Electronic Nose Technology Applied to Air Pollution from Solid Waste

Transcription

1 Electronic Nose Technology Applied to Air Pollution from Solid Waste Allison N. Cox and Aaron A. Jennings Department of Civil Engineering Case Western Reserve University Cleveland, OH ABSTRACT Electronic nose technology provides the attractive possibility of determining quantitative measurements of solid waste odors. This manuscript describes the results of efforts to use a Cyranose 320 electronic nose to quantify odors at a landfill excavation site in the City of Garfield Heights, Ohio as well as other solid waste related odors. Sampling procedures developed to overcome problems of headspace dilution and background contamination, which were experienced in the field, are also discussed. Results are presented to illustrate that under appropriate conditions, calibrations may be developed from raw electronic nose sensor data which can be used to identify variable concentrations of a single odor as well as aid in differentiation between component odors from a mixed source. KEY WORDS Electronic Nose, Solid Waste Odors, Gas Bag Sampling, Field Testing, Landfill Odors INTRODUCTION This paper will present results of a recent effort to apply electronic nose technology to air pollution in the form of solid waste odors. This was first attempted at a project involving the excavation and on-site re-deposition of over 1,000,000 yd 3 of solid waste deposited between 1970 and 1978 on 7 acres at the R&B Development Landfill in Cuyahoga County, Ohio.

2 Because of the size of this excavation and the potential for encountering industrial wastes, there was substantial concern about the odors that would be released. There was also frustration about the state of our ability to quantify odor. Although subjective measures exist, there are few procedures that are objective, quantitative, and reproducible. In response to this problem, a research project was conducted to examine the use of a Cyranose 320 electronic nose to track the development (magnitude, composition, distribution) of the odor signature during summer and fall excavations at the landfill excavation project. The Cyranose 320 uses electronic nose technology to generate smell prints from air samples, and onboard or post-processing algorithms to interpret results. Initial field data indicated that an electronic nose is able to capture and quantify characteristic solid-waste-derived smell prints. The field trail also produced improvements to field sampling techniques, which will be presented. The lessons learned from field trails have been applied to several sources of solid waste odor (transfer stations, composting facilities, dumpsters). The results help answer questions of how useful this technology will be in quantifying solid waste odors. Generally, electronic noses are designed to identify odor patterns, but not to quantify odor intensity. However, odor intensity controversies are at the heart of most solid waste odor problems. Results will also be presented to illustrate how one may use gas sample dilutions to construct concentration calibration curves to quantify odor intensity as well as differentiate between concentrations of more than one odor in a mixed sample.

3 ELECTRONIC NOSE TECHNOLOGY The human nose and sense of smell are sophisticated technologies. Inhaled air containing volatile chemical compounds is drawn into the nasal cavities in the upper respiratory tract. This portion of the nasal passage contains approximately 50 million sensory receptor cells located on the olfactory epithelium which covers an area of about 2.5 square centimeters (Cyrano Sciences, 2001; Leffingwell, 2001). These receptor cells react with the incoming volatile chemical compounds stimulating nerves to send impulses to the brain. In the brain, these impulses are processed and assembled into a pattern which is interpreted using what it has learned about the form of previous odor signals. Single cells do not identify odors. Instead, odor identification is the result of the collective responses of many sensors in the array combined with the brain s pattern recognition abilities. Odors are often made up for hundreds of chemicals, yet the olfactory sense is able to distinguish among a nearly infinite number of chemical compounds at very low concentrations (Cyrano Sciences, 2001). An electronic nose is a sensor technology designed to mimic the sense of human smell. The system uses an array of non-specific sensors to create an odor signal pattern, and a computer to analyze and interpret the results. This new technology has many potential applications. Current applications include uses in the food, chemical, medical, process control, defense and environmental industries (Cyrano Sciences, 2000a,b,c,d; Gibson et al., 2000). A Cyranose 320 from Cyrano Sciences was used in this research (see Fig. 1).

