Data-Centric Performance Improvement Strategies for Few-Shot Classification of Chemical Sensor Data

Metal oxide (MOX) sensors offer a low-cost solution to detect volatile organic compound (VOC) mixtures. However, their operation involves time-consuming heating cycles, leading to a slower data collection and data classification process. This work introduces a few-shot learning approach that promotes rapid classification. In this approach, a model trained on several base classes is fine-tuned to recognize a novel class using a small number (n = 5, 25, 50 and 75) of randomly selected novel class measurements/shots. The used dataset comprises MOX sensor measurements of four different juices (apple, orange, currant and multivitamin) and air, collected over 10-minute phases using a pulse heater signal. While high average accuracy of 82.46 is obtained for five-class classification using 75 shots, the model’s performance depends on the juice type. One-shot validation showed that not all measurements within a phase are representative, necessitating careful shot selection to achieve high classification accuracy. Error analysis revealed contamination of some measurements by the previously measured juice, a characteristic of MOX sensor data that is often overlooked and equivalent to mislabeling. Three strategies are adopted to overcome this: (E1) and (E2) fine-tuning after dropping initial/final measurements and the first half of each phase, respectively, (E3) pretraining with data from the second half of each phase. Results show that each of the strategies performs best for a specific number of shots. E3 results in the highest performance for five-shot learning (accuracy 63.69), whereas E2 yields the best results for 25-/50-shot learning (accuracies 79/87.1) and E1 predicts best for 75-shot learning (accuracy 88.6). Error analysis also showed that, for all strategies, more than 50% of air misclassifications resulted from contamination, but E1 was affected the least. This work demonstrates how strongly data quality can affect prediction performance, especially for few-shot classification methods, and that a data-centric approach can improve the results.

Applications and Limitations of Quantifying Speciated and Source-Apportioned VOCs with Metal Oxide Sensors

While low-cost air quality sensor quantification has improved tremendously in recent years, speciated hydrocarbons have received little attention beyond total lumped volatile organic compounds (VOCs) or total non-methane hydrocarbons (TNMHCs). In this work, we attempt to use two broad response metal oxide VOC sensors to quantify a host of speciated hydrocarbons as well as smaller groups of hydrocarbons thought to be emanating from the same source or sources. For sensors deployed near oil and gas facilities, we utilize artificial neural networks (ANNs) to calibrate our low-cost sensor signals to regulatory-grade measurements of benzene, toluene, and formaldehyde. We also use positive matrix factorization (PMF) to group these hydrocarbons along with others by source, such as wet and dry components of oil and gas operations. The two locations studied here had different sets of reference hydrocarbon species measurements available, helping us determine which specific hydrocarbons and VOC mixtures are best suited for this approach. Calibration fits on the upper end reach above R2 values of 0.6 despite the parts per billion (ppb) concentration ranges of each, which are magnitudes below the manufacturer’s prescribed detection limits for the sensors. The sensors generally captured the baseline trends in the data, but failed to quantitatively estimate larger spikes that occurred intermittently. While compounds with high variability were not suited for this method, its success with several of the compounds studied represents a crucial first step in low-cost VOC speciation. This work has important implications in improving our understanding of the links between health and environment, as different hydrocarbons will have varied consequences in the human body and atmosphere.

Biological Waste Air and Waste Gas Treatment: Overview, Challenges, Operational Efficiency, and Current Trends

International contracts to restrict emissions of climate-relevant gases, and thus global warming, also require a critical reconsideration of technologies for treating municipal, commercial, industrial, and agricultural waste gas emissions. A change from energy- and resource-intensive technologies, such as thermal post-combustion and adsorption, as well to low-emission technologies with high energy and resource efficiency, becomes mandatory. Biological processes already meet these requirements, but show restrictions in case of treatment of complex volatile organic compound (VOC) mixtures and space demand. Innovative approaches combining advanced oxidation and biofiltration processes seem to be a solution. In this review, biological processes, both as stand-alone technology and in combination with advanced oxidation processes, were critically evaluated in regard to technical, economical, and climate policy aspects, as well as present limitations and corresponding solutions to overcome these restrictions.

Volatile organic compound profiles in outlet air from extracorporeal life-support devices differ from breath profiles in critically ill patients

IntroductionIt is highly uncertain whether volatile organic compounds (VOCs) in exhaled breath of critically ill intensive care unit patients are formed in the lung locally, in the air compartment or lung tissue, or elsewhere in the body and transported to the lung via the bloodstream. We compared VOC mixtures in exhaled breath and in air coming from extracorporeal support devices in critically ill patients to address this issue.MethodsFirst, we investigated whether it was safe to connect an electronic nose (eNose) or a gas sampling pump to extracorporeal support membranes. Then, breath and air from extracorporeal support devices were collected simultaneously for continuous monitoring of VOC mixtures using an eNose. In addition, samples for gas chromatography/mass spectrometry (GC-MS) analysis were taken daily at the two measurement sites.Results10 critically ill patients were monitored for a median (interquartile range) duration of 73 (72–113) h; in total, we had 887 h of air sampling. The eNose signals of breath correlated moderately with signals of air from the extracorporeal support devices (R2=0.25–0.44). After GC-MS analysis, 96 VOCs were found both in breath and air from the extracorporeal support devices; of these, 29 (30%) showed a significant correlation (p<0.05) between the two measurement sites, of which 17 were identified. VOCs that did not correlate were found in a higher concentration in breath than in air from the extracorporeal support devices.ConclusionThis study suggests VOC analysis in the extracorporeal circulation is safe, and that VOCs of nonpulmonary origin can be measured in the breath and in the extracorporeal circulation of critically ill patients. For VOCs that did not correlate between the two measurement sites, the breath concentration was higher, suggesting pulmonary production of these molecules in a highly selected population of patients that received extracorporeal support.