Analysing an aerosol releasing article

Abstract

An apparatus for analysing an aerosol releasing article, such as a product of the tobacco or vaping industry, is disclosed The apparatus comprises means (22) for drawing an aerosol from the article, a plurality of sensors arranged to sense characteristics of the aerosol to produce sensor data, means (60) for predicting at least one sensory response from the sensor data using a model (70) relating sensor data to at least one sensory response, and means (72) for outputting a predicted sensory response to a user. This may allow an indication of how the article will be experienced by a user to be provided, which may help to reduce or avoid the need for human sensory panels.

Description

[0001] ANALYSING AN AEROSOL RELEASING ARTICLE

[0002] The present invention relates to techniques for analysing an aerosol releasing article, and in particular techniques for predicting sensory responses associated with delivery of an aerosol to the human mouth, lungs and / or respiratory tract. The present invention has particular, but not exclusive, application in analysing products of the tobacco and vaping industries.

[0003] In the tobacco industry, changes in product formulation / blend are a frequent occurrence. When such changes are made, sensory testing may be required to assure a similar or better sensory experience for the user. Other uses of sensory testing may be to match a product to the sensory experience of a competitor product; to produce a replacement product such as an e-cigarette or heated tobacco product with similar characteristics to an existing product; or to match a product to a particular market.

[0004] Currently in the tobacco industry, smoking panels are made up of trained and selected individuals that can smoke I vape and give standard responses regarding the sensory experience. Useful data is achieved by “calibrating” the individuals on a smoking panel with products that are designed to exemplify a specific sensory attribute such as “irritation”.

[0005] Assessment by panellists in the tobacco industry has some clear drawbacks that translate into practical limitations. These include cost, repeatability and reproducibility, ethics, and the declining population of adult smokers willing to participate in these studies. Considerable sums of money are spent annually by major tobacco companies on consumer and taste panels. An alternative that reduces these considerable disadvantages would be of major benefit.

[0006] It is known to analyse a smoking article using a smoking machine. For example, WO 2006 / 100493 A1 , the subject matter of which is incorporated herein by reference, discloses a smoking machine comprising one or more electrochemical cells for analysing the composition of a smoke stream from a smoking article. This can allow a gas or vapour phase chemical to be analysed directly in real time. The concentration of particular chemicals may be indicative of the brand of cigarette being smoked.

[0007] Known mechanical smoking systems are able to detect concentrations of particular chemicals, but they do not provide any indication of how the product might be experienced by the user.

[0008] It would therefore be desirable to provide techniques for determining a predicted sensory response to an aerosol which can reduce or avoid some of the drawbacks associated with the use of a smoking panel.

[0009] According to one aspect of the invention there is provided an apparatus for analysing an aerosol releasing article, the apparatus comprising: means for drawing an aerosol from the article; a plurality of sensors arranged to sense characteristics of the aerosol to produce sensor data; means for predicting at least one sensory response from the sensor data using a model relating sensor data to at least one sensory response; and means for outputting a predicted sensory response to a user.

[0010] The present invention may provide the advantage that, by providing a plurality of sensors arranged to produce sensor data, and means for predicting at least one sensory response from the sensor data using a model relating sensor data to at least one sensory response, it may be possible for the apparatus to provide an indication of how the article will be experienced by a user. This may help to reduce the need for human sensory panels, which in turn may help to address the costs, ethics and other drawbacks associated with the use of such panels.

[0011] The present invention is based on the realisation that the sensor data from a plurality of sensors may provide a profile or “fingerprint” which can be used to infer a human sensory experience. The sensors do not have to be specific for a particular sensation, but can instead be untargeted, in that they are not looking for a particular outcome. However, a combination of sensor data from a plurality of sensors may characterise the sensory response which it is desired to predict. In this way, it may be possible to infer a human sensory experience using sensor data from a plurality of disparate sensors.

[0012] Thus, the model may relate sensor data from a plurality of sensors to the or each sensory response. In this way, sensor data from a plurality of sensors may be used to produce the or each predicted sensory response. In other words, a single predicted sensory response may be produced using sensor data from two or more (disparate) sensors. This may allow a more accurate prediction of how the product will be experienced by the user to be produced.

[0013] The aerosol releasing article is preferably of a type which is used to deliver aerosol to the human body. For example, the aerosol releasing article may be of a type used to deliver an aerosol to the human mouth, lungs and / or respiratory tract. In one embodiment, the aerosol releasing article is a product of the tobacco or vaping industries. For example, the aerosol releasing article may be a cigarette, cigar, heated tobacco product, e-cigarette, vaping device, or any similar article. However, embodiments of the invention may be used with articles that deliver aerosols for other purposes, such as therapeutic applications.

[0014] The predicted sensory response may be a prediction of a response that will be produced in a human body when the aerosol is drawn into the human body. For example, the predicted sensory response may be a prediction of a response by one or more sensory receptors in the human body (for example, in the mouth, tongue, throat, lips, nose, eyes, ears and / or skin) which create a neurological impulse in the brain.

[0015] The predicted sensory response may comprise a quantitative description of at least one sensation reported by a (human) sensory panel. For example, the predicted sensory response may be in the form of a numerical value, or a classification such as “high”, “medium” or “low”, or any other appropriate description of a human sensory experience.

[0016] The predicting means is preferably arranged to predict a plurality of sensory responses from the sensor data. This may allow a more complete picture of how the product is likely to be experienced by a user to be provided. In one embodiment, the sensory response may comprise at least one of: Impact; Irritation in Throat; Tingling in Mouth; Mouthful of Aerosol; Heat Sensation; Mouth Cooling; Draw Resistance; Mouth Drying; or any equivalents thereof. In this context, “Impact” may defined as a sudden, sharp but short lived sensation which is noticed when the aerosol makes contact with the chest during inhalation - sometimes described as “chest hit”; “Irritation in Throat” may be defined as a burning, prickling sensation in the throat that lingers; Tingling in Mouth may be a prickling, tingling, peppery sensation in the mouth; Mouthful of Aerosol may be the impression of the amount of smoke entering the mouth and occupying the mouth during puff taking; Heat Sensation may be a combination of temperature of filter on lips and aerosol in mouth; Mouth Cooling may relate to the perceived presence and quantity of menthol or similar sensation; Draw Resistance may relate to the perceived amount of effort required to draw the smoke from the lit product into the mouth; and Mouth Drying may relate to the drying effect in mouth after the aerosol exhaled. Of course, these descriptors and definitions are examples only, and other descriptors and definitions may be used instead or as well. For example, in some cases other terms may be used to describe similar, related or equivalent sensory experiences.

[0017] It has been found that it meaningful results may be obtained if a number of different types of sensor are used to produce the sensor data. Thus, the plurality of sensors may comprise a plurality of different types of sensor. For example, the plurality of sensors may be arranged to sense at least one (and preferably at least two) of: a physical characteristic of the aerosol; a chemical characteristic of the aerosol; and a size characteristic of the aerosol. The physical characteristic may comprise, for example, temperature, pressure drop, etc. The chemical characteristic may relate to the presence and / or concentration of a chemical substance, and may be obtained using, for example, an electrochemical sensor such as a Tin Oxide environmental sensor, spectrometry, pH sensors, colorimetric methods, etc. The size characteristic may relate to the size of aerosol particles and may be obtained, for example, using dynamic aerosol diameter measurement, optical scattering, etc. If desired, other types of measurement could be used as well as or instead. This may allow a characterising pattern or fingerprint for the article to be produced. In one embodiment of the invention, the characteristics of the aerosol comprise at least one of: an aerosol density and / or obscuration; aerosol particle size distribution; presence and / or concentration of one or more chemicals; aerosol temperature at one or more locations; aerosol humidity; and pressure and / or pressure drop. It has been found that this combination of characteristics (or a subset thereof) may be particularly effective in predicting a human sensory experience. However, one or more other characteristics may be used instead or as well.

