A failed attempt to demystify the most wintry of miracles
Winter has already gently tickled the noses of the editorial staff here at the Institute of Winter Wonders, prompting a question: can we actually "sniff" frosty freshness with instruments and determine exactly what this scent is made of? To find out for sure, we spoke with Fedor Fedorov, head of the "electronic nose" development team.
Fedor
Fedorov //
Fedorov //
Assistant Professor, Laboratory of Nanomaterials, Skoltech Center for Photonic Science and Engineering.
Lead developer of the electronic nose.
Lead developer of the electronic nose.
Seventeen
Before meeting the scientist on Zoom, science journalist Nikita Lavrenov hastily finishes his aromatic filter coffee, unaware of just how many (and which) substances define its scent. In a few minutes, his screen will display photos of an experimental setup designed to sniff the most diverse aromas and substances.
Infinity
Bordering on Magic
Our electronic nose is a regular participant in various technology exhibitions. Specifically for these events, Fedor Fedorov’s lab 3D-printed a plastic nose (bearing a striking resemblance to Michelangelo's David!). It looks exactly like a real electronic... nose. After one such exhibition, this model was gifted to a colleague in the Communications Department. While in our department, the nose found only aesthetic application, the possibilities for its actual use are infinite.
If this nose, like the one in Gogol's story, could recount its own olfactory adventures, the world would have one more anthology of amazing tales. We’ve already discussed its moonlighting gig as a coffee taster. It has also served as a petroleum expert, determining the origin of crude oil samples by their light volatile fractions. Once, it monitored meat freshness, guessing the moment of spoilage before a microbiologist even knew. And it has entertained crowds by distinguishing primary plastic from recycled material and effortlessly locating “off-notes”—a task that would take a human tasting panel a bit of a time. One single device—dozens of professions: from sommelier to environmental inspector.
If this nose, like the one in Gogol's story, could recount its own olfactory adventures, the world would have one more anthology of amazing tales. We’ve already discussed its moonlighting gig as a coffee taster. It has also served as a petroleum expert, determining the origin of crude oil samples by their light volatile fractions. Once, it monitored meat freshness, guessing the moment of spoilage before a microbiologist even knew. And it has entertained crowds by distinguishing primary plastic from recycled material and effortlessly locating “off-notes”—a task that would take a human tasting panel a bit of a time. One single device—dozens of professions: from sommelier to environmental inspector.
But even if frost is the zero point of scent for us, for the electronic nose, silence is never absolute. The device is capable of registering even those gases that the human nose doesn't notice at all. And sometimes—even those that don't technically have a scent, even for the device.
The electronic nose brings us closer to a future where robots will be able to sense even the absence of a smell. But only a human can call that freshness.
Eight
17 organic compounds, primarily aldehydes. Now the science journalist knows what his morning smells like—as well as the morning of the billions who wake up with coffee. To us, it is a single "coffee scent," within which only the most sophisticated can distinguish individual notes.
In reality, the electronic nose—which is comparable in size to our own—perceives the scent as a whole, just like we do. To break a scent down into its components, scientists use cumbersome gas analyzers. But the e-nose is compact. And it is not a single shiny metal probe, as we tend to imagine lab instruments, but a small team of sensors.
Instead of olfactory receptors, it uses the thinnest films of metal oxides. There are usually eight of them on a chip: oxides of manganese, zinc, cobalt, tin, and other elements from the middle of the periodic table that react to volatile compounds appearance. With every breath—or rather, every intake of the gas mixture—these oxide films change their electrical resistance (conductivity). In some films, it drops; in others, it rises; in yet others, it barely budges. The combination of these fluctuations transforms into a "digital scent fingerprint"—a multidimensional vector where each sensor is responsible for its own note.
Then the mathematical part of the system steps in: algorithms compress this polyphonic signal down to a couple of axes, highlight patterns, and learn to distinguish images. In our press releases, this is described dryly—PCA, LDA, Random Forest... But in reality, it is much more beautiful: the electronic nose looks at a scent like a familiar silhouette. It doesn’t try to find a specific molecule of para-methyl-phenoxy-acetaldehyde in the air; it recognizes what it has already seen hundreds of times: fresh plastic, spoiled meat, light petroleum fractions. And it does this not by sniffing in the dark, like a human, but through machine learning methods. The more scents it has been introduced to during training, the subtler its ability to distinguish them later.
In reality, the electronic nose—which is comparable in size to our own—perceives the scent as a whole, just like we do. To break a scent down into its components, scientists use cumbersome gas analyzers. But the e-nose is compact. And it is not a single shiny metal probe, as we tend to imagine lab instruments, but a small team of sensors.
Instead of olfactory receptors, it uses the thinnest films of metal oxides. There are usually eight of them on a chip: oxides of manganese, zinc, cobalt, tin, and other elements from the middle of the periodic table that react to volatile compounds appearance. With every breath—or rather, every intake of the gas mixture—these oxide films change their electrical resistance (conductivity). In some films, it drops; in others, it rises; in yet others, it barely budges. The combination of these fluctuations transforms into a "digital scent fingerprint"—a multidimensional vector where each sensor is responsible for its own note.