4 Fig 1 Features of the Cyranose 320 Electronic Nose. The Cyranose 320 uses an array of 32 polymer composite sensors. Each sensor is composed of a pair of electrical contacts connected by a composite film consisting of conductive carbon black particles blended throughout a non-conducting polymer. When the composite is exposed to a sample vapor it absorbs mass causing it to swell. The swelling of the polymer causes the connections between the conductive carbon-black particles to be severed which results in an increase of resistance across the electrical contacts of the sensor. When the sample is purged from the system the polymer releases the absorbed mass thereby returning the film to its original size and resistance. The baseline resistance of the sensor (R) is measured while a background vapor passes over the sensor array, and the maximum resistance (R max ) is measured as the sample vapor passes. The response from a sensor is measured as the nondimentionalized bulk relative resistance

5 ((R max R)/R). Since each sensor is comprised of a unique polymer and amount of black carbon a reproducible combination of resistances or smellprint for each sample is created. The Cyranose 320 can display the results of odor analysis in several ways. Up to 5 methods can be stored on the Cyranose at one time, and any number of methods can be saved and loaded from a computer. Each method stored on the nose can be used to identify up to 6 different odors. The sensors in the nose are non-specific, meaning that the nose must be trained on an odor source in order for it to identify that odor in future samples. This training is accomplished by sampling an odor at least 5 times, and recording the sensor array responses from these samples to build a representative smell print for that odor. The sensor array responses generated by future samples are compared to all the smellprints in a method to determine the identity of an odor. Once training is complete the nose can be used on a stand-alone basis to identify unknown odors. If an unknown odor falls into one of the training set classes the Cyranose will identify the class, to which it belongs, and will indicate how accurately it belongs to that class. When the Cyranose is connected to a computer additional information can be acquired. A scrolling strip chart showing the real-time response of all 32 sensors can be displayed and or recorded. This proves very useful in determining if and when a stable background reading has been attained, if a sample is producing a significant sensor array response, or if the sampling configuration (sample draw duration, purge duration, pattern, etc.) is adequate to allow sensors to reach equilibrium. Additionally, the software package uses the sensor array responses from the training sets to create a characteristic smellprint for each training class as well as a Canonical Projection Plot and a Principal Component Analysis (PCA) Plot. The smell print is a column

6 chart that displays the relative response of each sensor (R max R)/R) compared to the range of corresponding values for any of the selected method s training sets. The Canonical Projection Plot and PCA Plot illustrate where a sample falls relative to the clusters created by the training set (see Figs. 2, 3, and 4). These plots can also be used to assure that outliers are not included in a training set. For further analysis the resistance values of all the sensors can be streamed directly into a spreadsheet program which is particularly useful for developing odor magnitude calibrations. Fig. 2 - Illustration of a Scrolling Strip Chart During Odor Sampling Showing Real Time Sensor Response vs. Time.

7 Fig 3 - Example of a Cyranose 320 Smellprint Showing Similarity Between a Sample and a Trainnig Set. Fig. 4 - Example of a PCA Projection Plot Showing Grouping of Traing Classes.

8 THE GARFIELD HEIGHTS LANDFILL PROJECT Interest in the application of electronic nose technology to landfill odors originated with an 4/10/00 Ohio EPA Rule 13 Application seeking authorization for the excavation and on-site re-deposition of 600,000 yd 3 of solid waste at the R&B Development Landfill in Cuyahoga County, Ohio (McCabe Corp., 2000). This landfill operated from 1970 to 1978, ans it is estimated to have received 6 million yd 3 of residential, commercial, and industrial waste. The site was closed in August of 1978 by a Cuyahoga County Court order because of controversy surrounding operational practices at the landfill. The current owner of the site proposed moving a portion of the old waste because it would free access to developable land at an Interstate 480 exchange that would then become valuable commercial real estate. The city of Garfield Heights was supportive of the project because of the potential economic benefits and tax revenue. However, the city leadership and many citizens expressed concerns about the potential environmental problems associated with the project (Longo, 2000, 2001). Because the waste was deposited prior to RCRA regulations and because the site of the landfill was near Cleveland s major industrial areas, it was feared that some very unpleasant industrial wastes could be uncovered. Additionally, the contractor proposed a 24 hours a day, 7 days a week work schedule to complete the project within a Summer/Fall work schedule. There was concern that, due to the magnitude of the project, even the more benign wastes would produce serious odor problems. The homes that were once encircled by the landfill site were acquired by the developer and demolished, but there were residential and commercial areas located in close proximity to the landfill (see Fig. 5).