[0018] The appropriate sensors may be provided in order to sense the above aerosol characteristics. Thus, in one embodiment of the invention, the plurality of sensors comprises at least one of: an aerosol density and / or obscuration sensor; a particle size distribution sensor; an electrochemical sensor; a temperature sensor; a humidity sensor; and a pressure sensor.

[0019] At least one of the plurality of sensors may be arranged to sense a physical characteristic of the aerosol in the intensity and time domains. For example, samples of the physical characteristic may be taken at different points in time (for example, during a period of aerosol residency in a chamber). In this case, the model may relate sensor data in the intensity and time domains to at least one sensory response. This may allow the apparatus to produce a predicted sensory response which is related to how long a particular sensation lasts. This may help to provide a predicted sensory response which corresponds to a human sensory experience.

[0020] The means for predicting a sensory response may comprise the model relating sensor data to at least one sensory response. The model may be a mathematical model, such as a statistical model, and may comprise an algorithm for inferring a human sensory response from the sensor data. By using a model relating sensor data to at least one sensory response, it may be possible to infer a human sensory experience from the sensor data.

[0021] In one embodiment, the model comprises a machine learning algorithm. The machine learning algorithm may use, for example, a neural network, linear regression, logistic regression, clustering, decision trees, random forests, or any other appropriate approach. The machine learning algorithm is preferably trained in advance. For example, the machine learning algorithm may be trained in advance using sensor data and reported sensations from a sensory panel. The sensor data are preferably obtained from a plurality of test products (which may be “standard” products), and the reported sensations are preferably obtained from the same or corresponding test products, preferably under the same or corresponding conditions.

[0022] In one embodiment, the model uses a decision tree algorithm to predict the sensory response. Decision trees use a branching sequence of linked decisions that can be represented with a tree diagram. The decision tree algorithm may use a tree model, such as decision tree, random forest or a gradient boosted tree model such as XGBoost. It has been found that this approach may be effective in predicting a sensory response from the sensor data.

[0023] However, in alternative embodiments, the model may use another machine learning approach, such as artificial neural networks (ANNs), support-vector machines (SVMs), regression analysis, Bayesian networks, Gaussian processes, etc. instead or as well. Furthermore, in some embodiments, it may be possible for the sensory response model to be in the form of one or more mathematical formula such as a polynomial equation, for example, provided by a linear mixed effect model.

[0024] If desired, a plurality of models could be provided, each corresponding for example to a particular type of product and / or set of test variables. In this case, the apparatus may be arranged to select the appropriate model, for example, automatically, or in response to a user input.

[0025] The apparatus may further comprise means for pre-processing the sensor data before it is input to the model. The means for pre-processing the sensor data may be arranged, for example, to reduce the amount of data which is input to the model. This may be desirable, for example, where the characteristics of the aerosol are sampled at a relatively high sampling rate (compared to a puff), in order to reduce the amount of data which needs to be processed. In one embodiment, the apparatus may comprise means for flattening the sensor data to obtain a set of features for each sensor. In this case, the model may be arranged to predict at least one sensory response from the flattened sensor data. The process of flattening may comprise, for example, transforming the data into another form and / or reducing the amount of data. The features may be for example synthetic information such as mean, standard deviation, linearity, entropy, curvature etc. This may help to reduce the amount of processing and facilitate the matching of sensor data to panellist responses.

[0026] The means for drawing an aerosol from the article may comprise a puff engine. This may allow aerosol to be drawn from the article in a controlled manner. The puff engine may comprise any suitable device for applying a negative pressure to the article. The puff engine may be programmed with an appropriate puff profile, such as volume and / or shape of puff.

[0027] The apparatus may further comprise a chamber, and the means for drawing an aerosol from the article may be arranged to draw the aerosol into the chamber. In this case, at least one of the plurality of sensors may be arranged to sense a physical characteristic of the aerosol in the chamber. The chamber may for example mimic physical features of a human, such as the mouth, throat or lungs. This may allow the apparatus to mimic the conditions which would be experienced by the aerosol in the human body.

[0028] If desired, the volume of the chamber may be adjustable. This may allow, for example, the testing to be adapted to different products, conditions, or intended end uses.

[0029] If desired, some or all of the sensors could be provided in the same chamber. However, it has been found that this may lead to space constraints or may compromise the placement of the sensors. In order to address this issue, the apparatus may comprise a plurality of chambers and / or a plurality of puff engines. In this case, each of the chambers may comprise at least one sensor. This may allow a larger number of sensors to be deployed without compromising their placement due to space constraints. Each chamber may be provided with its own puff engine, or a puff engine may be shared by one or more chambers.

[0030] In one embodiment, the apparatus is arranged to draw an aerosol from a plurality of aerosol releasing articles, and the plurality of sensors is arranged to sense at least one characteristic of the aerosol from each of the plurality of articles. The sensory response may then be predicted from sensor data from a plurality of articles. Each of the articles may be a different sample of the same product. This may allow a larger number of sensors to be deployed without compromising their placement. Each article may be puffed with its own puff engine, or a puff engine may be shared by one or more articles. The aerosol from each article may be drawn into a separate chamber, or alternatively, the aerosol from two or more articles could be drawn into the same chamber.

[0031] In one embodiment, the plurality of sensors comprises a plurality of temperature sensors arranged to sense a temperature of the aerosol in a plurality of different locations. For example, where the aerosol is drawn into a chamber, the plurality of temperature sensors may sense the temperature at different locations in the same chamber and / or at different locations in different chambers (or elsewhere e.g. the temperature of the smoking article itself at one or more locations such as the filter or mouthpiece). It has been found that this may help to predict one or more reported sensations, such as impact and / or taste intensity, and / or a reported sensation which can be assessed at different sites, such as different areas of the mouth, throat or nose.

[0032] In one embodiment, the plurality of sensors comprises a humidity sensor arranged to sense a humidity of the aerosol. For example, where the aerosol is drawn into a chamber, the humidity sensor may sense a humidity of the aerosol in the chamber. The humidity sensor may sense relative humidity, which is preferably the amount of moisture present relative to the amount that would be present if the air were saturated. This may help to predict one or more reported sensations, such as irritation, impact and / or mouth drying.

[0033] In one embodiment, the plurality of sensors comprises a pressure sensor arranged to measure a pressure change during a puff. For example, where the aerosol is drawn into a chamber, the pressure sensor may sense a pressure drop in the chamber during a puff. This may help to predict one or more sensations, such as draw effort.

[0034] In one embodiment, the plurality of sensors comprises a weighing device arranged to determine a mass of particle deposition. For example, where the aerosol is drawn into a chamber, the weighing device may be arranged to determine a mass of particle deposition in the chamber. This may help to predict one or more sensations, such as mouth coating and / or taste intensity. The weighing device may be for example a quartz microbalance or any other appropriate device for measuring mass.