Then the mathematical part of the system steps in: algorithms compress this polyphonic signal down to a couple of axes, highlight patterns, and learn to distinguish images. In our press releases, this is described dryly—PCA, LDA, Random Forest... But in reality, it is much more beautiful: the electronic nose looks at a scent like a familiar silhouette. It doesn’t try to find a specific molecule of para-methyl-phenoxy-acetaldehyde in the air; it recognizes what it has already seen hundreds of times: fresh plastic, spoiled meat, light petroleum fractions. And it does this not by sniffing in the dark, like a human, but through machine learning methods. The more scents it has been introduced to during training, the subtler its ability to distinguish them later.
«Он “видит” запах, он чувствует запах, как мы, — говорит Фёдоров. — Но для него нет валентности запаха. Он не может, как мозг, сказать: это приятный запах, а это неприятный. Ему всё равно. Он просто видит: вот запах номер раз, вот запах номер два, вот запах номер три. Только человек оценивает совокупность ароматов как приятную или неприятную. Если говорить образно, то наш электронный нос — это штука, которая ищет отпечатки пальцев. Мы работаем как с преступником: ищем те отпечатки пальцев, которые уже исследовали и можем снова увидеть. У нас даже есть такой рисунок — отпечаток пальца, наложенный на сенсор. Его использовали в качестве графической аннотации в журнал при публикации научной статьи».
Striving for Zero
The electronic nose hasn't been applied to symbols of Christmas yet, but that doesn't mean science is ignorant of their scents. And we are certain our nose wouldn't be bewildered. If we trained it on a massive dataset of coniferous scents and then held it up to a spruce, it would effortlessly identify the tree by β-pinene. It would distinguish firs, often sold in supermarkets as Christmas trees, by a mixture of α-pinene and camphene, and pines (rarely sold before the holidays, but coniferous nonetheless) by a mix of α-pinene and 3-carene. Where a human perceives just a "piney smell," the nose would see subtle divergences in the ratios of terpenes, manool, bornyl acetate, and dozens of other compounds.
With tangerines, it would be even simpler. Limonene would give a confident peak, γ-terpinene would add a soft green aroma, and methyl anthranilate would provide that specific, almost candy-like tangerine hue we carry from childhood into adulthood. The electronic nose would decompose all of this into neat vectors—and in the two dimensions of a PCA plot, a distinct tangerine "cluster" would instantly manifest.
It could even handle Olivier salad (the Russian festive staple). Moreover, it would likely distinguish yesterday's Olivier from today's better than we could. But the moment we moved to the main holiday question—the one from the article's headline—something happened that you least expect from the developer of an electronic nose. When asked, 'So, what does frosty freshness smell like?' the scientist replied...
With tangerines, it would be even simpler. Limonene would give a confident peak, γ-terpinene would add a soft green aroma, and methyl anthranilate would provide that specific, almost candy-like tangerine hue we carry from childhood into adulthood. The electronic nose would decompose all of this into neat vectors—and in the two dimensions of a PCA plot, a distinct tangerine "cluster" would instantly manifest.
It could even handle Olivier salad (the Russian festive staple). Moreover, it would likely distinguish yesterday's Olivier from today's better than we could. But the moment we moved to the main holiday question—the one from the article's headline—something happened that you least expect from the developer of an electronic nose. When asked, 'So, what does frosty freshness smell like?' the scientist replied...
And surprisingly, his answer turned out to be extremely precise. Because frosty freshness is not a scent, but rather its absence. In cold air, there are almost no volatile molecules: at sub-zero temperatures, evaporation slows down so drastically that the air becomes almost empty, scrubbed clean of background aromas of soil, plants, bacteria, and urban impurities. Snow further "sweeps" scents from the atmosphere like a soft adsorbing sponge. Our nose perceives the cold not as an aroma, but as irritation of the trigeminal nerve—that very physiological "chill" we habitually mistake for the scent of purity.
So, what do we actually have here, according to laboratory studies? Acetaldehyde, methyl formate, acetone, propanal, furan, isoprene, 2-methylpropanal, 2,3-butanedione, 2-methylfuran, 3-methylbutanal, 2-methylbutanal. These are all ketones and aldehydes. Often, the scent of fruits and beverages is defined specifically by aldehydes, as well as esters and terpenes. And the list of substances defining the smell of coffee, which I have open on my screen right now, contains 17 different organic compounds. Sometimes even more. There are even substances on this list that can be very dangerous in high concentrations. Take acetone, for example.
