Classifying a consumable inserted into an aerosol-generating device using an ai model

Abstract

A computer-implemented method of determining a type of an aerosol-generating article or substrate couplable to an aerosol-generating device, the method comprising: receiving a sensing profile by at least one sensor, wherein the sensing profile comprises information about the evolution in time of a parameter indicative of and / or influenced by the composition of the aerosol-generating article or substrate, determining the type of the aerosol-generating article or substrate from the sensing profile based on evaluating the sensing profile with an artificial intelligence- based, AI-based, model.

Description

[0001] CLASSIFYING A CONSUMABLE INSERTED INTO AN AEROSOL-GENERATING DEVICE USING AN Al MODEL

[0002] The present disclosure relates to a computer-implemented method of determining a type of an aerosol-generating article or substrate couplable to an aerosol-generating device. The present disclosure also relates to an aerosol-generating device as well as an aerosol-generating system comprising an aerosol-generating device and at least one of a companion device configured to charge the aerosol-generating device with electrical energy, a computing device, for example a smartphone, and a server device. Further, the present disclosure relates to a computer program and a computer-readable medium.

[0003] Aerosol-forming or aerosol-generating devices are typically designed as handheld devices that can be used by a user for consuming or experiencing, for instance in one or more usage sessions, aerosol generated from an aerosol-forming substrate or an aerosol-forming article, for example by heating. The aerosol-forming devices the present disclosure pertains to are mainly directed to the field of tobacco and tobacco-substitute products, as well as e-vapor devices, for example heated tobacco products (HTP), heat-not-burn devices, electronic cigarettes, e-vapor devices, and / or vaporisers. The aerosol-forming devices of the present disclosure may also pertain to other types of inhalers, dispensers, or atomizers, for example inhalers, dispensers, or atomizers for medical applications.

[0004] Typical aerosol-forming systems can be designed as one-part systems or devices including an aerosol-forming device that can be operated by a user to generate aerosol. Alternatively, aerosol-forming systems can be designed as two-part systems or devices comprising an aerosolforming device and a companion device for storing and / or charging the aerosol-forming device. In either design or configuration, the aerosol-forming system or device can be used by a user for consuming or inhaling, for instance in one or more usage sessions, aerosol generated based on heating an aerosol-forming article or substrate couplable to the aerosol-forming system. In the context of the present disclosure, an aerosol-forming device can refer to both a one-part device and a two-part device, unless explicitly specified otherwise.

[0005] The aerosol-forming article, also referred to as aerosol-generating article, can comprise an aerosol-generating or aerosol-forming substrate, such as a tobacco or nicotine-containing substrate. The aerosol-forming article may be configured in shape and size to be inserted at least partially into the aerosol-forming device or system. In conventional systems or devices, the aerosol-forming article is usually formed as a stick that can be at least partly inserted into a cavity or heating chamber of the aerosol-forming device for aerosol consumption. Insertion of the sticklike shaped aerosol-forming article into the cavity, however, can potentially damage the aerosolforming article, which may potentially affect an experience for a user, for example in terms of taste or homogeneity of the experience. Also, inserting the aerosol-forming article into and removing itfrom the cavity of conventional systems may, at least for some users or in certain scenarios, be cumbersome.

[0006] Exemplary aerosol-forming substrates can comprise solid substrate material, such as tobacco material or tobacco cast leaves (TCL) material. The substrate material can, for example, be assembled, often with other elements or components, to form a substantially stick-shaped aerosol-forming article. Such a stick or aerosol-forming article can be configured in shape and size to be inserted at least partially into the aerosol-forming device. The aerosol-forming device may comprise a heating element or heater device for heating the aerosol-forming article and / or the aerosol-forming substrate. The heating element or heater device may be part of the aerosolforming article and / or the aerosol-forming device. Alternatively or additionally, aerosol-forming substrates can comprise one or more liquids and / or solids, which can, for example, be supplied to the aerosol-forming device in the form of a cartridge or container. Corresponding exemplary aerosol-forming articles can, for example, comprise a cartridge containing or fillable with the liquid and / or solid substrate, which can be vaporized during aerosol consumption by the user based on heating the substrate and / or liquid. Usually, such cartridge or container can be coupled to, attached to or at least partially inserted into the aerosol-forming device. Alternatively, the cartridge may be fixedly mounted to the aerosol-forming device and refilled by inserting liquid and / or solid into the cartridge. The aerosol generated from the aerosol-forming substrate or article may comprise or include one or more of nicotine, aroma, sugar, moisturising agent, botanicals, preservative, flavouring, for example cocoa, liquorice, menthol and lactic acid or other additives. The aerosol generated from the aerosol-forming substrate or article may additionally or alternatively comprise one or more pharmaceutical agents or drugs and may include one or more adjuvants.