9 N Fig. 5 - Garfield Heights Landfill Site Showing Proximity of Residential and Commercial Areas. Note a Commercial Area to the West, Homes to the North and South West, and a Community to the East of the Landfill OEPA granted approval of the project subject to 31 conditions, some of which stipulated that odors deemed to be causing a nuisance must be controlled (Jones, 2001). Given the history of this and other landfill sites, the public was familiar with the types of odors that can be released from landfills, and were concerned about health hazards as well as their quality of life. Realizing that odor was likely to be the cause of disputes, many questions had been posed concerning how bad the odor would be. Residents, the developer, and city officials all wanted to know how the odor would be measured, who would determine what were nuisance levels, and what control measures would be applied if nuisances did occur. There was much concern that many of these questions would be answered using highly subjective methods.

10 The application of an electronic nose was proposed as one way to quantitatively assess the odor problems generated from the project. By providing quantitative assessment it was thought that the quality of debate regarding acceptable and unacceptable odor levels could be improved. FIELD SAMPLING Electronic nose monitoring began at the Garfield Heights Landfill Site on September 16, By the time the Cyranose was available the landfill excavation project was in full operation. Large areas of waste were already exposed on the working face of the project, and odors (and complaints) were being generated. The intent of sampling at and near the landfill site was to quantify various strengths of the landfill odor, and then use this information to determine an acceptable threshold of landfill odor. The nose was trained on two fresh air samples at public parks far from the site, at the site, and on odor from two 5-gallon buckets of landfill material removed from the site. Monitoring was conducted in areas North East (downwind) of the landfill. The proximity of residential and commercial areas to the landfill is illustrated in see Fig. 6. While results were encouraging, several operational complications became apparent. As a result of these complications only a relatively small number of samples could be taken at any one time. Additionally, because monitoring with the electronic nose began quite late in the excavation process, after odor complaints had been voiced, odor suppressants and masking chemicals were being used aggressively at the site. Because of the combination of these factors, it seemed more desirable to refine the technique of using an electronic nose for field work rather than to continue

11 sampling in an effort to determine a nuisance threshold or concentration profile of the odor plume (See Jennings and Cox 2002, for odor plume details). Fig. 6 Scrolling Strip Chart Showing Sensor Response During Landfill Odor Sampling FIELD SAMPLING CONSIDERATIONS Sampling in a contaminated environment As mentioned in the description of electronic nose technology, the sensors in the Cyranose 320 generate a smellprint by comparing the relative resistance change in each sensor. In order to determine a relative change, a sensor must be purged between samples to re-establish a baseline value. When sampling in the filed, this can become a significant problem. In applications with a small point odor source, such as a bottle containing a volatile compound, the nose is able to use ambient air to purge the sensors between samples. However, at the landfill the entire air space

12 contained odors generated from the excavation. When the Cyranose attempted to purge, it pulled in contaminated air which effectively set the baseline as the landfill odor. The initial approach to solving this problem was to carry a tank of compressed air to the field to use to reset the meter during the purge cycle. This method was somewhat successful, but the number of samples that could be taken in was seriously limited by the air volume supply. Additionally, the Cyranose s purge inlet flow rate is important to proper purge operation. An pump inside the Cyranose is designed to use ambient air at atmospheric pressure. Therefore, it is difficult to use compressed air under field conditions. Also, it is necessary to ensure that the tanks used to carry the air do not impart odors that will further alter the baseline measurements. The Cyranose does have a small carbon filter that can be installed along the pathway from the purge inlet to the sensor array to clean the purge airflow. However, the filter is small and not capable of handling the magnitude of global odors that were present at the landfill. An alternative replaceable purge filter was constructed by attaching a 10ml syringe filled with activated carbon to the purge inlet. This made sampling in the filed possible because the carbon could easily be replaced. However it was inconvenient and makes the nose more awkward to handle. Headspace Dilution Effects To achieve reliable results the sensor array must reach equilibrium with the sample gas. By watching the scrolling chart sensor resistance in real time one can observe when the sensors reach equilibrium. At a field site with a global odor source the composition of the air being

