription and then click on
Click here for the digital edition of the current issue