[0035] In one embodiment, the plurality of sensors comprises an optical sensor arranged to measure particulate scattering of the aerosol. This may help to predict one or more sensations, such as mouthful of smoke. The optical sensor may be provided, for example, in a conduit through which the aerosol is drawn, or elsewhere such as in a chamber.

[0036] In one embodiment, the plurality of sensors comprises at least one electrochemical cell. The electrochemical cell may be arranged to sense a concentration of one or more substances such as a compound or type of compound. For example, the electrochemical sensor may be arranged to sense a concentration of volatile organic compounds (VOC), oxygen (O2), carbon monoxide (CO), carbon dioxide (CO2), nicotine and / or any other compound or element. This may help to predict one or more sensations, such as impact and taste intensity.

[0037] At least one electrochemical cell may be a solid-state sensor such as a tin oxide sensor. This may allow a relatively inexpensive and readily available sensor to be used.

[0038] In one embodiment, the plurality of sensors comprises a plurality of electrochemical cells. In this case, each of the electrochemical cells may be arranged to sense a different substance (such as a different compound or type of compound) from at least one of the other electrochemical cells. This may help to predict one or more sensations, such as impact and taste intensity. Each of the electrochemical cells may be provided at the same location or a different location from the others.

[0039] In one embodiment, the plurality of sensors comprises a particle sizing device. This may help to predict one or more sensations, such as mouth coating, mouthful of smoke and mouth drying. The particle sizing device may be provided, for example, at the exhaust side of a puff engine, or elsewhere.

[0040] In one embodiment, the plurality of sensors comprises a spectroscopic sensor. This may help to predict one or more sensations, such as impact and taste intensity. The spectroscopic sensor may be provided, for example, at the exhaust side of a puff engine, or elsewhere.

[0041] The means for outputting a predicted sensory response to a user may comprise a display, or any other appropriate means for presenting information to the user such as a printer, set of lights, speaker, or other visual or audio indicator. The predicted sensory response may be presented to the user in any appropriate way, such as a series of heat maps, spider graphs or pseudo videos of data change. Alternatively, or in addition, the means for outputting a predicted sensory response may be arranged to store the predicted sensory response and / or to transmit the predicted sensory response to another device.

[0042] According to another aspect of the invention there is provided an apparatus arranged to analyse an aerosol releasing article, the apparatus comprising: a device (such as a puff engine) arranged to draw an aerosol from the article; a plurality of sensors arranged to sense characteristics of the aerosol to produce sensor data; a processor arranged to predict at least one sensory response from the sensor data using a model which relates the sensor data to the sensory response; and a user interface arranged to output a predicted sensory response to a user. Corresponding methods may also be provided. Thus, according to another aspect of the invention there is provided a method of analysing an aerosol releasing article using an analysis apparatus, the method comprising: drawing an aerosol from the aerosol releasing article; sensing characteristics of the aerosol using a plurality of sensors to produce sensor data; predicting a sensory response from the sensor data using a model relating sensor data to at least one sensory response; and outputting the predicted sensory response to a user.

[0043] The method may comprise converting the sensor data to at least one predicted sensory response using the model. The method may further comprise training the model using training data. For example, the model may be trained in advance using sensor data obtained from a plurality of test products, and reported sensations obtained from the same or corresponding test products. Thus, the method may further comprise the step of training the model using sensor data and reported sensations from a sensory panel.

[0044] Features of one aspect may be used in combination with any other aspect. Any of the apparatus features may be provided as method features and vice versa.

[0045] Preferred features of the present invention will now be described, purely by way of example, with reference to the accompanying drawings, in which:

[0046] Figure 1 shows parts of an apparatus that may be used to collect sensor data in an embodiment of the invention;

[0047] Figure 2 shows parts of a sensory testing apparatus in an embodiment of the invention;

[0048] Figure 3 shows in more detail parts of a supervisory and control unit;

[0049] Figure 4 shows how sensory descriptors may be presented to the user in one exemplary embodiment; and

[0050] Figure 5 shows steps carried out by a sensory testing apparatus in one embodiment. Various industries rely on the human sensory response for the success of the products that they provide. One such industry is the tobacco industry. Changes in tobacco, or product formulation / blend, are a frequent occurrence in the industry. Furthermore, market-to-market sensory expectations may vary.

[0051] Sensory testing may therefore be used to ensure that the sensory experience matches market expectations.

[0052] It should be noted that “sensory experience” when considered by the tobacco industry is more than might be defined as “taste” or “smell” but describes the whole mouth, lung and nose experience that occurs when taking an aerosol into the body. Terms such as “mouth feel”, “fullness”, “after-taste” and “dryness” are common and do not feature at all in term of taste but are what might be called the “hepatic” experience of the product aerosol. It is notable that a single response will be the result of a number of physical reactions to the smoke.

[0053] Currently in the tobacco industry, smoking panels are made up of trained and selected individuals that can smoke and give standard responses regarding the sensory experience. These sensory experiences are by necessity qualitative in nature although there is an approximate quantitative element (e.g. high, medium, low, or a number from 1 to 5). Useful data is achieved by “calibrating” the individuals on a smoking panel with products that are designed to exemplify a characteristic such as “irritation”.

[0054] Table 1 lists various potential sensory descriptors that could be used by panellists to describe a product.

[0055] Table 1

[0056] It will be appreciated that the sensory descriptors shown in Table 1 are exemplary only, and other descriptors and sensations could be used and described instead or as well.

[0057] Considerable sums of money are spent annually by major tobacco companies on consumer and taste panels. An alternative that reduces these considerable costs would be of major benefit to manufacturers.

[0058] Reproducibility and repeatability are generally a problem when using people’s perception to assess a subjective effect or impact. This can be in part mitigated by good practice. This involves training the people involved, creating a common language of descriptors that explain the sensory experience (although the language may change from company to company) and “calibration” of an individual’s responses with others using standard or control products. However, despite these precautions, variability is not unknown.

[0059] The ethics of using human panellists remains a significant problem. Panellists are exposed to cigarette smoke and despite the selection of existing smokers for the task, this is considered an ethical grey area. Furthermore, using human smoking panels does not align with strategies of tobacco companies to transition to a smoke free world with a range of low harm products.

[0060] If the variability in results from human panellists can be further minimised, the cost per test reduced, and the ethical dilemma reduced or eliminated even as part of a screening tool this would form a significant advance in the study of sensory response to tobacco aerosols.

[0061] Embodiments of the present invention aim to at least partially replace human assessment with an objective assessment from electronic or other sensing means. It is recognised that electronic systems are unlikely to mimic fully a sensory panel so the breadth of experience / data may be narrowed. Moreover, the final arbitrator will normally be human as to the acceptability or otherwise of a product. However, an electronic system may be able to provide an early stage screening device that can eliminate some of the routine testing used in product development, allowing only the most promising candidates to go to full panel assessment or possibly allowing more product iterations to find a sensory sweet spot.

[0062] A further use would be in establishing if a deliberate or accidental change in blend in a product has any noticeable impact on consumer experience, for example a change in blend (Virginia / Oriental / Burley tobacco ratio), or a change in casings (e.g. excess liquorice). This approach may then become part of the quality assurance of a product and not just a research tool.