13 Fig.7 Ethyl Alcohol Suffering from Headspace Dilution Fig.8 Ethyl Alcohol Sampled From a Stable Headspace

14 sampled can constantly vary with the wind making it very difficult to obtain a stable reading (see Fig. 7 and 8). Fig. 7 illustrates the result of sampling the headspace of approximately 25ml or Ethyl Alcohol placed in an uncovered 250ml beaker. The sample was taken by placing the sample needle about half way into the beaker. The sensor readings fluctuate drastically as a result of small air currents forming around the needle, which are induced by sample pumping. When these currents draw air from above the beaker they lead to dramatic variations in the headspace concentration of the beaker. Fig. 8 illustrates the result of suppressing the effects of headspace dilution. This was accomplished by placing the sample in a capped bottle and allowing the headspace to come to equilibrium before sampling. The sample needle was then inserted through a paraffin cover to sample the contained headspace. The sensors rapidly approached equilibrium and clearly drop back to background levels after the sample was collected. Under field conditions, the effect of varied concentrations resulting from wind-induced concentration transients is similar to that of head space dilution, and yields diminished data quality. Field Sampling Solutions The Cyranose is ideal for use in a clean airspace and sampling from a closed container. Sampling in an open, windy, contaminated environment makes acquiring accurate results significantly more difficult. As suggested, there are solutions to the complications attributed to field sampling. However, rather than attempting to continue using the Cyranose under difficult and variable conditions, further research is being conducted using gas sample bags which can be analyzed in a more controllable environment. In addition to solving the background

15 contamination and headspace dilution challenges, this method greatly simplifies the sampling process. OTHER ENVIRONMENTAL ODOR SOURCES To follow up on the research started at the Garfield Heights landfill, other environmental odor sources were identified for potential further research. Using a field sampling cart (see Fig. 9) containing a Cole Palmer Air Admiral non-contact vacuum pump, 20 Liter tedlar gas sampling bags were filled with odors from municipal and private solid waste transfer stations, yard waste composting facilities, and dumpsters. Like landfill odors, these sources are often the subject of nuisance complaints. Fig. 9 Gas Sampling Cart Used in the Filed to Collect Odor Samples for Laboratory Analysis

16 The majority of the sampling locations had distinct odors that were easily detectable with the human sense of smell. None of these sources, however, caused significant sensor responses in the Cyranose. While the Cyranose is able to very accurately distinguish between some compounds which smell similar to humans (xylene, toluene, and ethyl alcohol or various types of soda) it has had limited success in identifying odors generated by solid wastes. The change in relative sensor resistance from these odors is typically not discernable from the baseline values of the sensor. QUANTIFYING ODOR MAGNITUDE The original goal of the application of electronic technology was to quantify the intensity of an odor. The ability of the Cyranose to function in this manner has been demonstrated in a number of ways. Head space dilutions based on Henry s Law were used to demonstrate the Cyranose s ability to recognize differences in concentration. Liquid phase dilutions of Toluene, to which the nose was known to be sensitive, were made in 1L bottles, and the headspace was allowed to reach equilibrium. As a result of Henry s Law the gas phase concentrations of Toluene were proportional to the liquid phase concentrations. Sampling the bottles with the Cyranose yielded the expected result. A smellprint for each concentration was created, and the raw data from the sensors were streamed into a spreadsheet for further analysis. The changes in sensor response is illustrated in Fig. 10.

17 Fig. 10 Cyranose Sensor Response for Headspace Dilutions of Toluene

18 In addition to using the headspace from liquid phase dilutions it is desirable to have the ability to create dilutions directly from a gas phase sample. A device was created that used a supply of water to replace a volume of air or sample gas in a dilution chamber. The dilution chamber would start out full of water, as a specified volume of water was released an equal volume of sample gas or dilution air would be pulled into the chamber. When the desired dilution was accomplished the dilution chamber could be re-filled with water, and in the process expel the newly diluted sample into a tedlar bag for analysis. This device worked with some success. The dilution device process was time consuming, but the correct volumes could be measured to create dilutions. However, problems with this method became apparent. Many Cyranose sensors are sensitive to the moisture content of samples. In this device the dilutions come into contact with water, which altered their moisture content. This challenge was solved by adding desiccants to the inlet of the tedlar bag the mixture was being expelled into. More importantly, because the gas dilutions were made in contact with water, this device can not be used with water-soluble compounds. Because many of the odors derived from environmental sources are quite water soluble, this device has been replaced with other methods. A method was also developed to make gas/gas dilutions at one atmosphere directly from gas bag samples. To make concentration calibration curves 20 liter tedlar bags were filled with gas vapors. Using an A M Systems 2 Liter calibrated syringe, measured amounts of sample gas or dilution air were extracted and used to fill smaller tedlar bags. Laboratory dilution air was passed through a carbon filter used to eliminate any background odors. The Cyranose was used to sample the bags, while data were streamed into a spreadsheet. The relative resistance change in each sensor was then calculated and plotted versus the percent concentration in the bag (see

