Multi-Parametric Machine Olfaction

Abstract

A system includes an array of chemical, pressure, and temperature sensors, and a temporal airflow modulator configured to provide sniffed vapors in a temporally-modulated sequence through a plurality of different air paths across multiple sensor locations.

Claims

1. A system for machine olfaction comprising: an array of temperature-controlled sensor pairs; and a computer configured to provide power to the array of sensor pairs, and to wirelessly transmit data from the array of sensor pairs to a host computer.

2. The system of claim 1, wherein each pair of sensors comprises one Volatile Organic Compound (VOC) sensor and one digital barometer.

3. The system of claim 1, wherein said computer is a plurality of computers.

4. A method of machine olfaction comprising the steps of: passing an analyte vapor through a temporal airflow modulator configured to selectively pass an airflow and the analyte vapor; and passing the analyte vapor through a multi-parametric sensing unit.

5. The method of claim 4, wherein said multi-parametric sensing unit is configured to react to a chemical parameter, a pressure parameter and a temperature parameter.

6. The method of claim 4, wherein said multi-parametric sensing unit comprises: an array of temperature-controlled sensor pairs; and a computer configured to provide power to said array of sensor pairs, and to wirelessly transmit data from said array of sensor pairs to a host computer.

7. The method of claim 6, wherein each pair of sensors comprising one Volatile Organic Compound (VOC) sensor and one digital barometer.

8. The method of claim 7, wherein said computer is a plurality of computers.

9. A system comprising: an array of temperature-controlled sensor pairs; and a single-board computer configured to provide power to the array of sensor pairs, and to wirelessly transmit data from the array of sensor pairs to a host computer configured to analyze the data.

10. The system of claim 9, wherein said array of sensor pairs comprises at least eight sensor pairs.

11. The system of claim 9, wherein each pair of sensors comprises one Volatile Organic Compound (VOC) sensor and one digital barometer.

12. The system of claim 11, wherein each VOC sensor comprises a micro-hotplate metal-oxide (MOX) sensor with integrated resistive heaters configured to respond to a presence of volatile molecules.

13. The system of claim 11, wherein each digital barometer comprises a small MEMS sensor with piezoresistive elements on a thin suspended membrane.

14. The system of claim 9, further comprising a DAC for controlling temperature of said temperature-controlled pairs.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0011] These and other features, aspects, and advantages of the present invention will become better understood with reference to the following description, appended claims, and accompanying drawings where:

[0012] FIG. 1 is a block diagram of an exemplary system.

[0013] FIG. 2 is an exemplary graph.

[0014] FIG. 3 is an exemplary test bench.

[0015] FIG. 4 is an exemplary comparison.

[0016] FIGS. 5(a), 5(b) and 5(c) illustrate representative data for odors.

[0017] FIG. 6 is a table.

DETAILED DESCRIPTION

[0018] The subject innovation is now described with reference to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present invention. It may be evident, however, that the present invention may be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form in order to facilitate describing the present invention.

[0019] Most implementations of electronic noses (e-noses) include an array of chemical sensors whose outputs are analyzed in parallel at one discrete point in time. These designs are widely employed across military, industrial, medical, and environmental sciences, with applications ranging from explosives and disease detection to environmental and industrial monitoring. Recent advances in compact, portable, and low-cost sensor designs have been complemented by aggressive microelectronic integration.

[0020] The present invention is an electronic platform (sometimes referred to herein as TruffleBot) which classifies odors using multi-parametric environmental information in order to improve upon traditional e-noses. The TruffleBot simultaneously samples pressure, temperature, and chemical time series, while sniffing in a temporally modulated sequence which introduces spatiotemporal time signatures, such as transport delays and diffusive dynamics. These multidimensional signals depend on chemical and physical properties which can be unique to a particular chemical. Additionally, the odor plumes traverse a set of four unique physical pathways which have the aggregate effect of expanding the feature space and separability of odors. The system, which mirrors some of the dynamic contextual features of animal olfaction, improves the performance and accuracy of chemical sensing in a simple and low-cost hardware platform.