[0063] A fundamental problem which embodiments of the invention aim to address is the replacement, or part replacement, of human sensory panels with screening equipment with a meaningful output (at least one predicted sensory response) built on sensors with an electronic output.

[0064] It has been found, pursuant to the present invention, that, by deploying multiple sensors that together form a pattern of response characteristic of the aerosol under test, it may be possible to infer human sensory experience through pattern matching.

[0065] In some embodiments, a puff engine, or series of puff engines, is provided that can be programmed with standardised puffing profiles (that could be mathematically defined or replicate recordings of human puffing topography), pulling on a smoking article such as a cigarette, heated tobacco product, e- cigarette or vaping device. Aerosol from the article enters into one or more chamber. The chamber(s) could be size adjustable to represent the volume of the mouth, a single puff or lungs. The aerosol resides in the chamber for analysis until another puff is drawn or evacuated after a set residency time. The chamber is fitted with sensors and possibly a facility for drawing smoke aliquots into more sensitive apparatus where dilution may be necessary. The aerosol is held for a set time - a bypass for flushing with clean air may also be available - and then exhausted or possibly exhaled into another chamber with a second array of sensors possibly through some form of filter that mimics the body’s filtration mechanisms.

[0066] A series of chambers which can simultaneously puff on different articles of the same product using the same profile of puffing could be used, so allowing a large array of sensors to be deployed without having to compromise the number of sensors used or their placement due to space constraints. In this case it is assumed that if the same product is puffed the same sensory experience is present.

[0067] During puffing and smoke residency, the sensor electronic output is logged (time stamped for synchronicity) and this creates an array of data (electronic sensor pattern) that is characteristic of the sensory experience that a panellist experiences. Both the intensity of sensor response and the time when this response occurs is relevant (the pattern of response is not just a single reading).

[0068] By combining the individual sensor patterns of response in the time domain, a sensory “fingerprint” of the product under test can be created. This is extended by considering each puff in a sequence to be part of the fingerprint for the whole user experience, not just a single puff.

[0069] In order to be able to provide a predicted sensory response, the system is first calibrated against a panellist response.

[0070] There are in existence “standard” products that have been characterised with panellists such that the products have a known sensory evaluation in the terms specified by the tests, including the sensations and the intensity of these sensations. With these products, patterns can be generated that characterise the smoking mechanics, the smoking sensations, the taste sensations and the residual sensations. By using a range of standards, the specific patterns associated with a reported sensation can be identified. In this way a set of analytical measurements can be made that directly correlate to the panellist responses in terms of intensity and presence. Alternatively, a series of “head-to-head” comparisons can be made of sensory panel data and sensor array response that is characteristic of a particular product that creates a set of standardised data upon which to create a model for sensor / sensory response that can be generally applied to new systems under test.

[0071] It is unlikely that there is a direct correlation between a single sensor measurement and a panel assessment. It is more likely that there will be a range of sensor outputs that correlate to a sensory panel data point. This panel response will not in all cases “scale” simply or linearly with the sensor matrix response - some will be dominant, some counter, and some only having a measurable effect when viewed in context of other sensor responses. It is possible from an array of sensory panel data and sensor data to establish those sensors that can be used to define a sensory response and those that have little or no correlation with the sensory response under consideration.

[0072] Moreover, the continuing sensitivity of a system (which may be poisoned by the chemicals in the aerosol or physically coated making a barrier to the aerosol) can be established and confirmed by periodic checks with these standard products.

[0073] The sensors do not have to be specific for a particular sensation but can instead be “untargeted” in that they are not looking for a particular outcome but when combined with other sensors can form an array that characterises the identified response. The arrays can have strong correlations to reported sensations, weak correlation, or even negative correlation. For example, particle size may have a negative or positive correlation to mouth feel; aerosol density a strong correlation; temperature of aerosol a weak correlation; humidity a strong correlation; pH a negative correlation; and so on.

[0074] The “untargeted” nature of the approach (the intention is not necessarily to measure specific chemicals, for example, but instead to map a human response onto a number of connected measurements) combines physical measurements (weight, pressure drop, etc.) with chemical measurements (Tin Oxide environmental sensors, spectrometry, pH, potentially colorimetric methods and so on) and classical sizing (dynamic aerosol diameter measurement, optical scattering, etc.) and so on into a pattern for a product experience that changes with time (puff), in the intensity and time domain forming a characterising pattern or fingerprint for a product.

[0075] The data array can be processed in a number of different ways and displayed to the system user, for example, as a series of heat maps, spider graphs or pseudo videos of data change. Defined algorithms are used to “translate” the array of sensor data to a set of panel sensory descriptors such as “high throat hit”.

[0076] This data transformation forms a major challenge in developing sensory replacement technology as is the cross correlation to human sensory panel response. The system of sensing needs to correlate with the sensory panel. For example, each sensory panel member’s “mark” may need to be mapped onto the sensors, with different magnitudes involved and then validated against “unknowns”.

[0077] This becomes more complex when the time domain is added to the equation. Some of the panel responses may refer to how long a flavour or experience lasts. This means that the sensor data should be collected in a time domain and in some cases a spatial domain as well as in intensity.

[0078] Figure 1 shows parts of an apparatus that may be used to collect sensor data in one exemplary embodiment of the invention. Referring to Figure 1 , the apparatus comprises a holder 10 for holding an aerosol producing article 12, a first conduit 14, an analysis chamber 16, a particulate filter 18, a second conduit 20, a puff engine 22 and a regulated air supply 24.

[0079] The aerosol producing article 12 may be, for example, a cigarette, cigar, heated tobacco product (HTP), e-cigarette, vaping device, or any similar article. The holder 10 is arranged to hold the end of the article 12 which interfaces with the consumer, such as the filter or mouthpiece. The holder 10 comprises at least one seal, such as a lip seal, which forms a seal with the article and mimics the lips of the consumer. The holder 10 is in fluid communication with the analysis chamber 16 via the first conduit 14. The analysis chamber 16 is in fluid communication with the puff engine 22 via the second conduit 20. The analysis chamber 16 could be size adjustable, for example, to represent the volume of the mouth, a single puff or lungs (although this is optional). This could be achieved, for example, by means of extending bellows, extension rings or sliding seals and could be automated. For example, a stepper motor may be provided which can be actuated to increase and decrease the size of the chamber. The chamber 16 is fitted with sensors and possibly a facility for drawing smoke aliquots into more sensitive apparatus where dilution may be necessary.

[0080] The puff engine 22 is used to draw puffs from the article 12 to produce an aerosol stream. The puff engine 22 is programmed with standardised puffing profiles, which may be mathematically defined or may replicate recordings of human puffing topography. The particulate filter 18 is used to collect particles in the aerosol stream. The particulate filter 18 is optional and can be removed or replaced.

[0081] In operation, the puff engine 22 draws puffs from the aerosol producing article 12 to produce an aerosol stream. The aerosol stream is drawn from the article 12, through the first conduit 14 into the analysis chamber 16. The aerosol resides in the chamber 16 for a set residency time. The aerosol stream then passes through the particulate filter 18 and the second conduit 20 to the puff engine 22. The aerosol is exhausted from the puff engine 22 through an exhaust 44.