19 Fig. 11- Toluene Odor Concentration Calibration Curves Average R^2 = Fig. 12- Ethyl Alcohol Odor Concentration Calibration Curves Fig. 11 and 12). Trend lines fit to these responses can then be used to predict unknown concentrations from the same odor source. Trend lines fit the data extremely well. To show this

20 regression analysis correlation coefficients (R 2 ) values were calculated to determine the accuracy of the linear regressions. An R 2 of 1 represents a prefect correlation; the average R 2 values to linear regression for Toluene and Ethyl Alcohol calibrations were and respectively. Because efforts to find an environmental odor to which the Cyranose is sufficiently sensitive have not been successful laboratory experiments using substances to which the sensors are known to be sensitive (Toluene, Ethyl Alcohol) are underway. DIFFERENTIATING ODORS IN A MIXED SAMPLE Having the ability to accurately predict the unknown concentrations of a sample can provide useful information. However, it would be even more valuable if concentrations from more than one odor source could be differentiated in a sample. For example, This information could be used as a way to control vapor extraction. Using a column packed with sand, and a source of compressed air, soil vapor extraction techniques were used to create gas dilutions from liquid phase samples. Individual samples of Toluene and Ethyl Alcohol as well as a 50% Toluene-50% Ethyl Alcohol were tested. In each case, 10ml of liquid sample was added to a sand column and a steady flow of air was passed through the column, the effluent gas was used to fill a sequence of small tedlar bags. The Cyranose was then used to samples the bags, and the data were streamed into an Excel file. The relative resistance change in each sensor were then calculated and plotted versus the bag number (each bag corresponds to 1 minute of column discharge) to create a set of dilution curves for the sensor array response (Fig. 13 and 14).

21 It is our hope that further research will reveal how the raw sensor data collected from these experiments can be decoded into information that can quantify the mixed composition of these samples. Mixing theories capable of accomplishing this are currently being tested. Fig. 13- Sensor Response to Soil Vapor Extraction of Toluene Fig. 14- Sensor Response to Soil Vapor Extraction of Ethyl Alcohol

22 SUMMARY AND CONCLUSIONS Electronic nose technology offers promise for identifying and quantifying air pollution related to solid waste odors. Field conditions at landfills and other sites can cause sampling difficulties. Bag sampling for laboratory analysis may be required. Sensor response calibration is useful for quantifying odor magnitude, but not all odor sources yield dramatic sensor responses. The best results using the Cyranose 320 were obtained from simple chemical odors. There are air pollution and environmental sources (such as industrial air pollution) where this applies, but this may not be typical of odors from solid wastes. ACKNOWLEDGEMENT This research was supported by a grant from the Environmental Research and Education Foundation and by Case Western Reserve University. REFRENCES Cyrano Sciences Inc., 2000a, Smell What You ve Been Missing, Pasadena, CA. Cyrano Sciences Inc., 2000b, Technical Note 1 A Tour of the New Cyranose 320, Pasadena, CA. Cyrano Sciences Inc., 2000c, Technical Note 2 Cyrano Sciences Sensor Technology: The Heart of the Cyranose 320 Electronic Nose, Pasadena, CA. Cyrano Sciences Inc., 2000d, Technical Note 3 Cyrano Sciences On Board Signal & Processing: The Brains of the Cyranose 320, Pasadena, CA.

23 Cyrano Sciences Inc., 2001, Science Forum: Biological Smell, Gibson, T. Prossser, O. and Hulbert, K., 2000, Electronic Noses: An Inspired Idea, Chemistry & Industry, 8, Jennings, A. and Cox, A., 2002, Electronic Nose Technology Applied to Landfill Odors, Proceedings from the Solid Waste Association of North America s 7 th Annual Landfill Symposium, Jones, C., (Director of Ohio EPA), 2001, R & B Development Landfill, Cuyahoga County OCA Authorization Ohio Environmental Protection Agency, Feb. 2, Leffingwell, John C., 2001, Olfaction A Review, Longo, Thomas, J., (Mayor, The City of Garfield Heights), 2000, personal communication. Longo, Thomas, J., (Mayor, The City of Garfield Heights), 2001, personal communication. McCabe Corporation, 2000, Rule 13 Application Submittal for the R & B Development Landfill, Garfield Heights, Cuyahoga County, Ohio prepared for Peter J. Limited, April 7, 2000.