Tutorial A: Multivariable sensors for selective and stable gas monitoring
- Radislav A. Potyrailo (GE Global Research)
Abstract: Modern monitoring requirements of gases for demanding applications such as industrial safety, environmental surveillance, medical diagnostics, biopharmaceutical process control, and homeland security push the limits of existing detection concepts where we may reach their fundamental performance limits.
This tutorial is focused on multivariable sensing concepts and implementations that bridge the gap between the existing and required sensing capabilities. This tutorial will stimulate scientific and engineering senses of the participants by (1) posing several fundamental and practical questions on possibilities for new principles of gas sensing and (2) by demonstrating on how these questions are addressed in the developments of sensors with previously unthinkable capabilities in wearable, disposable, wireless, and other formats.
Multivariable gas sensors utilize multi-dimensional response principles to overcome insufficient selectivity and stability limitations of existing sensors. The design rules of these multivariable sensors involve a sensing material with multi-response mechanisms to different chemicals and a multivariable transducer with independent outputs to recognize these different responses. The independent outputs of the sensor are processed using multivariate data analysis tools. We will illustrate the capabilities of these sensors to quantify individual chemical components in mixtures, reject interferences, and enhance response stability. Such performance is attractive when selectivity advantages of traditional mature instruments are cancelled by application-specific requirements.
Tutorial B: Quantification of odors and the role of electronic noses
- Jinichi Kita (Shimadzu Corporation)
Abstract: It is difficult to express odor quantitatively because the primary elements of odors have not been discovered yet unlike the primary colors in vision. In sensory panel tests, quantitative expression has been attempted with some procedures, in which the odor intensity is decided by determining the dilution factor making the odor not-detectible, or odor types are expressed using reference odors. In component analysis like GC/MS, the odor is expressed as the combination of the odor components which are effective to give the total odor perception using GC-Olfactmetory, with considerable effort. Unfortunately, there is a limit in the reproducibility for the sensory evaluation. There are also limits in the component analysis. First, portability cannot be expected for well sophisticated GC/MS instruments, and second, it is not easy to pick up all the effective components in the case of multicomponent odor giving the same odor as the sample odor.
Because, in case of multicomponent odor, the lower threshold odor components (which are odorless when they exist individually), are needed and components with completely different odors from the total odor, are needed to reproduce the original odor.
In the situation of odor quantification, the electronic nose has potential portability and is expected to express the odor the same as that of real nose sense. However, several questions are still unanswered: which components of odor are detected by the electronic nose’s sensors, and whether the outputs of the electronic nose are well expressed with multicomponent odor features. Occasionally, serious problems with the electronic nose arise described below. High concentration of disturbing molecules (termed here as large interferences), consisting of little or no odor, and non-target odors sometimes hide the low concentration of indispensable molecules (termed here as small contributors). Some kinds of electronic noses do not respond to the small contributors because of low sensitivity.
Some trials to solve these problems are introduced.
Practical examples will also be shown in this tutorial for explanation.
Tutorial C: Principle of Gas phase biosensors: Biosniffer & Sniff-cam for medical and healthcare applications
- Kohji Mitsubayashi and Takahiro Arakawa (Tokyo Medical and Dental University)
Abstract: In the fields of medical and healthcare, novel gas sensors are required not only for medical inspection of some diseases at hospital, but also for early diagnosis in daily life. The sensing devices need a great performance (i.e. sensitivity, target selectivity, insensibility to humidity, prompt responsiveness, etc.) for real-time sensing & imaging of breath and skin volatiles related some disease and health condition. In 1994, we reported a gas-phase biosensor for real-time sensing of ethanol using a gas/liquid flow cell with a porous hydrophobic diagram on Analytical Chemistry, ACS. So far, many types and kinds of the gas phase biosensors (Bio-sniffer and Sniff-cam [imaging system]) have been developed using general enzymes and/or drug metabolizing enzymes for human volatiles (EtOH, MeOH, acetaldehyde, acetone, isopropanol, trimethylamine, methyl mercaptan, formaldehyde, nicotine, 2-nonenal, choline, etc.) related some diseases and health conditions. They shows good performance (sensitivity [ppb], selectivity [enzyme specificity], humidity insensibility, wide dynamic range, etc). In this tutorial lecture, we will introduce the principle of the gas phase biosensors and some applications of the real-time sensing and imaging of human volatiles (breath, transcutaneous gases) n the fields of medical & dental and healthcare. The session would be useful for young researchers to develop the novel gas sensors.
Tutorial D: Challenges of breath sampling in medical applications
- Jan Mitrovics (JLM Innovation GmbH)
Abstract: Analyzing exhaled breath is a very attractive way to detect diseases, monitor fitness and wellbeing or follow the progress of the treatment of patients. Breath analysis is very low risk, painless for the patient and quick to perform. Advances on technology and analytical instrumentation have led to large scientific interest and many studies being performed.
Sampling is a crucial step in any measurement to ensure reliable, reproducible results. Breath sampling poses special challenges due to high humidity, large variability and measurement conditions in the practical application.
In this tutorial we will give an overview of the different factors that may influence the results of breath analysis. We will introduce several concepts for breath sampling and discuss their advantages and disadvantages. For offline sampling methods using sampling bags and desorption tubes will be shown. For real time measurements we present two different methods, buffered and non-buffered sampling.
Examples of breath analysis systems will be presented. Typical analytical instrumentats used are GC-MS and PTR-MS systems. While these systems are very powerful to measure individual VOCs and analysis the composition of breath, they are also very expensive. A broad application of breath analysis will require cheaper and possibly mobile systems. Hence there is large interest in using gas sensor arrays gas sensor arrays for breath analysis to reduce cost and size. We will give examples of breath analysis systems using sensor arrays with metal oxide and gold nanoparticle sensors and discuss practical aspects of breath sampling in the context of clinical trials.
Index Terms— breath analysis, sampling, gas sensors
Tutorial E: Introduction to deep learning applications in robotic olfaction
- Emily Stark (Florida Atlantic University)
Abstract: The application of deep-learning techniques has led to rapid and significant progress in many research and development areas including computer vision, natural language processing and autonomous-systems control. To date, deep learning has primarily been applied to large, multi-sample data sets. Robotic olfaction has previously not needed such techniques, as the dimensionality and sample size is often small and traditional machine learning approaches have been sufficient. However, the latest generation of such instruments now produce very large, data rich results from every sample, making the adoption of deep learning approaches a clear next step in the field.
This tutorial will introduce participants to general deep learning techniques as well as specific tools for robotic olfaction and chemical sensing. It will include an overview of deep learning, including a brief history, basic concepts, and hands-on coding in Python (prior installation is not necessary, as a free online platform will be used with Google Drive) showing some popular network architectures (e.g. Perceptrons, Convolutional Neural Networks, and Recurrent Neural Networks). In addition,we will introduce a set of pre- and post-processing methods that have yielded success in analyzing data sets with a small sample size (N < 100) and high dimensionality, typical of chemical sensing. As an example of these methods, a GC-IMS dataset of three types of juice will be preprocessed, used to train a model, and post-processed with a pre-trained model.
Tutorial F: Development & validation of machine learning predictive models
- Santiago Marco (University of Barcelona)