[0082] In the arrangement of Figure 1 , the apparatus is provided with an optical sensor 30, a first temperature sensor 32, a second temperature sensor 33, a third temperature sensor 49, a humidity probe 34, a quartz microbalance 36, a pressure drop sensor 38, an electrochemical sensing array 40, an impinger or other capture device 42, a particle sizing device 46 and a spectral sensor 48. An optional dilution valve 50 is provided for the particle sizing device 46 and spectroscopic sensor 48.

[0083] The optical sensor 30 is provided in the first conduit 14 and is used to measure particulate scattering or “thickness” of the aerosol. The first temperature sensor 32 is provided towards the front of the chamber 16, used to measure the temperature of the aerosol close to the point at which it enters the chamber. The second temperature sensor 33 is provided towards the rear of the chamber 16, used to measure the temperature of the aerosol close to the point at which it exits the chamber. The third temperature sensor 49 is provided towards the smoking article 10 and is used to measure the temperature of one or more points of the smoking article such as the filter temperature. The humidity probe 34 is provided in the chamber 16, and used to measure the humidity of the aerosol in the chamber 16. The quartz microbalance 36 is used to measure the weight of particles in the chamber 16. The pressure drop sensor 38 is used to measure a change in pressure in the chamber 16.

[0084] The electrochemical sensing array 40 is provided in the second conduit 20 and comprises a plurality of electrochemical cells. Each of the electrochemical cells is used to measure the concentration of a particular gas or vapour in the aerosol stream. By way of example, the electrochemical sensing array 40 may include sensors which detect concentrations of volatile organic compounds (VOCs), oxygen, carbon monoxide (CO), carbon dioxide (CO2), nicotine, etc. Amplifiers may be included to amplify the output of the sensors. In this example, the electrochemical sensors are shown in the second conduit 20, downstream of the particulate filter 18, to reduce the risk of contamination. However, some or all of the electrochemical cells could be provided elsewhere, such as in the chamber 16 or exhaust 44.

[0085] The impinger 42 may be provided at the entry to the puff engine 22, and is used to collect particle samples from the aerosol stream. The particle sizing device 46 is used to measure the size of particles in the exhaust stream from the puff engine 22 when the filter 18 is not fitted. The spectroscopic sensor 48 is provided on an outlet from the particle sizing device 46. The spectroscopic sensor 48 is a device such as a nondispersive infrared (NDIR) device or a mass spectrometer (MS). The spectroscopic sensor 48 is tuned to quantify several chemicals in real time. The spectroscopic sensor 48 may use, for example, time-of-flight mass spectrometry (TOFMS), Ion mobility spectrometry (IMS), Quadrupole mass spectrometry, or any other suitable technique. Also shown in Figure 1 are two shut valves 26, 28. The shut valves 26, 28 are in fluid communication with the analysis chamber 16. The first shut valve 26 opens to the atmosphere and the second shut valve 28 is connected to the air supply 24. The shut valves 26, 28 can be actuated so as to allow the analysis chamber 16 to be purged between puffs.

[0086] It will be appreciated that the arrangement shown in Figure 1 is purely illustrative and should not be seen as limiting the type or arrangement of sensor or other component. For example, some sensors may be omitted, some sensors may be located elsewhere, and other sensors may be provided instead or as well.

[0087] During puffing and aerosol residency, data from the optical sensor 30, temperature sensors 32, 33, 49, humidity probe 34, quartz microbalance 36, pressure drop sensor 38, electrochemical sensing array 40 and impinger 42 are logged and time stamped for synchronicity. After the aerosol has been exhausted, data from the particle sizing device 46 and spectroscopic sensor 48 are also logged. This creates an array of sensor data (electronic sensor pattern) that is characteristic of the sensory experience that a panellist experiences. Both the intensity of sensor response and the time when this response occurs is recorded (the pattern of response is not just a single reading). Control of puff is by a computer system (not shown in Figure 1 ) as is the data gathering.

[0088] In this embodiment, specific analysis, or even highly accurate analysis, is not the objective. By using an array of sensors, “patterns” of response are created, so relative sensitivities and the stability of these sensitivities is a desirable feature / outcome.

[0089] Table 2 summarises results of investigations carried out to determine where each sensor might contribute to one of the sensory responses that a panellist might rate. Table 2

[0090] The sensor data will be in different forms and will be a mixture of continuous data generated in the time domain (for example, from the electrochemical array), readings at a discrete time intervals through the puff cycle (for example, from the optical sensor) or “spectra” taken at discrete time intervals that contain multiple areas or interest (for example, from the spectroscopic sensor or the particle sizing device). Consequently, data will schematically have a time and numerical element.

[0091] Data analysis is then used to correlate the sensor data to specific sensory experiences. The analytical problem to be solved is to exploit the sensor data to predict new unknown panellist responses. In order to achieve this, the system is first calibrated against a panellist response.

[0092] For example data is gathered from a sensory panel on a range of representative products under different environment conditions. In each case the panellists scored a set of sensations on a 1-10 scale for each puff taken. In parallel, data is gathered via the electronic sensory rig for the same products tested under the same conditions at the same time. From this, it is possible to develop sensory predictive models.

[0093] Feature engineering is preliminary to train a predictive model. Two issues to be addressed are: • Data misalignment. Panel data and sensor data are not natively connected at a single row level, meaning that the data have an N x M potential relationship, where N is the number of panellists and M is the number of runs on the sensory rig for a given device under a given environmental condition.

[0094] • Featuring of time series. Sensor data consists of a set of time dependant variables with sampling rate of, for example, 1 or 0.1 seconds, meaning that it has collected 30 or 300 data points per puff, whereas the panellist response is 1 data point per puff.

[0095] This assures that panellists responses are independent, as well as runs of the sensory rig. This allows the data to be “flattened” and responses to be represented as mean, median and quantiles for each puff across each panellist, leading to a single value for device, condition and puff.

[0096] Table 3 shows an example of how the panellist responses may be flattened to mean, median and quantiles for each puff.

[0097] Table 3

[0098] Q1 is the generic question submitted to panellists

[0099] The sensor data once flattened on the time dimension, create a set of features for each device, condition, run and puff (in this context, “features” being synthetic information such as mean, standard deviation, linearity, entropy, curvature and so forth).

[0100] Table 4 shows an example of how the sensor data may be flattened to a set of features for each device, condition, run and puff. Table 4

[0101] Vk is the k-th variable; fh is the h-the feature (for the current variable)

[0102] Standards (known products with panel sensory responses at the ends of the range, e.g. 1 or 9) is an information set that proved effective for increasing the quality of predictions, helping models to calibrate on new devices or conditions.

[0103] A set of models can be tested under different situations. These are summarised in Table 5. Table 5

[0104] Combinations of all possible situations can be evaluated using Minimum Absolute Error (MAE) (for regression) and accuracy (for classification). MAE is the average of absolute differences between predicted and true value. Accuracy is the ratio between correct predictions and all predictions. Considerations are also given in terms of expected inferential power of new devices and conditions. From this, the following results and conclusions may be obtained:

[0105] • Standards of new devices or conditions are desirable to provide reliable predictions.

[0106] • Classification models are accurate.

[0107] • Puff results as an important predictor.

[0108] • Linear mixed effect model was found to be a suitable model, although it does not overperform random forest or decision tree.

