The ability to monitor the air quality in a closed environment, such as the Shuttle, the International Space Station (ISS), and future human missions to Mars or the Moon, is important to ensure health and safety of astronauts. Postmission analyses of grab air samples from the Shuttle have confirmed the occasional presence of onboard contaminants. Accordingly, a need exists for a lightweight, low-power, miniature instrument that can monitor contaminants at trace levels in real time. One promising technology is the Electronic nose (E-nose). Commercial E-nose instruments are now available, and several are being evaluated at the Kennedy Space Center for space program applications (figure 1).

An E-nose consists of an array of nonspecific vapor sensors. Typically, each individual sensor responds to a broad range of chemicals, albeit with a unique sensitivity relative to the other sensors. Upon exposure to a test vapor, the pattern and magnitude of response across the sensing array is compared to previously stored response patterns. The E-nose is selective because individual vapors induce unique response patterns. Representative response patterns for hypergolic fuels (i.e., hydrazine [HZ] and monomethylhydrazine [MMH]) are presented graphically in figure 2. Although chemically similar, the patterns for HZ and MMH are clearly distinct. The collection of data for the E-nose library and the development of mathematical models for using that data for identification of a test vapor are known as “training.” A properly trained E-nose could provide notification of sudden adverse events, such as leaks, spills, or even fire. The program at KSC is evaluating the E-nose technology for monitoring organic vapors and other chemicals in breathing air.

One critical parameter for space applications is the detection and identification of hypergolic fuel. Present allowable vapor levels in breathing air are set at 10 parts per billion (ppb). One particularly challenging application is the detection of hypergols in the Shuttle airlock at these levels. During space walks, the Orbital Maneuvering System is controlled by a Hypergolic Propulsion System. Prior to reentry to the crew quarters cabin through the airlock, it is important to verify that no residual vapor is present. This must be done at the operating pressures of the airlock, which range from about 3 to 15 pounds per square inch (psi). Although numerous monitors exist for hypergolic fuel vapors at higher concentrations, few technologies have been identified that reliably respond at this low concentration. Thus far, the Kamina is the only commercial E-nose technology that can readily respond to hypergolic fuels at this level (figure 3).

To train the Kamina E-nose to hypergolic fuels, over 50 individual exposures were performed at pressures ranging from 3 to 14.5 psi and relative humidity ranging from 5 to 80 percent for concentrations in the 10- to 1,000-ppb range. Using the vendor-supplied modeling software to perform a principal component analysis (PCA) followed by linear discrimination analysis (LDA), a model was created that assigned individual compounds to well-defined regions in two-dimensional space. This demonstrates that the Kamina should identify these vapors. In addition to developing models using vendor-supplied software, there is an active in-house program to develop algorithms to identify which sensor elements within the array provide the most information, to identify the best procedure to extract information from the data, and to determine the best classifier for the application.

Key accomplishments:

• Evaluated four commercial E-nose technologies for low-level detection of hypergolic fuels and identified at least one instrument (Kamina) that can readily detect HZ and MMH at 10 ppb.
• Using the Kamina E-nose, collected a training set composed of over 50 independent hypergolic fuel (HZ and MMH) measurements at concentrations ranging from 10 to 1,000 ppb. The effects of ambient relative humidity (5 to 85 percent) and pressure (3 to 15 psi) were included in the training set.
• Developed criteria to extract analytically significant information (e.g., features) from the training set to develop models with improved identification efficiencies. Using the newly developed feature extraction method, self-validation of the training set predicts an improved probability of identification near 90 percent.
• Obtained comparable identification efficiencies with a second training set composed of five common volatile organic compounds.
• Demonstrated that both short-term and long-term exposures provide similar identification information.