A number of organizations, such as the Washington Metropolitan Area Transit Authority (D.C. Metro), are installing networks of gas sensors to detect and mitigate the effects of terrorist chemical attacks. Many companies, universities, and government agencies are developing novel sensor networks to this end. Our team aims to determine whether the current state of this technology is mature enough to meet its goals of increased public safety.
Over the past year, we have been doing background research on sensor and sensor network technology. This year, we plan to build our own gas sensor network. We decided to use surface acoustic wave (SAW) sensors, which can detect a wide variety of toxic gases at very low concentrations. We want to find a commercial gas sensor manufacturer willing to donate SAW sensor elements. We will then build a network out of these sensors and test how well it can detect non-toxic simulated chemical attacks.
Gas sensor networks have a wide variety of applications in environmental and safety monitoring that can be very useful to businesses and the general public. Environmentalists can use sensors to measure atmospheric pollution and monitor industrial emissions, and safety monitors can use sensors to detect harmful chemical vapors and explosives in public spaces, government and military facilities, and chemical processing plants. The Washington, D.C. Metro system, for example, has been testing a biochemical sensor system in the wake of recent anthrax scares. However, gas sensor technologies are still developing and have yet to reach their full potential in capabilities and usage.
The scientific community provides a myriad of methods for sensing chemicals. Some technologies are particularly accurate but prohibitively expensive for large-scale networking, such as thermal neutron activation, neutron scattering, x-ray imaging, and photoionization. Our team has chosen to focus on the less expensive and smaller scale technology of solid-state sensors. These come in many shapes and sizes, from electrochemical sensors to polymer film micro-balances and surface acoustic wave devices. They have the advantage of detecting a wide array of gases, and they respond very quickly to stimulants.
Sensors that detect a wide array of toxic gases are desirable, but they frequently detect a large variety of non-toxic chemicals as well. Solid-state sensors suffer from this difficulty, and sensor network designers must create processes that mitigate the possibility of such false alarms. We aim to confront the problem of false alarms with as high a priority as detecting real attacks.
By using a sensor network, the problem with false alarms could potentially be reduced. Multiple outputs can be compared for a more accurate analysis. By using advanced learning algorithm, the network can be trained to recognize different scenarios and false alarm signatures. The network provides exponential complexity in outputs as oppose to a few individual sensors output.
We propose to deploy a system of networked gas sensing elements throughout a suitably chosen indoor test site, and perform a series of controlled releases of gas simulants to test the functional parameters of the system. Our deployment will consist of three phases: a setup phase, a deployment phase, and a cleanup phase.
Knowing the geometry of the test building, we can assume the air in each room is well-mixed and do a material balance on the simulant release. The resulting differential equations can be integrated to watch the spread through the building and out the windows. This will give us an idea on what to expect from our network. When the network is operational, we will fine-tune the model, and then invert it as part of Objective 2, so that we may infer release information from observations, rather than observations from release information.
In research authorized by the Army at the Aberdeen Proving Grounds in Aberdeen, Maryland, the ADP2000 (an advanced sensor by manufactured Environmental Technologies Group) triggered an alarm for the extremely toxic nerve agent VX when exposed to a 0.1% concentration of Windex vapor (with no VX present). Using these results as a guide, we will release 2 cubic meters of vapor (Windex or another chosen simulant) for every 2000 cubic meters of air in the testing facility, thereby achieving a similar concentration level. If our sensors do not detect the simulant, we will release a 1% concentration. Other potential test simulants include ammonia, antifreeze, and floor wax.
At each sensor location, we will need to pass the raw sensor signal through several layers of electronics to transmit the signal to a computer. Below is a list and explanation of a proposed communications module for each sensor.
The SAW sensors are most likely to be driven by a high frequency signal (~200MHz) and will output analog voltage signals with a frequency that differs from the driver frequency. To process these signals, each sensor needs:
To simplify network setup, each sensor must send its output through a standard Ethernet cable and in the form of packets that follow the UDP protocol. Adhering to these standards allows us to use a standard Ethernet switch to route packets coming from multiple sensors through multiple cables. This switch will then route these packets through one cable to the computer for data analysis. That is, by increasing the complexity of each individual sensor module, we can greatly decrease the difficulty associated with aggregating the data in a central processing location.