— Fedor Fedorov, Lead developer of the electronic nose
— Fedor Fedorov, Lead developer of the electronic nose
Он “видит” запах, он чувствует запах, как мы, — говорит Фёдоров. — Но для него нет валентности запаха. Он не может, как мозг, сказать: это приятный запах, а это неприятный. Ему всё равно. Он просто видит: вот запах номер раз, вот запах номер два, вот запах номер три. Только человек оценивает совокупность ароматов как приятную или неприятную. Если говорить образно, то наш электронный нос — это штука, которая ищет отпечатки пальцев. Мы работаем как с преступником: ищем те отпечатки пальцев, которые уже исследовали и можем снова увидеть. У нас даже есть такой рисунок — отпечаток пальца, наложенный на сенсор. Его использовали в качестве графической аннотации в журнал при публикации научной статьи».
We are working on many projects, and each one is important to me; every application is interesting. We would like to integrate the electronic nose with robots and gadgets so it can analyze food freshness, storage conditions, and indoor air quality. It could monitor microclimate, help in the kitchen, assess a person's health by the smell of their breath... Essentially, it can become part of the Internet of Things, making our lives simpler and better.
— Fedor Fedorov, Lead developer of the electronic nose
We are working on many projects, and each one is important to me; every application is interesting. We would like to integrate the electronic nose with robots and gadgets so it can analyze food freshness, storage conditions, and indoor air quality. It could monitor microclimate, help in the kitchen, assess a person's health by the smell of their breath... Essentially, it can become part of the Internet of Things, making our lives simpler and better.
— Fedor Fedorov, Lead developer of the electronic nose
Что здесь, собственно, есть, согласно лабораторным исследованиям. Ацетальдегид, метилформиат, ацетон, пропаналь, фуран, изопрен, 2-метилпропаналь, 2,3-бутандион, 2-метилфуран, 3-метилбутаналь, 2-метилбутаналь. Это всё кетоны и альдегиды. Зачастую запах фруктов и напитков определяют именно альдегиды, а также сложные эфиры и терпены. А в списке веществ, определяющих запах кофе, который открыт у меня сейчас на экране, 17 разных органических соединений. Иногда бывает больше. В этом списке есть и такие вещества, которые в высокой концентрации могут быть очень опасны. Вот, например, ацетон.
— Fedor Fedorov, Lead developer of the electronic nose
We are working on many projects, and each one is important to me; every application is interesting. We would like to integrate the electronic nose with robots and gadgets so it can analyze food freshness, storage conditions, and indoor air quality. It could monitor microclimate, help in the kitchen, assess a person's health by the smell of their breath... Essentially, it can become part of the Internet of Things, making our lives simpler and better..
To be honest, I don't know.
It 'sees' the scent, it feels the scent like we do. But for it, the scent has no valence. Unlike the brain, it cannot say: 'this is a pleasant smell, and this is an unpleasant one.' It is indifferent. It simply sees: here is scent number one, here is scent number two, here is scent number three. Only a human evaluates a combination of aromas as pleasant or unpleasant.
Speaking figuratively, our electronic nose is a device that searches for fingerprints. We work just like we do with a criminal: we look for those fingerprints we have already investigated and can recognize again. We even have an image like that—a fingerprint superimposed on a sensor. It was used as a graphical abstract for a journal when we published our scientific paper.
— Fedor Fedorov, Lead developer of the electronic nose
It 'sees' the scent, it feels the scent like we do. But for it, the scent has no valence. Unlike the brain, it cannot say: 'this is a pleasant smell, and this is an unpleasant one.' It is indifferent. It simply sees: here is scent number one, here is scent number two, here is scent number three. Only a human evaluates a combination of aromas as pleasant or unpleasant.
Speaking figuratively, our electronic nose is a device that searches for fingerprints. We work just like we do with a criminal: we look for those fingerprints we have already investigated and can recognize again. We even have an image like that—a fingerprint superimposed on a sensor. It was used as a graphical abstract for a journal when we published our scientific paper.
There are substances that have no smell for humans: carbon monoxide, carbon dioxide, oxygen. But the electronic nose senses them. There are gases that even it doesn't sense directly—nitrogen, for example. But the electronic nose can detect a change in the nitrogen content of the air by how the proportion of oxygen shifts. We see only a signal. But from this signal, we can understand that something in the air has changed—even if not a single human receptor would notice it.
— Fedor Fedorov, Lead developer of the electronic nose
So, what do we actually have here, according to laboratory studies? Acetaldehyde, methyl formate, acetone, propanal, furan, isoprene, 2-methylpropanal, 2,3-butanedione, 2-methylfuran, 3-methylbutanal, 2-methylbutanal. These are all ketones and aldehydes. Often, the scent of fruits and beverages is defined specifically by aldehydes, as well as esters and terpenes. And the list of substances defining the smell of coffee, which I have open on my screen right now, contains 17 different organic compounds. Sometimes even more. There are even substances on this list that can be very dangerous in high concentrations. Take acetone, for example.
Off-notes — a term used in sensory evaluation within the beverage, coffee, perfumery, food chemistry, and packaging industries. It designates an extraneous nuance that disrupts the familiar aroma and signals a defect. In coffee, this might be mustiness, rubber, or inappropriate burnt tones; in meat—hydrogen sulfide and cadaveric hints; in plastic—a characteristic "chemical" stench; in wines—sulfurous or vinegary notes.