[0007] For generating the aerosol during use or consumption, heat can be supplied by a heating element, heater device or heat source to heat at least a portion or part of the aerosol-forming substrate. The heating element, heater device or heat source can be arranged in the handheld device or a handheld part of the aerosol-forming device. Alternatively or additionally, at least a part of or the entire heating element or heater device or heat source can be fixedly associated with or arranged within an aerosol-forming article, for instance in the form of a stick or cartridge, which can be attached to and / or powered by the handheld device or handheld part of the aerosolforming device.

[0008] Exemplary heating elements or heater devices can be based on one or more of resistive heating, inductive heating and microwave heating using electrical energy supplied via, drawn from or stored in an energy storage or battery of the aerosol-forming device. As used herein, a battery of the aerosol-forming device can generally refer to an energy storage of the aerosol-forming device configured to store electrical energy. Accordingly, the term energy storage can include oneor more batteries, one or more capacitors, one or more accumulators or other types of energy storage. Also, any reference to a battery herein can include a plurality of batteries.

[0009] Typically, aerosol-forming devices comprise an energy storage, for example a battery, providing the electrical energy needed to operate the aerosol-forming device and especially for heating the aerosol-forming substrate and / or article, for example to generate aerosol in one or more usage sessions using one or more aerosol-forming articles. The battery may, for example, be a lithium-ion battery.

[0010] As used herein, a usage session may refer to a period of time, during which a user may use the device to generate, consume, experience or inhale aerosol using the aerosol-forming device. Therein, a usage session may be finite. In other words, a usage session may have a start, an end and a duration. The duration of the usage session as measured by time may be influenced by use during the usage session. The duration of the usage session may have a maximum duration determined by a maximum time from the start of the usage session. The duration of the usage session may be less than the maximum time if one or more monitored parameters reaches a predetermined threshold before the maximum time from the start of the usage session. By way of example, the one or more monitored parameters may comprise one or more of: i) a cumulative puff count of a series of puffs drawn by a user since the start of the usage session, and ii) a cumulative volume of aerosol evolved from the aerosol-forming substrate since the start of the usage session.

[0011] Aerosol-generating articles or substrates are typically mass-produced and need to fulfil strict quality standards to ensure that health risks for the user are at a minimum. At the same time, being one-use consumables, aerosol-generating articles or substrates need to be as cost-effective and environmentally friendly as possible. Therefore, they are typically analogue in design, meaning that they typically do not contain any electronic sensors or features which communicate with the device to provide information and data about them. The reason such a design is generally preferred may be to reduce any risk of contamination or toxicology issues as well as keeping complexity and cost low. Examples of specifically marking aerosol-generating articles or substrates for identification may include special inks, printed ID codes, embossing, and RFID or EEPROM memory chips, among others. Apart from making the consumables more expensive, there may be a risk that the additional substances used could release toxic gases or fumes during use.

[0012] However, when the aerosol-generating device cannot identify what type of consumable is used, the device will typically work with and heat any consumable, regardless of its origin or constituents. Therefore, the manufacturer of the aerosol-generating device may not be able to protect the user from using lower quality consumables. Additionally, since in this case, the device is unaware of exactly which consumable is inserted, performance may not be optimized sincesettings such as heating profiles to perform optimally with a particular consumable specification cannot be chosen or adjusted.

[0013] It may therefore be desirable to provide for a possibility of recognizing the type of aerosolgenerating article or substrate being used in an aerosol-generating device. Specifically, it may be desirable to determine whether or not the aerosol-generating article or substrate fulfils a specification or whether unwanted substances are present.

[0014] These advantages may be achieved by the features described herein.