[0021] In FIG. 1, a block diagram of an exemplary system 10 is illustrated. Eight analog metal-oxide gas sensors 12 are digitized while a DAC 14 controls their heater voltage while eight digital barometers 16 measure pressure and temperature. The array of eight sensor pairs 12, 16 are arranged in four rows of two, with each position containing one Volatile Organic Compound (VOC) sensor and one digital barometer. The array of eight sensor pairs 12, 16 are linked to a Raspberry Pi 18 (85 mm?56 mm).

[0022] The VOC sensors 12 (e.g., AMS CCS801) are micro-hotplate metal-oxide (MOX) sensors with integrated resistive heaters. In a MOX gas sensor, a metal oxide film is heated to several hundred degrees Celsius, to a temperature where its electrical conductivity becomes sensitive to chemical interactions with nearby gases. These interactions are complex and non-specific, and MOX sensors will respond to the presence of many different volatile molecules. The heaters of the eight MOX sensors are driven from the common buffered DAC 14, whose voltage controls the temperature of the sensors, and in turn, affects their chemical sensitivity. The MOX resistivity is converted to a voltage and routed through a multiplexer 20 into a high precision ADC 22 (e.g., TI ADS1256). Components 12, 14, 16, 18, 20 and 22 are referred to herein collectively as a TruffleBot.

[0023] In one embodiment, the digital barometers (e.g., ST LPS22HB) are small MEMS sensors with piezoresistive elements on a thin suspended membrane. These chips measure both temperature and absolute pressure at up to 75 samples per second through a serial peripheral interface (SPI) bus.

[0024] The TruffleBot is powered entirely through the 5V and 3.3V rails of a Raspberry Pi 18, and consumes approximately 77 mW. The TruffleBot also hosts several other supporting circuits, including a precision reference generator for the MOX sensors, and transistors to switch external 5V peripherals which may include solenoids and small air pumps. Other peripherals can also be connected through a Universal Serial Bus (USB) (not shown). Components for one TruffleBot cost approximately $150 US.

[0025] The TruffleBot connects to a host computer 24 over Ethernet or WiFi, and multiple TruffleBots can co-exist on the same network. A host program initiates an experiment by broadcasting a command for all TruffleBots to begin data collection. Each TruffleBot saves its sensor traces locally, and when the trial concludes, the host automatically retrieves each client's dataset and compiles them all into a single HDF5 file for analysis in MATLAB? from The Mathworks.

[0026] FIG. 2 is an exemplary graph that plots the temperature, pressure and chemical response to a five second exposure to odors from beer (?6% ethanol). The output of the VOC sensor is expressed as a percentage of its full scale range, and the pressure and temperature signals deviate only slightly from ambient. When beer odors are introduced, the pressure decreases and the temperature increases; both the polarity and magnitude of these changes depend on the physical properties of the analyte vapor including its vapor pressure, density, and molecular weight. These differences contribute to TruffleBot's overall chemical selectivity.

[0027] In FIG. 3, an exemplary test bench 100 is illustrated in (a). The output of a pump 102 is regulated to a constant flow by a flow controller 104, and a three-way solenoid valve 106 (e.g., Takasago CTV-3) selectively bypasses the analyte vapor 108. The solenoid 106 is controlled with a short pseudorandom (PN) binary sequence. The analytes used in these experiments were ambient air (control), apple cider vinegar, lime juice, beer (6.2% ABV), wine (chardonnay, 13% ABV), vodka (40% ABV), ethanol (100% ABV), isopropanol, and acetone. A manifold 109 splits the fluid flow between four small plastic columns 110, 112, 114, 116 containing different obstructions before reaching the sensor array 118. This arrangement allows one to adjust multiple parameters including the overall airflow, the solenoid's temporal sequence, the analyte, and the geometries and contents of the columns.

[0028] In (b), an exemplary graph 120 shows differences in the positions and obstructions of four air paths produce different signals in each column, in response to ethanol.