[0109] Calibration follows in general the same process as data acquisition, the principal difference being the use of multiple calibration pieces of known sensory attributes. The sensor array responds to these test pieces and characteristic levels are set so the “pattern” created matches the ideal pattern that would be stored in a database. Hence a sensor-by-sensor calibration adjustment can be obtained by stretching or squashing the response to align with one that is expected.

[0110] Tests carried out by the present applicant have revealed that meaningful results (in terms of predicted sensory responses) can be obtained using the following sensor data:

[0111] • Aerosol density I obscuration

[0112] • Particle size distribution

[0113] • Data from at least one chemical sensor (organics / volatiles)

[0114] • Aerosol relative humidity (RH)

[0115] • Aerosol temperature at two or more places

[0116] • Smoking article temperature at one or more places

[0117] • Pressure drop

[0118] However, not all of the above sensor data need be present, and other sensor data could be used as well or instead. It has been found that, using the above sensor data it may be possible to predict one or more of the following sensory responses:

[0119] • Impact (Chest Hit - sudden and short lived)

[0120] • Irritation in Throat (burning, prickling sensation that lingers)

[0121] • Tingling in Mouth (also part of irritation)

[0122] • Mouthful of Aerosol

[0123] • Heat Sensation (combination of temperature of filter on lips and aerosol in mouth)

[0124] • Mouth Cooling (menthol or similar sensation)

[0125] • Draw Resistance

[0126] • Mouth Drying (drying effect in mouth after aerosol exhaled)

[0127] Thus, in preferred embodiments, at least some of the above sensor data is used to predict one or more of the above sensory responses (or equivalents thereof).

[0128] Where a number of sensors are to be used, a series of chambers can be provided which can simultaneously puff on the same product using the same profile of puffing. This can allow a large array of sensors to be deployed without having to compromise the number of sensors used or their placement due to space constraints.

[0129] Figure 2 shows parts of a sensory testing apparatus in an embodiment of the invention. Referring to Figure 2, the apparatus comprises a plurality of holders 10i - 10N, a plurality of analysis chambers 161 - 16N, a plurality of puff engines 22i - 22N, a data logging unit 56, a puff control unit 58, a supervisory and control unit 60, a display unit 62 and a user interface 64. Each of the holders 10i - 10N is arranged to hold a respective aerosol producing article 12i - 12N. Each holder 10i - 10N is in fluid communication with a corresponding analysis chamber 161 - 16N. The analysis chambers 161 - 16N may be size adjustable, although this is optional. Each analysis chamber 161 - 16N may be the same size as or a different size from the others. Each analysis chamber 161 - 16N is in fluid communication with a corresponding puff engine 22i - 22N. The holders 10i - 1 ON, analysis chambers 161 - 16N, and puff engines 22i - 22N may be for example as described above with reference to Figure 1 . Any of the other components shown in Figure 1 may also be provided (although some may be omitted).

[0130] In the arrangement of Figure 2, a first sensor array 52i - 52N is associated with each of the analysis chambers 161 - 16N. The sensors in each first sensor array may be for example in the analysis chamber, or in a conduit leading to or from the analysis chamber. A second sensor array 54i - 54N is associated with each of the puff engines 22i - 22N. Each second sensor array 54i - 54N is on the exhaust side of the corresponding puff engine. Each of the sensor arrays 52i - 52N, 54I - 54N contains one or more sensors, which may be, for example, one or more of the sensors described above with reference to Figure 1 . For example, each first sensor array 52i - 52N may comprise one or more of an optical sensor, a temperature sensor, a humidity probe, a quartz microbalance, a pressure drop sensor, a electrochemical sensing array and an impinger. Each second sensor array 54i - 54N may comprise one or more of a particle sizing device and spectral sensor.

[0131] Each sensor array 52i - 52N, 54I - 54N may contain different sensors from the other sensor arrays, or at least some of the same sensors, or any appropriate combination. If desired, one or more of the sensor arrays may be omitted. If desired, any other appropriate sensor may be provided as well or instead. In Figure 2 the sensor arrays 52i - 52N, 54I - 54N are shown associated with the analysis chambers and puff engines for simplicity. However, in practice, at least some of the sensors themselves may be in conduits between components or at the exhaust, for example, as shown in Figure 1 , or elsewhere in the system.

[0132] In one non-limiting embodiment, the sensor arrays 52i - 52N, 54I - 54N comprise multiple (two or more) temperature sensors, multiple (two or more) chemical sensors based on tin oxide technology, an optical obscuration sensor, a pressure drop sensor, a humidity sensor, a particle size sensor, a pH sensor and / or a nicotine sensor.

[0133] In operation, each of the puff engines 22i - 22N is used to draw puffs from each of the aerosol producing article 12i - 12N to produce aerosol streams. In some arrangements, one puff engine could be used to puff two or more articles. The puff engines 22i - 22N are controlled by the puff control unit 58. The puff control unit 58 is programmed with puffing profiles, which may be mathematically defined or may replicate recordings of human puffing topography. The puffing profiles may be the same for each puff engine, or they may differ.

[0134] The aerosol streams from the articles 12i - 12N are drawn into the analysis chambers 161 - 16N. The aerosols are held in the chambers 161 - 16N for a set residency time and are then exhausted. Each sensor in the sensor array 52i - 52N senses one or more parameters of the aerosol. The output of each of the sensor is fed to the data logging unit 56. After the aerosols have been exhausted, the sensor arrays 54i - 54N are used to sense parameters of the exhausted aerosols. Data from the sensor arrays 54i - 54N are also fed to the data logging unit 56. This creates an array of data that is characteristic of the sensory experience that a panellist experiences. Both the intensity of sensor response and the time when this response occurs is recorded.

[0135] Overall system control and data analysis is performed by the supervisory and control unit 60. The data logging unit 56, puff control unit 58, and supervisory and control unit 60 may be implemented, for example, as software routines running on one or more processors. Results of the analysis are displayed on the display 62. A user interface 64 is used to receive inputs from and / or output data to the user. The user interface may comprise, for example, a keyboard, mouse, touchscreen, display etc.

[0136] Figure 3 shows in more detail parts of the supervisory and control unit 60. Referring to Figure 3, the supervisory and control unit 60 comprises system control unit 66, data flattening unit 68, sensory response model 70 and reporting unit 72. The supervisory and control unit 60 may be implemented as software running on a processor. It will be appreciated that other hardware and / or software components may also be present, but for simplicity these are not shown. Furthermore, some of the units shown may be combined.

[0137] The system control unit 66 is used for overall control of puffing and data collection. The system control unit 66 receives inputs from the user interface 64, and instructs the puff control unit 58 to perform the appropriate puff routines. The system control unit 66 may also control other system variables, such as chamber volumes, valve settings and / or sensor settings.

[0138] The data flattening unit 68 receives sensor data from the data logging unit 56, and flattens the sensor data to a set of features for each puff of the product under test. The flattened sensor data are provided to the sensory response model 70.

[0139] The sensory response model 70 is a statistical model that relates sensor data to sensory profiles. The sensory response model 70 is produced in advance by mapping sensory panel data to sensor data from the sensor array. To achieve this, the sensory response model 70 is trained on training data using machine learning. This is preferably performed using a set of “standard” products.

[0140] Sensory panel responses to these products are mapped to sensory data obtained using the same products under the same conditions.