[0015] According to an aspect of the present invention, there is provided a computer-implemented method of determining a type of an aerosol-generating article or substrate inserted into or insertable into or couplable to an aerosol-generating device, the method comprising: receiving a sensing profile by at least one sensor, wherein the sensing profile comprises information about the evolution in time of a parameter indicative of and / or influenced by the composition of the aerosol-generating article or substrate, and determining the type of the aerosol-generating article or substrate from the sensing profile based on evaluating the sensing profile with an artificial intelligence-based, Al-based, model.

[0016] In other words, the at least one sensor, particularly a sensing part of the sensor, may be configured to react to the presence and / or concentration of one or more constituents of the aerosol-generating article or substrate. These may be present in the aerosol-generating article or substrate or may be released from the aerosol-generating article or substrate, for example during heating, for example as one or more gases or volatile organic compounds (VOC). The response of the at least one sensor to the presence and / or concentration of one or more of these constituents may be recorded in the sensing profile. For example, the sensing profile may comprise and / or represent the evolution in time of a measurable and / or measured parameter of the sensor, particularly a sensing part of the sensor. For instance, the sensing profile may comprise two or more or a plurality of measured values of the parameter at different points in time. The parameter may be determined or measured at discrete intervals or continually.

[0017] It may also be provided that more than one sensor, for example two, three, four, five or more sensors, each provide one sensing profile, and that all sensing profiles are used for determining the type of the aerosol-generating article or substrate as explained herein. At least part of the sensors, for example all sensors, may be similar sensors, for example obtaining the sensing profiles by a similar interaction with the aerosol-generating article or substrate. The sensing profiles obtained in this way may still be different from each other as a result of, for example, different positions of the sensors, for example in relation to the aerosol-generating article or substrate. Alternatively, at least part of the sensors, for example all sensors, may be configured to obtain sensing profiles based on different interactions with the aerosol-generatingarticle or substrate. Also, some sensors may be similar and some sensors may be different. Any of the sensors may be of any suitable type, for example of any type as mentioned herein.

[0018] An Al-based model may be a computational system or algorithm designed to simulate human-like intelligence by learning patterns from data and making predictions or decisions based on that learning. These models may leverage techniques from machine learning and / or deep learning to process and / or analyze complex datasets. Thus, the Al-based model may be or may comprise a machine learning model and / or algorithm. In the present disclosure, the Al-based model is used for classifying data, specifically for classifying or determining the type of the aerosol-generating article or substrate from the sensing profile. To this end, an Al-based model may work as follows: The model may receive raw or preprocessed data, i.e. the input data, for example in the form of numerical, categorical, or textual values. Preprocessed data may be called features or feature information, as explained in more detail below. Then the model may be trained, meaning that the Al-based model may learn to identify patterns in the data by analyzing labeled examples (e.g., input data with known outcomes) using a learning algorithm. Once trained, the model may classify new, unseen data by assigning it to one of the predefined categories or classes. The model’s performance may be assessed or evaluated using metrics like accuracy, area under the curve (AUG), precision, recall, or F1-score, and optimized to improve its classification or predictive capability. In the present disclosure, the Al-based model may comprise one or more learning algorithms, which may, for example, be used on the data one after the other or which may be used in conjunction to enhance performance. The Al-based model may comprise or perform several computational steps, for example steps in which data from the sensing profile is modified, used as basis for deriving further information, clustered, classified, merged, smoothed, filtered, aggregated, split and / or combined. Any one or any combination of these steps may utilize a learning algorithm as described herein. Such Al-based models which may implement learning techniques are especially useful in the present application because, as mentioned above, the aerosol-generating articles or substrates are not specifically marked for identification. Thus, the determination of the type of aerosol-generating article or substrate relies on sensor data generated in response to constituents of the aerosol-generating article or substrate. However, sensed data may slightly vary depending on tolerances of the aerosol-generating article or substrate or environmental conditions or the type of interaction with the sensor along with the sensor’s internal variations. Reliably determining the correct type of aerosol-generating article or substrate from this varying input data may be achieved by the Al-based model as explained herein.