[0029] FIG. 4 shows an exemplary comparison of the responses to air, vodka and acetone, at a single array position. The baseline signal levels are affected by noise and uncontrolled parameters including ambient temperature, humidity, and atmospheric pressure. The solenoid PN sequence is the same for all trials, and the signal is represented by temporally correlated changes in the sensor outputs. Assuming a lossless system with fixed volumetric flow, the total absolute pressure in each column head can be represented as

[00001]$\begin{array}{cc}{P}_{\mathrm{abs}}=P_{\mathrm{atm}}+P_{\mathrm{ext}}+P_{\mathrm{analyte}}& \left(1\right)\end{array}$ [0030] where P.sub.ext is the resulting pressure from the constant regulated airflow and P.sub.atm and P.sub.analyte are the partial pressures exerted by atmospheric air and the analyte vapor. (When the solenoid bypasses the analyte, P.sub.analyte=0.) According to the Darcy-Weisbach equation, Newtonian fluid flowing through a cylindrical tube experiences a pressure drop given by

[00002]$\begin{array}{cc}?p=?{\mathrm{fLv}}^2/2D& \left(2\right)\end{array}$ [0031] where ? is the fluid density, v is the fluid velocity and f, L, D are the friction coefficient, length, and diameter of the tube. Since the flow is constant and tube properties do not change, ?p only depends on ?. Thus an analyte with vapor density greater than air would incur more pressure loss in the tube, resulting in a decrease in measured air pressure. For example, P.sub.abs decreases during the release of beer (?=1.05 g/cm.sup.3) but increases for vodka (?=0.95 g/cm.sup.3).

[0032] These pressure changes, in combination with the analyte's physical properties (e.g. heat capacity), produce analyte-specific temperature fluctuations. Using this information, TruffleBot can distinguish between analytes which have similar MOX sensor responsivity, provided the pressure and temperature changes observed are a systematic result of the analyte's physical properties. For example, in FIG. 4, vodka and acetone could have been easily discriminated by temperature and pressure alone.

[0033] The arrayed sensors and diverse airflow paths support the extraction of temporal and spatial features. Using the setup in FIG. 3, the same sniffing sequence of 40 pseudorandom bits was applied at 1 bit/second for 8 different analytes. Representative data for each odor is shown in FIG. 5(a). The first eight rows represent VOC sensor traces, followed by eight rows of pressure readings and eight rows of temperature readings. The mean value has been subtracted from each trace. The control trials with ambient air show only small deviations, while VOC magnitudes appear to correlate with alcohol content, as one might anticipate. Some odors do not have significant VOC sensor response (lime, vinegar), but do show appreciable pressure and temperature responses.

[0034] The experiment was repeated ten times for each analyte, and feature vectors containing the mean, derivative, and standard deviation were assembled from 0.5 second windows of each of the 24 time series. We performed principal-component analysis (PCA) on the combined sensor data of the nine odor classes (FIG. 5(b)). Even with a comparison of only the first two principal components, tightly grouped clusters emerged. We then performed 2-fold cross-validation of the results using a simple k-means algorithm over 1000 iterations. The classification approach is comparable to other e-nose demonstrations, and is one of many possible classification strategies (See Table I in FIG. 6).

[0035] A cross validation accuracy of 90.9% was achieved using only the transient time series from the MOX sensors, compared to 79.8% if the data is condensed to only 1 average value per MOX sensor. Adding temperature and pressure data, error rates reduced by a factor of 2 and accuracy improved to 95.8%. The confusion matrix in FIG. 5(c) shows that most of the errors occurred between lime and vinegar. This can also be seen in their overlapping PCA clusters (FIG. 5(b)), and it is intuitive since citric acid and acetic acid contain chemically similar functional groups. Repeating the classification for the 8 datasets excluding lime gives a cross validation accuracy of 98.5%.

[0036] It would be appreciated by those skilled in the art that various changes and modifications can be made to the illustrated embodiments without departing from the spirit of the present invention. All such modifications and changes are intended to be within the scope of the present invention except as limited by the scope of the appended claims.