[0141] In one embodiment, the sensory response model 70 is a tree model, such as decision tree, random forest or a gradient boosted tree model such as XGBoost. A decision tree is a decision support hierarchical model that uses a tree-like model of decisions and their possible consequences. Random forest is a machine learning algorithm that combines the output of multiple decision trees to reach a single result. Gradient boosting is a machine learning technique based on boosting in a functional space. It gives a prediction model in the form of a plurality of weak prediction models, which are typically simple decision trees. Such tree-based models are known in the art and therefore not described further.

[0142] In alternative embodiments, the sensory response model 70 may use other machine learning approaches, such as artificial neural networks (ANNs), supportvector machines (SVMs), regression analysis, Bayesian networks, Gaussian processes, etc. instead or as well. Furthermore, in some embodiments, it may be possible for the sensory response model to be in the form of one or more mathematical formula such as a polynomial equation provided by linear mixed effect model.

[0143] If desired, a plurality of sensory response models could be provided, each corresponding to a particular type of product and / or set of test variables. In this case, the system control unit 66 may select the appropriate sensory response model to be used, for example, based on the test conditions or user input.

[0144] In operation, the data flattening unit 68 receives sensor data from the data logging unit 56. The sensor data may be data from any of the sensors described above. For example, in one embodiment the sensor data may comprise at least one of: aerosol density / obscuration; particle size distribution; chemical sensor data (organics / volatiles); aerosol residual humidity (RH); aerosol temperature, for example, at two or more places; and pressure drop. The sensor data comprises samples from the various sensors taken at different points in time (i.e. in the intensity and time domains). The data flattening unit 68 flattens the sensor data to a set of features (synthetic information such as mean, standard deviation, linearity, entropy, curvature and so forth) for each puff of the product under test. The flattened data are provided to the sensory response model 70.

[0145] The sensory response model 70 receives as an input the flattened sensor data, and uses the sensor data to predict panellist responses to the product under test. The output of the sensory response model 70 is a set of sensory descriptors, which describe how a panellist is predicted to respond to the product under test. For example, in one embodiment the output of the sensory response model 70 may comprise sensory descriptors such as: Impact; Irritation in Throat; Tingling in Mouth; Mouthful of Aerosol; Heat Sensation; Mouth Cooling; Draw Resistance; and / or Mouth Drying. The sensory descriptors are provided to the reporting unit 72.

[0146] The reporting unit 72 receives the set of sensory descriptors, and arranges it into a form in which it can be presented to the human operator. The sensory descriptors are then presented to the user on the display 62. The sensory descriptors can be presented to the user in a number of different ways, for example, as a series of heat maps, spider graphs, pseudo videos of data change, etc. The sensory descriptors may also be provided to a communications unit 74, for communication to another device and / or stored for later retrieval.

[0147] Figure 4 shows how sensory descriptors may be presented to the user in one exemplary embodiment. In this example, the sensory descriptors are presented as a spider graph. The spider graph has five axes, with each axis corresponding to a specific sensory descriptor (in this case, “mouthful”, “throat irritation”, “impact”, “heat sensation” and “draw resistance”). Each sensory descriptor has a score of between zero and five. The results from three different puffs are shown in Figure 4. This can allow an immediate visual comparison of different puffs and different products to be made.

[0148] It will be appreciated that the visualisation shown in Figure 4 is given by way of example only, and the data may be presented in any appropriate manner, such as tables, bar graphs, heat maps, videos etc. Furthermore, other sensory descriptors may be shown as well or instead.

[0149] Figure 5 shows steps carried out by a sensory testing apparatus in order to test a product in one embodiment. Referring to Figure 5, testing begins in step 100.

[0150] The testing may be initiated, for example, by the user inputting a command to the system control unit 66 via the user interface 64. Alternatively, testing may be initiated automatically, or on receipt of a command from another device.

[0151] In step 102, the system control unit 66 selects profiles of the puffs that are to be taken during the test. The puffing profiles may include for example the volume of each puff, and the shape of each puff (change in volume with time). The puffing profiles may be selected, for example, based on the type of product under test or the type of sensory response to be predicated. The puffing profiles may be standardised volumes and shapes, or based on data collected from human topography (e.g. human mouth or lungs or single puff) or anything else. The selected puffing profiles are output from the system control unit 66 to the puff control unit 58, for use in controlling the puff engines.

[0152] In step 104 the system control unit 66 selects the volumes of the analysis chambers 161 - 16N which are to be used during the test. The chamber volumes may be selected, for example, based on the type of product under test or the type of sensory response to be predicated. The chamber volumes may be standardised volumes or based on human topography or anything else. The selected volumes are output to the analysis chambers 161 - 16N and used to set the volumes of the chambers. In step 106 the system control unit 66 selects process variables which are to be used during the test. The process variables may include, for example, purge / no purge between puffs, hold time in the chambers, total number of puffs to be taken, and so forth. The process variables may be selected based on any appropriate criteria, such as type of product under test or the type of sensory response. The process variables are used to control the puff engines and / or other components such as shut valves during the puffing process.

[0153] In step 108 the puff engines are used to take an initial puff (before the product is activated). Then, in step 110, background clean air is established in the analysis chambers. This helps to ensure that residual aerosols and other possible contaminants are not present at the start of testing.

[0154] In step 112 the article or articles under test are lit or activated. In the case of a cigarette or similar smoking article, the article may be lit manually or automatically. For example, the article may be lit using an ignition system such as that disclosed in WO 2006 / 056747, the subject matter of which is incorporated herein by reference, or in any other appropriate manner. In the case of a heated tobacco product, e-cigarette or vaping device, the article may be activated manually or automatically in accordance with the product design.

[0155] In step 114, the data logging unit 56 begins logging data from the sensors. In step 116 the puff engines are used to take a puff from the products under test. In step 118 it is determined whether the number of puffs taken NP is greater than or equal to the total number of puffs NT to be taken during the test. If the number of puffs taken is less than the total number of puffs to be taken, processing returns to step 116. If the number of puffs taken is greater than or equal to the total number of puffs to be taken, processing proceeds to step 120. In step 120, the data logging unit 56 stops logging data from the sensors.

[0156] In step 122 the data flattening unit 68 is used to flatten the sensor data to a set of features for each puff of the product under test. In step 124 the sensory response of the product under test is modelled using the sensory response model 70. The input to the sensory response model is the flattened sensor data from the data flattening unit 68. The output of the sensory response model is one or more sensory descriptors, which predict panellist responses to the product under test. In step 126 the reporting unit 72 uses the sensory descriptors to produce a report of the sensory profile of the product in a standard form. In step 128 the report of the sensory profile is displayed on the display 62. The testing process then ends in step 130.

[0157] It will be appreciated that the steps shown in Figure 5 are given by way of example only and various changes can be made. For example, some of the steps may be omitted, and other steps which are not shown may be present. Furthermore, the order of the steps may be changed as appropriate.

[0158] The embodiments described above can provide a method of analytically measuring and describing the sensory and hepatic sensation associated with delivery of an aerosol to the human mouth, lungs and respiratory tract by means of a machine. In particular, a method of sensory correlation and measurement can be provided that deploys an array of multiple sensors that together form a pattern of response characteristic of the aerosol or smoke under test. Patterns gathered from sensing elements deployed in an aerosol path can be used to correlate to the findings of a sensory panel that uses similar or identical products to provide a correlation between analytical measurements and human sensory experience through pattern matching. The method may use a pattern of response from an array of sensors that changes in the time domain as well as in the intensity domain that is characteristic of a sensory experience of interest. The method may use sensors that are untargeted or non-specific but that in conjunction with other sensing elements can be used to create a pattern of response characteristic of a specific hepatic or sensory judgement of a panellist. The method may involve the use of sensors that produce an electronic signal.