[0019] The method may comprise classifying the sensing profile into one of a plurality of classes, each class being representative of and / or associated with a particular type of aerosol-forming article or substrate. The classes or types of aerosol-generating articles or substrates may refer todifferent compositions, for example in terms of the aerosol-forming material or material mixture and / or additives. For example, each class or type of aerosol-generating article or substrate may be identical or congruent with a commercially available type or brand of aerosol-generating article or substrate, for example sold by the manufacturer of the aerosol-generating device or a third party. As an example, different classes or types of aerosol-generating articles or substrates may refer to the different variants of TEREA sticks available for the IQOS devices from Philip Morris Products S.A., such as Russet, Bronze, Sienna, Teak, Yellow Green, Amber, Yellow, Warm Fuse, and Turquoise. However, other types of aerosol-generating articles or substrates may be determined and / or classified by the method according to the present disclosure. The classes or types of aerosol-generating articles or substrates may have the same shape and / or outer contour and / or form factor and / or weight. They may be free of special markings identifying their class or type. They may be both heat-not-burn (HNB) or vaping articles or another kind of consumable.

[0020] The Al-based model may be configured and / or trained to classify the sensing profile into one of the plurality of classes. The Al-based model may, in other words, determine that the aerosol-generating article or substrate in question is part of a known class of aerosol-generating article or substrate and / or may determine to which class the aerosol-generating article or substrate in question belongs. The classification may be based on the sensing profile, meaning on the data constituting the sensing profile, or data derived from the sensing profile. The Al-based model may be configured to output, for example as a result, the determined class of aerosolgenerating article or substrate. In other words, the Al-based model may be configured to provide, based on the sensing profile, a classification result as output data, the classification result being indicative of the type of aerosol-generating article or substrate. In this way, the type of an aerosolgenerating article or substrate being in use in the aerosol-generating device may be determined and a reaction to this determined type may be initiated, as explained in more detail below.

[0021] The data comprised in or constituting the sensing profile may be used directly by the Al-based model. However, the sensing profile, i.e. the data comprised in or constituting the sensing profile, may be preprocessed before being used by the Al-based model. Preprocessing the data may lead to more reliable classification results. For example, preprocessing may be performed by extracting feature information from the sensing profile. Thus, the method may comprise extracting feature information from the sensing profile. Feature information may refer to measurable attributes or characteristics which may be extracted from the sensing profile as raw data. The feature information may then be used as input to train the Al-based model or to infer or determine a class or type of the aerosol-generating article or substrate. The features may represent aspects of the data that are relevant to the specific problem the model is designed to solve, in the present case classification of the type of aerosol-generating article or substrate. Therefore, the method may comprise generating input data for the Al-based model based on theextracted feature information of the sensing profile. The Al-based model may be configured and / or trained to evaluate or analyze the sensing profile in terms of the feature information. During training, the extracted features may be used to build a representative dataset that the model learns from to classify correctly. During inference, i.e. classification or determining the type of aerosol-generating article or substrate, the same feature extraction process may be applied to new input data, meaning the same features are extracted. These extracted features or feature information may then be used as inputs to the trained model, enabling it to perform classification or prediction. Thus, feature extraction may be used in both training and inference.

[0022] Feature information may be derived from raw data, i.e. the sensing profile(s), through preprocessing, which may include a series of transformations and techniques to make the raw data suitable for analysis. This process may ensure that the features capture meaningful patterns or structures while reducing noise and irrelevant details. Features or feature information used should be informative and contribute to the predictive or analytical task. They should accurately capture the underlying properties of the raw data and should be clean, consistent, and free of errors or biases. Feature information may be derived from the sensing profile by mathematical and / or statistical analysis and / or signal processing methods. Thus, the feature information may comprise mathematical, statistical, time-domain, and / or frequency-domain features. For example, the feature information may comprise one or more of mathematical feature information, one or more mathematical features, one or more derivatives, one or more time derivatives, one or more statistical features, statistical feature information, and a standard deviation of the sensing profile. Additionally or alternatively, the feature information may comprise one or more of a time series complexity, an autocorrelation, a maximum derivative, a number of peaks, a permutation entropy, a forward Sharpe ratio, a backward Sharpe ratio, a minimum derivative, a logarithmic sum, an integral, a difference measure, a peak-to-peak value, a root mean square, a mean slope, a kurtosis, a skewness, a median, a mean absolute deviation, an interquartile range, a zerocrossing rate, an entropy, an energy, a wavelet energy, and an instantaneous frequency. The feature information used may comprise any one or any combination of two or more of the listed items. For example, in one embodiment, the feature information extracted from a sensing profile may at least comprise a time series complexity, an autocorrelation, and a maximum derivative. These features have proven very suitable for the present use case. When more than one sensing profile is used, i.e. when more than one sensor is used, each sensing profile may be preprocessed using the same set of feature information or at least a part of the sensing profiles or all of the sensing profiles may be preprocessed using a different set of feature information.

