Publications

Conventionally, odours emitted by different sources present in wastewater treatment plants (WWTPs) are measured by dynamic olfactometry, where a human panel sniffs and analyzes air bags collected from the plant. Although the method is considered the gold standard, the process is costly, slow, and infrequent, which does not allow operators to quickly identify and respond to problems. To better monitor and map WWTP odour emissions, here we propose a small rotary-wing drone equipped with a lightweight (1.3-kg) electronic nose. The "sniffing drone" sucks in air via a ten-meter (33-foot) tube and delivers it to a sensor chamber where it is analyzed in real-time by an array of 21 gas sensors. From the sensor signals, machine learning (ML) algorithms predict the odour concentration that a human panel using the EN13725 methodology would report. To calibrate and validate the predictive models, the drone also carries a remotely controlled sampling device (compliant with EN13725:2022) to collect sample air in bags for post-flight dynamic olfactometry. The feasibility of the proposed system is assessed in a WWTP in Spain through several measurement campaigns covering diverse operating regimes of the plant and meteorological conditions. We demonstrate that training the ML algorithms with dynamic (transient) sensor signals measured in flight conditions leads to better performance than the traditional approach of using steady-state signals measured in the lab via controlled exposures to odour bags. The comparison of the electronic nose predictions with dynamic olfactometry measurements indicates a negligible bias between the two measurement techniques and 95 % limits of agreement within a factor of four. This apparently large disagreement, partly caused by the high uncertainty of olfactometric measurements (typically a factor of two), is more than offset by the immediacy of the predictions and the practical advantages of using a drone-based system.Copyright © 2022. Published by Elsevier B.V.

JTD Keywords: calibration, chemical sensors, drone, dynamic olfactometry, electronic nose, odourquantification, olfaction, volatile organic-compounds, wwtp, Calibration, Chemical sensors, Drone, Dynamic olfactometry, Electronic nose, Environmental monitoring, Odour quantification, Olfaction, Variable selection methods, Wwtp

The human olfactory system remains one of the most challenging biological systems to replicate. Humans use it without thinking, where it can equally offer protection from harm and bring enjoyment in equal measure. It is the system’s ability to detect and analyze complex odors, without the need for specialized infra-structure, that is the envy of many scientists. The field of artificial olfaction has recruited and stimulated interdisciplinary research and commercial development for several applications that include malodor measurement, medical diagnostics, food and beverage quality, environment and security. Over the last century, innovative engineers and scientists have been focused on solving a range of problems associated with measurement and control of odor. The IEEE Sensors Journal has published Special Issues on olfaction in 2002 and 2012. Here we continue that coverage. In this article, we summarize early work in the 20th Century that served as the foundation upon which we have been building our odor-monitoring instrumental and measurement systems. We then examine the current state of the art that has been achieved over the last two decades as we have transitioned into the 21st Century. Much has been accomplished, but great progress is needed in sensor technology, system design, product manufacture and performance standards. In the final section, we predict levels of performance and ubiquitous applications that will be realized during in the mid to late 21st Century.

JTD Keywords: air-quality, breath analysis, calibration transfer, chemical sensor arrays, chemosensor arrays, drift compensation, electronic nose, gas sensors, headspace sampling, machine learning, machine olfaction, odor detection, plume structure, voc analysis, Artificial olfaction, Electrodes, Electronic nose, Electronic nose technology, Headspace sampling, Instruments, Machine learning, Machine olfaction, Monitoring, Odor detection, Olfactory, Sensor phenomena and characterization, Sensors, Temperature sensors, Voc analysis