Thus, for networking each sensor needs:
Furthermore, each sensor needs:
Here is a list of probable components needed, with prices:
|Frequency to Voltage Converter (National LM2917)||$5|
|ADC (Texas Instruments ADS8 series):||$7|
|MPU (Motorola 68000 series):||$10|
|Network controller (Epson S1S60000):||$10|
|PHY with an Ethernet connector (Intel LXT9763):||$15|
|25 MHz clock crystal and driver (estimate):||$5|
|200 MHz clock crystal and driver (estimated):||$5|
|Circuit board, wiring, power, electrical components, etc. (estimate):||$10-20|
If DSP processing is required, we will also need a separate DSP chip In this case, we could use an MPU with a built-in DSP, such as the Hitachi SuperH-3. This alternative would add about $10 to the total cost of the sensor.
The setup phase will consist of installing the gas sensors and networking components, as well as the data acquisition system for monitoring their output. We will plant sensors at key points throughout the building. In the case of the Maryland Fire and Rescue Institutes College Park burn building, a multistoried structure with on the order of 1000 sq ft per floor (and currently our first choice as a test site), sensors could be placed in several different arrangements. Starting with one sensor in the middle of each wall of each floor above the first, we could experiment with such permutations as removing an arbitrary number of sensors to simulate malfunctioning sensors. Theoretically, the more densely the sensors are placed throughout the structure, the more precise and error-tolerant the system will be.
The sensors will be connected with a hard-wired system of cables leading to a central data-acquisition station, located on the first floor, where a personal computer will run Matlab and/or Java software we write to gather, record, and analyze the sensor output. Hopefully, this analysis will be displayed in real-time, or at least faster than the frequency of the sensor output (approximately 250 MHz). Our processing needs may increase as the complexity of our computer analysis increases.
During the deployment phase, we will run several trials to simulate the release of dangerous gases into the building. These trials will make use of a chemical simulant, such as Windex, which is chosen to mimic the behavior of dangerous gases without posing a serious risk to the testing environment. In each trial, a predetermined amount of simulant will be released in a controlled fashion in a controlled location, and the reactions of the network will be monitored, recorded, and analyzed.
Trials will consist of controlled releases of various known concentrations of simulant vapor in various locations. After each trial, the system will be cleaned and reset to ensure the validity of subsequent trials.
The cleanup phase will involve removing the simulant from the testing environment (i.e., mopping the floors, etc.) and dismantling the sensor network (taking down wiring and sensor elements and removing the data acquisition system).
|Fall 2002 Budget||Spring 2003 Budget|
|Item||Cost ($)||Item||Cost ($)|
|Smoke detectors||40||Gas sensors||0|
|Splicing tools||0||Used Laptop||600|
|Ethernet cable 100 ft||80||Ethernet hub||50|
|Sensor electronics (outlined above)||80||Ethernet cable||250|
|Smoke generating supplies||30||Sensor electronics||420|
|Misc. supplies||50||Misc. supplies||50|
|Borrowed EEPROM burner||0||Matlab student version||100|
We will begin network construction by creating a single communications module and attaching it to a smoke detector. A smoke detector will behave similarly to a gas sensor, and will output an analog voltage. Thus, we can create and test a prototype communications module without waiting for acquisition of actual chemical sensors. After we have tested our communications module, we can create a network of smoke detectors and use smoke as a chemical simulant to begin testing data analysis.
Once we have created a prototype network and received gas sensors from a willing sensor manufacturer, we will replace the smoke detectors with gas sensors. We should, in theory, easily transition to the chemical sensors, as they will provide output in the same form as the smoke detectors.
Domestic Preparedness Program: Testing of ADP2000 chemical warfare agent detector against chemical warfare agents: Summary Report
Delphian Corporation: Electrochemical Sensors (reprinted from The Gas Monitoring Handbook, Anderson and Hadden, Avocet Press Inc, 1999)
Department of Energy's Characterization, Monitoring, & Sensor Technology Crosscutting Program: 1994 Technology Summary: Sensing Of Head Space Gases: Continuous In Situ Monitoring Of Gaseous Components In Underground Storage Tanks Using Piezoelectric Thin Film Resonator Sensors
Epson Introduces Intelligent Network Controller with Built-in TCP/IP