[0023] Additionally or alternatively, the feature information may comprise a difference measure in comparison to one or more reference profiles. Such measures may be used to capture how an individual data point or group of data points relates to a predefined reference. This approach mayprovide valuable insights into patterns or anomalies and enhance the predictive power of the Al-based model. The reference profiles may have been previously determined using known types of aerosol-generating articles or substrates. Therefore, the reference profiles may serve as examples, i.e. training data, and references for determining difference measures during inference. Suitable difference measures may include deviation from mean or median, distance metrics, similarity measures or others.

[0024] For example, a matrix may be constructed in which the columns may correspond to features and the rows may correspond to different aerosol-generating articles or substrates. The last column may contain known target features which may represent the outcomes to be predicted using the Al-based model. This may be the variable to be estimated or classified based on the available data. The structure of the matrix may be as shown in Table 1 below. While in the example, only one row per type is shown, the actual matrix may contain a plurality of rows per type.

[0025] Table 1: Exemplary matrix containing feature information and target classes

[0026]

[0027] The Al-based model may be trained on a training data set comprising sensing profiles of known types of aerosol-generating articles or substrates, for example as shown in Table 1. The data contained in the matrix may be the training and / or reference data used for the Al-based model. The training phase may consist in associating the value that the consumable or aerosolgenerating article or substrate has per each variable / feature with the specific type and calculating how much it is distant from the target (Terea, Levia, Menthol etc.). The so trained model may be used to predict new consumables, or, more in general, new elements that belong to a certain class.

[0028] In general, the complete sensing profile may be used for determining the type of the aerosolgenerating article or substrate. However, the sensing profile may also only comprise and / or represent a relevant portion of all of the data gathered by the specific sensor. For example, the sensor may already collect data before an aerosol-generating article or substrate is inserted into the aerosol-generating device. This data may not be relevant for the determination of the type of the aerosol-generating article or substrate and may therefore be disregarded and / or discarded. Similarly, the sensor may still collect data after a usage session using the aerosol-generatingarticle or substrate has begun. This data may also be irrelevant and may therefore be disregarded and / or discarded. Therefore, the sensing profile may start or begin when the aerosol-generating article or substrate is inserted or has been inserted into the aerosol-generating device and / or may end when a usage session is started. The data gathered between these points in time may be regarded as the sensing profile. Therefore, feature information may be extracted from the whole sensing profile. Alternatively, feature information may be extracted from one or more parts of the sensing profile, particularly a first part of the sensing profile at the beginning of the sensing profile and / or a second part of the sensing profile at the end of the sensing profile. Feature information may also be extracted from a part of the sensing profile between the beginning and the end of the sensing profile. For example, if the sensing profile is parted in three parts of the same length, the feature information may be extracted from only one of the first, the second, and the third part, or feature information may be extracted from any combination of these parts. Depending on the specific sensor and the specific data gathered by this sensor, different parts of the sensing profile may be of different importance or value for the determination of the type of the aerosol-generating article or substrate. Therefore, a focus may be put on the more relevant parts of the data.

[0029] In an embodiment, the method may comprise extracting feature information from the sensing profile, wherein at least 5 or at least 10 or at least 15 or at least 20 or at least 30 or at least 40 or at least 50 features are derived or extracted from the sensing profile as feature information. This may mean that at least 5 or at least 10 or at least 15 or at least 20 or at least 30 or at least 40 or at least 50 different pieces of feature information may be extracted or derived from the sensing profile. Larger numbers of feature information may represent more details about the original data. Combining at least 5 or more pieces of feature information may therefore make the original data, or at least its pertinent content, available for analysis by the Al based model. Such data preprocessing may make the data accessible for the Al based model in the first place. Additionally, the selection of the amount and nature of the specific features or feature information extracted or derived from the sensing profile tailors the method to the use case at hand, particularly the determination of the type of the aerosol-generating article or substrate.