The difficulty to obtain ground truth (i.e. empirical evidence) about how a gas disperses in an environment is one of the major hurdles in the field of mobile robotic olfaction (MRO), impairing our ability to develop efficient gas source localization strategies and to validate gas distribution maps produced by autonomous mobile robots. Previous ground truth measurements of gas dispersion have been mostly based on expensive tracer optical methods or 2D chemical sensor grids deployed only at ground level. With the ever-increasing trend towards gas-sensitive aerial robots, 3D measurements of gas dispersion become necessary to characterize the environment these platforms can explore. This paper presents ten different experiments performed with a 3D grid of 27 metal oxide semiconductor (MOX) sensors to visualize the temporal evolution of gas distribution produced by an evaporating ethanol source placed at different locations in an office room, including variations in height, release rate and air flow. We also studied which features of the MOX sensor signals are optimal for predicting the source location, considering different lengths of the measurement window. We found strongly time-varying and counter-intuitive gas distribution patterns that disprove some assumptions commonly held in the MRO field, such as that heavy gases disperse along ground level. Correspondingly, ground-level gas distributions were rarely useful for localizing the gas source and elevated measurements were much more informative. We make the dataset and the code publicly available to enable the community to develop, validate, and compare new approaches related to gas sensing in complex environments.

JTD Keywords: Mobile robotic olfaction, Metal oxide gas sensors, Signal processing, Sensor networks, Gas source localization, Gas distribution mapping

Metal Oxide (MOX) gas sensors rely on chemical reactions that occur efficiently at high temperatures, resulting in too-demanding power requirements for certain applications. Operating the sensor under a Pulsed-Temperature Operation (PTO), by which the sensor heater is switched ON and OFF periodically, is a common practice to reduce the power consumption. However, the sensor performance is degraded as the OFF periods become larger. Other research works studied, generally, PTO schemes applying waveforms to the heater with time periods of seconds and duty cycles above 20%. Here, instead, we explore the behaviour of PTO sensors working under aggressive schemes, reaching power savings of 99% and beyond with respect to continuous heater stimulation. Using sensor sensitivity and the limit of detection, we evaluated four Ultra Low Power (ULP) sensors under different PTO schemes exposed to ammonia, ethylene, and acetaldehyde. Results show that it is possible to operate the sensors with total power consumption in the range of microwatts. Despite the aggressive power reduction, sensor sensitivity suffers only a moderate decline and the limit of detection may degrade up to a factor five. This is, however, gas-dependent and should be explored on a case-by-case basis since, for example, the same degradation has not been observed for ammonia. Finally, the run-in time, i.e., the time required to get a stable response immediately after switching on the sensor, increases when reducing the power consumption, from 10 minutes to values in the range of 10–20 hours for power consumptions smaller than 200 microwatts.

JTD Keywords: Robot sensing systems, Temperature sensors, Heating systems, Gas detectors, Power demand, Sensitivity, Electronic nose, gas sensors, low-power operation, machine olfaction, pulsed-temperature operation, temperature modulation

Inherent sensor variability limits mass-production applications for metal oxide (MOX) gas sensor arrays because calibration for replicas of a sensor array needs to be performed individually. Recently, calibration transfer strategies have been proposed to alleviate calibration costs of new replicas, but they still require the acquisition of transfer samples. In this work, we present calibration models that can be extended to uncalibrated replicas of sensor arrays without acquiring new samples, i.e., general or global calibration models. The developed methodology consists in including multiple replicas of a sensor array in the calibration process such that sensor variability is rejected by the general model. Our approach was tested using replicas of a MOX sensor array in the classification task of six gases and synthetic air, presented at different background humidity and concentration levels. Results showed that direct transfer of individual calibration models provides poor classification accuracy. However, we also found that general calibration models kept predictive performance when were applied directly to new copies of the sensor array. Moreover, we explored, through feature selection, whether particular combinations of sensors and operating temperatures can provide predictive performances equivalent to the calibration model with the complete array, favoring thereby the existence of more robust calibration models.