[0159] The method may involve the use of standard products whose characteristics are known in terms of the pattern obtained from the test chamber to ensure calibration is maintained and adjusted to give consistent results. Correlation of measured responses in a test chamber with panellist judgements for control products can be used to build a set of patterns that can be used to produce the same or similar judgements as panellists when the product is no longer one of the selected control products (predictive capability). The method may use a chamber to hold a volume of aerosol for a set time to mimic the hold and release of a human in smoking or vaping. The method may use different volumes of test chamber that can be matched to the drawn volume of aerosol. The method may use a puff engine that can be programmed for any desired puff shape and volume to mimic human profiles both in inhale and exhale functions. The method may use multiple chambers and puff engines that are synchronised in operation, the different chambers having different sensors that have their data captured synchronously to form a single picture of the product aerosol. The method may sense on the inhale and / or exhale side of a puff engine. The method may use sensing elements on the exhaust or exhale part of the puff engine. The method may use sensing elements that provide information in the time domain. The method may include control in terms of speed and / or volume for the exhale function. The aerosols tested may be generated from tobacco products where the products may be heated, combusted or formed from the heating of a liquid or gel. The method may include the storage and transformation of discrete sensor data, in different formats, into patterns characteristic of the panellists’ sensations profile, and the use of “exemplar” patterns to match against captured data so to produce a sensory profile from the equipment comparable to a panel profile. Corresponding apparatus configured to carry out the methods may also be provided.

[0160] Embodiments of the invention have been described above by way of example only, and various modifications are possible. For example, while the embodiments described are for use with tobacco products, other embodiments may be for use with vaping products and / or with products that delivery aerosols to the human body for other purposes such as therapeutic applications. Features of one embodiment may be used with any of the other embodiments. Various other modifications and variations in detail will be apparent to the skilled person within the scope of the appended claims.

Claims

CLAIMS1 . Apparatus for analysing an aerosol releasing article, the apparatus comprising: means for drawing an aerosol from the article; a plurality of sensors arranged to sense characteristics of the aerosol to produce sensor data; means for predicting a sensory response from the sensor data using a model relating sensor data to at least one sensory response; and means for outputting the predicted sensory response to a user.

2. Apparatus according to claim 1 , wherein the model relates sensor data from a plurality of sensors to the or each sensory response.

3. Apparatus according to claim 1 or 2, wherein the aerosol releasing article is of a type which is used to deliver aerosol to the human body.

4. Apparatus according to any of the preceding claims, wherein the aerosol releasing article is a product of the tobacco or vaping industries.

5. Apparatus according to any of the preceding claims, wherein the predicted sensory response is a prediction of a response that will be produced in a human body when the aerosol is drawn into the human body.

6. Apparatus according to any of the preceding claims, wherein the predicted sensory response comprises a quantitative description of at least one sensation reported by a sensory panel.

7. Apparatus according to any of the preceding claims, wherein the sensory response comprises at least one of: Impact; Irritation in Throat; Tingling in Mouth; Mouthful of Aerosol; Heat Sensation; Mouth Cooling; Draw Resistance; and Mouth Drying.

8. Apparatus according to any of the preceding claims, wherein the plurality of sensors comprises a plurality of different type of sensor.

9. Apparatus according to any of the preceding claims, wherein the plurality of sensors is arranged to sense at least one of: a physical characteristic of the aerosol; a chemical characteristic of the aerosol; and a size characteristic of the aerosol.

10. Apparatus according to any of the preceding claims, wherein the characteristics of the aerosol comprise at least one of: an aerosol density and / or obscuration; aerosol particle size distribution; presence and / or concentration of one or more chemicals; aerosol temperature at one or more locations; aerosol humidity; and pressure and / or pressure drop.11 . Apparatus according to any of the preceding claims, wherein the plurality of sensors comprises at least one of: an aerosol density and / or obscuration sensor; a particle size distribution sensor; an electrochemical sensor; a temperature sensor; a humidity sensor; and a pressure sensor.

12. Apparatus according to any of the preceding claims, wherein at least one of the plurality of sensors is arranged to sense a characteristic of the aerosol in the intensity and time domains.

13. Apparatus according to any of the preceding claims, wherein the model comprises a machine learning algorithm, and wherein the machine learning algorithm is preferably trained in advance using sensor data and reported sensations from a sensory panel.

14. Apparatus according to any of the preceding claims, wherein the model comprises at least a tree-based model and / or a linear mixed effect model.

15. Apparatus according to any of the preceding claims, wherein the apparatus comprises means for flattening the sensor data to obtain a set of features for each sensor, and the model is arranged to predict at least one sensory response from the flattened sensor data.

16. Apparatus according to any of the preceding claims, wherein the means for drawing an aerosol from the product comprises a puff engine.

17. Apparatus according to any of the preceding claims, further comprising a chamber, wherein the means for drawing an aerosol from the article is arranged to draw the aerosol into the chamber, and at least one of the plurality of sensors is arranged to sense a characteristic of the aerosol in the chamber.

18. Apparatus according to claim 17, wherein the volume of the chamber is adjustable.

19. Apparatus according to claim 17 or 18, the apparatus comprising a plurality of chambers and / or a plurality of puff engines.

20. Apparatus according to any of the preceding claims, wherein the apparatus is arranged to draw an aerosol from a plurality of aerosol releasing articles, and the plurality of sensors are arranged to sense at least one characteristic of the aerosol from each of the plurality of articles.21 . Apparatus according to any of the preceding claims, wherein the plurality of sensors comprises at least one of: a plurality of temperature sensors arranged to sense a temperature of the aerosol in a plurality of different locations; a humidity sensor arranged to sense a humidity of the aerosol; a pressure sensor arranged to measure a pressure change during a puff; a weighing device arranged to determine a mass of particle deposition; an optical sensor arranged to measure particulate scattering of the aerosol; at least one electrochemical cell arranged to sense one or more organic and / or volatile compounds; at least one temperature sensor arranged to sense the temperature of the item under test; a particle sizing device; and a spectroscopic sensor.

22. Apparatus according to any of the preceding claims, wherein the plurality of sensors comprises at least one electrochemical cell arranged to sense a concentration of a least one of: a volatile organic compound; oxygen; ammonia; carbon monoxide; carbon dioxide and nicotine.

23. Apparatus according to any of the preceding claims, wherein the plurality of sensors comprises a plurality of electrochemical cells, and each of the electrochemical cells is arranged to sense a different substance from at least one of the other electrochemical cells.

24. A method of analysing an aerosol releasing article using an analysis apparatus, the method comprising: drawing an aerosol from the aerosol releasing article; sensing characteristics of the aerosol using a plurality of sensors to produce sensor data; predicting a sensory response from the sensor data using a model relating sensor data to at least one sensory response; and outputting the predicted sensory response to a user.

25. A method according to claim 24, further comprising the step of training the model using sensor data and reported sensations from a sensory panel.

Images