[0030] In general, as many features as possible may be derived from the sensing profile. For example, at least 5 or at least 10 or at least 15 or at least 20 or at least 30 or at least 40 or at least 50 features may be derived from the sensing profile as feature information. However, large numbers of features may make the further analysis difficult, both by necessitating higher computational power and by possibly introducing errors, for example by overfitting. Therefore, the method may comprise reducing the dimensionality of the feature information. Dimensionality reduction may refer to the process of reducing the number of features or feature information while retaining as much relevant information as possible. This simplification may make the data easier to work with and may improve the performance of the Al-based model. For example, the featureinformation may be combined into a linear combination, for example wherein each feature or item of feature information or feature information is multiplied with a weighting factor or coefficient. The Al-based model may then use the linear combination as input data.

[0031] For example, the method may comprise extracting feature information from the sensing profile, for example as explained above. The method may further comprise reducing the dimensionality of the feature information, and generating input data for the Al-based model based on the extracted feature information of the sensing profile. These steps may be performed before the type of the aerosol-generating article or substrate is determined using the Al based model. The type of the aerosol-generating article or substrate may then be determined from the reduced dimensionality feature information. This data preprocessing is suitable for Al based models as explained herein. It was discovered that using the reduced dimensionality feature information to determine the type of the aerosol-generating article or substrate increases model performance both in terms of accuracy and power consumption. This may be based on an improved signal to noise ratio in the reduced dimensionality data as well as the reduction in memory and CPU needs for processing the reduced dimensionality data.

[0032] The feature information extracted from the sensing profile may at least comprise one or more of a time series complexity, an autocorrelation, and a maximum derivative. Particularly, the feature information extracted from the sensing profile may comprise at least two or all three of a time series complexity, an autocorrelation, and a maximum derivative. The mentioned feature information has proven particularly reliable and has provided the most robust results in testing. By extracting this particular feature information, the input data is made suitable for the analysis and the processing by the Al based model according to the present disclosure. Also, this particular feature information may be particularly useful for the present purpose of determining the type of the aerosol-generating article or substrate.

[0033] Any suitable technique or process may be used to create linear combinations from the feature information for dimensionality reduction. One particularly suitable method is linear discriminant analysis (LDA). LDA, also known as normal discriminant analysis (NDA) or discriminant function analysis (DFA), follows a generative model framework. This may mean that LDA algorithms model the data distribution for each class and use Bayes' theorem to classify new data points. Bayes may calculate conditional probabilities, i.e. the probability of an event given some other event has occurred. LDA algorithms may make predictions by using Bayes to calculate the probability of whether an input data set will belong to a particular output. LDA may identify one or more linear combinations of features that separate or characterize two or more classes of objects or events. To this end, LDA may project data with two or more dimensions, i.e. features, into one dimension, i.e. one linear combination, so that it can be more easily classified. The technique is, therefore, referred to as dimensionality reduction. This versatility may ensurethat LDA can be used for multi-class data classification problems, unlike logistic regression, which is limited to binary classification. LDA is thus often applied to enhance the operation of other learning classification algorithms such as decision tree, random forest, or support vector machines (SVM).

[0034] As mentioned, the method may comprise extracting feature information from the sensing profile and the feature information may be combined into a linear combination. The linear combinations may comprise the feature information, each item of feature information weighted by a weighting factor or coefficient. By adjusting the weighting factors, i.e. changing their numerical values, the linear combinations may be optimized, for example for optimal differentiation between the classes or types of aerosol-generating articles or substrates. Any suitable algorithm or Al-based model may be used to optimize the linear combinations in this way. For example, LDA and / or SVM may be used to optimize the linear combinations. Thus, the method may comprise optimizing one or more linear combinations of the feature information for differentiating between different types of the aerosol-generating article or substrate, for example by maximizing a distance between mean values associated with the different types, for example using LDA, and / or maximizing a margin between the values associated with the different types, for example using SVM.

[0035] Additionally or alternatively, the feature information may be grouped into clusters, each cluster representing a distinct type of aerosol-generating article or substrate. Any suitable clustering algorithm may be used. For example, the feature information may be grouped into clusters based on similarity, for example using a k-means model. A k-means model is an unsupervised machine learning algorithm used for clustering. It may partition a dataset into a number of k distinct clusters based on feature similarity, where each data point belongs to the cluster with the nearest mean. In this way, for example, the training data set may be grouped into a number of clusters corresponding to the number of different types of aerosol-generating articles or substrates covered by or comprised in the training data set. Clustering may simplify and / or increase the accuracy of further data analysis, for example using LDA or SVM.