JTD Keywords: Gas sensor array, MOX sensor, Robust calibration, Calibration transfer, Machine olfaction

A methodology to calculate analytical figures of merit is not well established for detection systems that are based on sensor arrays with low sensor selectivity. In this work, we present a practical approach to estimate the Resolving Power of a sensory system, considering non-linear sensors and heteroscedastic sensor noise. We use the definition introduced by Shannon in the field of communication theory to quantify the number of symbols in a noisy environment, and its version adapted by Gardner and Barlett for chemical sensor systems. Our method combines dimensionality reduction and the use of algorithms to compute the convex hull of the empirical data to estimate the data volume in the sensor response space. We validate our methodology with synthetic data and with actual data captured with temperature-modulated MOX gas sensors. Unlike other methodologies, our method does not require the intrinsic dimensionality of the sensor response to be smaller than the dimensionality of the input space. Moreover, our method circumvents the problem to obtain the sensitivity matrix, which usually is not known. Hence, our method is able to successfully compute the Resolving Power of actual chemical sensor arrays. We provide a relevant figure of merit, and a methodology to calculate it, that was missing in the literature to benchmark broad-response gas sensor arrays.

JTD Keywords: Gas sensor array, MOX sensors, Resolving Power, Sensor resolution, Dimensionality reduction, Machine olfaction

We present the estimation of a likelihood map for the location of the source of a chemical plume dispersed under atmospheric turbulence under uniform wind conditions. The main contribution of this work is to extend previous proposals based on Bayesian inference with binary detections to the use of concentration information while at the same time being robust against the presence of background chemical noise. For that, the algorithm builds a background model with robust statistics measurements to assess the posterior probability that a given chemical concentration reading comes from the background or from a source emitting at a distance with a specific release rate. In addition, our algorithm allows multiple mobile gas sensors to be used. Ten realistic simulations and ten real data experiments are used for evaluation purposes. For the simulations, we have supposed that sensors are mounted on cars which do not have among its main tasks navigating toward the source. To collect the real dataset, a special arena with induced wind is built, and an autonomous vehicle equipped with several sensors, including a photo ionization detector (PID) for sensing chemical concentration, is used. Simulation results show that our algorithm, provides a better estimation of the source location even for a low background level that benefits the performance of binary version. The improvement is clear for the synthetic data while for real data the estimation is only slightly better, probably because our exploration arena is not able to provide uniform wind conditions. Finally, an estimation of the computational cost of the algorithmic proposal is presented.

JTD Keywords: Machine olfaction, Odor robots, Chemical sensors, Bayesian inference

Conventional fire alarms are based on smoke detection. Nevertheless, in some fire scenarios volatiles are released before smoke. Fire detectors based only on chemical sensors have already been proposed as they may provide faster response, but they are still prone to false alarms in the presence of nuisances. These systems rely heavily on pattern recognition techniques to discriminate fires from nuisances. In this context, it is important to test the systems according to international standards for fires and testing the system against a diversity of nuisances. In this work, we investigate the behavior of a gas sensor array coupled to sensor fusion algorithms for fire detection when exposed to standardized fires and several nuisances. Results confirmed the ability to detect fires (97% Sensitivity), although the system still produces a significant rate of false alarms (35%) for nuisances not presented in the training set.

JTD Keywords: Fire alarm, Gas sensor array, Machine Olfaction, Multisensor system, Sensor fusion

Abstract Inherent variability of chemical sensors makes necessary individual calibration of chemical detection systems. This shortcoming has traditionally limited usability of systems based on Metal Oxide (MOX) sensor arrays and prevented mass-production for some applications. Here, aiming at exploring transfer calibration between electronic nose systems, we exposed five identical 8-sensor detection units to controlled gas conditions. Our results show that a calibration model provides more accurate predictions when the tested board is included in the calibration dataset. However, we show that previously built calibration models can be extended to other units using a reduced number of measurements. While baseline correction seems imperative for successful baseline correction, among the different tested strategies, piecewise direct standardization provides more accurate predictions.