[0036] The Al-based model may correlate the sensing profile, i.e. the dataset constituting the sensing profile, or the feature information to a class, i.e. a type of aerosol-generating article or substrate. To this end, the Al-based model may determine a statistical relationship or association between the sensing profile and / or its feature information and the class labels corresponding to the types of aerosol-generating articles or substrates. Specifically, changes or patterns in the sensing profile and / or its feature information, for example the numerical values representing the sensing profile and / or its feature information, may be used to predict or distinguish between the classes. Thus, determining the type of the aerosol-generating article or substrate from the sensing profile may comprise correlating the sensing profile, for example through the linear combinationsof feature information of the sensing profile, to the type of the aerosol-generating article or substrate. By virtue of the learning algorithms or models employed, inaccuracies introduced by naturally occurring variations, for example due to environmental influences or tolerances in the composition of the aerosol-generating article or substrate, may be minimized and reliable classification may be achieved.

[0037] In principle, any suitable Al-based model and / or learning model and / or learning algorithm may be used to classify the aerosol-generating article or substrate. For example, the Al-based model may be or may comprise a support vector machine, SVM, an XGBoost classifier, a linear discriminant analysis, LDA, classifier, a decision tree classifier, a random forest classifier, or a k-nearest neighbors, k-NN, algorithm classifier. These models and / or algorithms have been found especially suitable for the present invention. In fact, the inventors evaluated all of these models and / or algorithms in terms of the area under the curve (AUG) and accuracy. The AUG is a performance metric for classification models, representing the area under a receiver operating characteristic (ROC) curve. The ROC curve is the plot of the true positive rate (TPR) against the false positive rate (FPR) at each threshold setting. The AUC quantifies the model's ability to distinguish between classes, with values ranging from 0 to 1 , wherein 1.0 would be representative of a perfect classifier, 0.5 would be representative of no discrimination (i.e. random guessing), and values below 0.5 would be representative of worse than random guessing. In essence, the AUC reflects the likelihood that the model ranks a randomly chosen positive instance higher than a randomly chosen negative instance. Accuracy is the proportion of correct predictions made by a model out of the total number of predictions. It also ranges between 1 and 0, wherein 1 would be representative of all predictions or classifications being correct and 0 would be representative of all predictions or classifications being incorrect. An overview of the performance, as measured by AUC and accuracy, of the evaluated models is given in Table 2.

[0038] Table 2: AUC and accuracy of Al-based models comprising LDA and another learning model as indicated in the table (the LDA-row comprises LDA only)

[0039]

[0040] As can be taken from Table 2, the combination of LDA and SVM exhibited high performance in terms of AUC and accuracy. Another possibility of measuring the performance may be a correlation matrix. A correlation matrix may be a table with the correlation coefficients for differentvariables. The matrix may show how all the possible pairs of values in a table are related to each other. It may be a powerful tool for summarizing a large data set and finding and showing patterns in the data. It may be visualized as a table, as in Table 3 below, with each variable listed in both the rows and the columns and the correlation coefficient between each pair of variables written in each cell. The correlation coefficient ranges from -1 to +1, where -1 means a perfect negative correlation, +1 means a perfect positive correlation, and 0 means there is no correlation between the variables. A correlation of 0.5 or more may be considered moderate, wherein a correlation of more than 0.75 may be considered strong. As can be seen from Table 3 for an Al-based model comprising LDA and SVM, all of the types of aerosol-generating article or substrate have at least a moderate correlation with themselves, while all but two even have a strong correlation. The correlation between different types of aerosol-generating articles or substrates is universally low or nonexistent.

[0041] Table 3: Correlation matrix of an Al-based model comprising LDA and SVM ; the total AUC is 92.3 % and the total accuracy is 86.2 %

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] e a h c i d b f j g predicted type

[0051] One specific exemplary embodiment of the present invention therefore comprises both LDA and SVM in the Al-based model or one of the other combinations as given in Table 2. However, any other model may also be employed.