JTD Keywords: Electronic nose, Calibration, MOX sensor, Machine Olfaction

Biological olfaction outperforms chemical instrumentation in specificity, response time, detection limit, coding capacity, time stability, robustness, size, power consumption, and portability. This biological function provides outstanding performance due, to a large extent, to the unique architecture of the olfactory pathway, which combines a high degree of redundancy, an efficient combinatorial coding along with unmatched chemical information processing mechanisms. The last decade has witnessed important advances in the understanding of the computational primitives underlying the functioning of the olfactory system. EU Funded Project NEUROCHEM (Bio-ICT-FET- 216916) has developed novel computing paradigms and biologically motivated artefacts for chemical sensing taking inspiration from the biological olfactory pathway. To demonstrate this approach, a biomimetic demonstrator has been built featuring a large scale sensor array (65K elements) in conducting polymer technology mimicking the olfactory receptor neuron layer, and abstracted biomimetic algorithms have been implemented in an embedded system that interfaces the chemical sensors. The embedded system integrates computational models of the main anatomic building blocks in the olfactory pathway: The olfactory bulb, and olfactory cortex in vertebrates (alternatively, antennal lobe and mushroom bodies in the insect). For implementation in the embedded processor an abstraction phase has been carried out in which their processing capabilities are captured by algorithmic solutions. Finally, the algorithmic models are tested with an odour robot with navigation capabilities in mixed chemical plumes.

JTD Keywords: Antennal lobes, Artificial olfaction, Computational neuroscience, Olfactory bulbs, Plume tracking, Abstracting, Actuators, Algorithms, Biomimetic processes, Chemical sensors, Conducting polymers, Data processing, Flavors, Odors, Robots, Smart sensors, Embedded systems

Neural coding of chemical information is still under strong debate. It is clear that, in vertebrates, neural representation in the olfactory bulb is a key for understanding a putative odour code. To explore this code, in this work we have studied a public dataset of radio images of 2-Deoxyglucose uptake (2-DG) in the olfactory bulb of rats in response to diverse odorants using univariate pixel selection algorithms: rank-products and Mann-Whitney U (MWU) test. Initial results indicate that some chemical properties of odorants preferentially activate certain areas of the rat olfactory bulb. While non-parametric test (MWU) has difficulties to detect these regions, rank-product provides a higher power of detection.

JTD Keywords: 2-Deoxyglucose uptake, Chemotopy, Feature selection, Odour coding, Olfaction, Olfactory bulb

We propose the use of multi-way methods to analyze the contribution of diversity and redundancy to odor identification and concentration estimation in a large chemical sensor array. We use a chemical sensing system based on a large array of metal oxide sensors (MOX) and inspired on the diversity and redundancy of the olfactory epithelium. In order to analyze the role of diversity (different sensor type and temperature modulation) and redundancy (replicates of sensors and different load resistors) in odor quantification and discrimination tasks, we have acquired two datasets and modeled the data using multi-way techniques.

JTD Keywords: Artificial Olfaction, Large array, MOX gas sensor, Multi-way methods

In sensory processing of odors, the olfactory bulb is an important relay station, where odor representations are noise-filtered, sharpened, and possibly re-organized. An organization by perceptual qualities has been found previously in the piriform cortex, however several recent studies indicate that the olfactory bulb code reflects behaviorally relevant dimensions spatially as well as at the population level. We apply a statistical analysis on 2-deoxyglucose images, taken over the entire bulb of glomerular layer of the rat, in order to see how the recognition of odors in the nose is translated into a map of odor quality in the brain. We first confirm previous studies that the first principal component could be related to pleasantness, however the next higher principal components are not directly clear. We then find mostly continuous spatial representations for perceptual categories. We compare the space spanned by spatial and population codes to human reports of perceptual similarity between odors and our results suggest that perceptual categories could be already embedded in glomerular activations and that spatial representations give a better match than population codes. This suggests that human and rat perceptual dimensions of odorant coding are related and indicates that perceptual qualities could be represented as continuous spatial codes of the olfactory bulb glomerulus population.

JTD Keywords: Glomeruli, Memory organization, Odor quality, Olfaction, Olfactory bulb, Perceptual categories, Population coding, Spatial coding