[0052] For example, the Al-based model may be or may comprise an ensemble learning model for making the final classification decision, for example a random forest model. Ensemble learning models, such as random forests, may be accurate, robust and able to handle complex relationships. These properties may be taken advantage of when using an ensemble learning model as the final decision-maker in the Al-based model. This may be especially useful in combination with preprocessing the data, i.e. the sensing profiles, as described herein, for example by LDA and / or SVM. Combining preprocessing models like LDA or SVM with anensemble learning model, like a random forest, for final decision-making may therefore leverage the strengths of both approaches, leading to improved classification performance and robustness. The reduced data dimensionality and the improved class separation provided by LDA and / or SVM may help the ensemble learning model to achieve the best possible performance with low computational needs.

[0053] As mentioned previously, there may be a number of types of aerosol-generating articles or substrates commercially available by the manufacturer of the aerosol-generating device and / or an associated provider. However, in principle, the aerosol-generating device may also be used with aerosol-generating articles or substrates obtained from a third party. However, it may not be guaranteed that these third-party consumables may be safe to use in terms of their composition and / or compatibility with the device. To protect the user from potential harm and / or to protect the user from damage to the devices, it may be desirable to be able to determine whether the aerosolgenerating article or substrate used in the device has been provided by the manufacturer and therefore is safe or whether the aerosol-generating article or substrate used in the device has been provided by a third party and may therefore potentially be problematic or dangerous. It may therefore be provided that determining the type of the aerosol-generating article or substrate comprises determining whether the aerosol-generating article or substrate is an authentic aerosol-generating article or substrate provided by the manufacturer of the aerosol-generating device or whether the aerosol-generating article or substrate is provided by a third party. For example, it may be determined which one of the available aerosol-generating articles or substrates, i.e. which specific type, provided by the manufacturer of the aerosol-generating device is being used in the aerosol-generating device at that moment. As aerosol-generating articles or substrates provided by third parties and which may, at least in principle, also be compatible with use in the aerosol-generating device may be known, it may also be determined which one of the available aerosol-generating articles or substrates, i.e. which specific type, provided by a third party is being used in the aerosol-generating device at that moment. Additionally or alternatively, determining the type of the aerosol-generating article or substrate may comprise determining whether the aerosol-generating article or substrate comprises potentially toxic or toxic constituents, for example gases released from the aerosol-generating article or substrate during heating, in concentrations above a predetermined threshold.

[0054] In principle, simply determining the type of the aerosol-generating articles or substrates used in the aerosol-generating device is advantageous on its own. For example, the manufacturer of the aerosol-generating device may collect this information over a period of time and may therefore learn the preferences of the user. This may comprise information about which types of aerosol-generating articles or substrates are used at what times of day and / or what times of year and / or which days of the week, for example. Data about the popularity of third party or counterfeitconsumables being used in the aerosol-generating device may also be desirable. However, apart from collecting information for its own merit, the information about which specific type of aerosolgenerating article or substrate is being used may also be directly employed in controlling the aerosol-generating device. For example, the method may comprise controlling the heating device of the aerosol-generating device depending upon the determined type of the aerosol-generating article or substrate. For instance, the method may comprise controlling the heating device to provide or apply a specific heating profile depending upon the determined type of the aerosolgenerating article or substrate. The heating profile may be adjusted or configured specifically for the determined type of aerosol-generating article or substrate. For example, some types of aerosol-generating article or substrate may need different temperatures, for example peak temperatures, or temperature profiles during the usage session to provide the highest quality aerosol. Such specific necessities and therefore heating profiles may be implemented for each type of aerosol-generating article or substrate, or such specific heating profiles may only be used for some, for example at least one, type of aerosol-generating article or substrate. For example, different heating profiles may be provided for authentic aerosol-generating articles or substrates provided by the manufacturer of the aerosol-generating device and for aerosol-generating articles or substrates provided by a third party. As the quality and security standards of third parties may be unknown, the heating profiles for third party consumables may correspond to or use lower temperatures than the ones for authentic aerosol-generating articles or substrates.

[0055] As another possible reaction to the determined type of aerosol-generating article or substrate, the method may comprise allowing or disallowing generation of aerosol from the aerosol-generating article or substrate depending upon the determined type of the aerosolgenerating article or substrate. Allowing or disallowing the generation of aerosol may be implemented in the method and / or device by controlling the heating device of the aerosolgenerating device, for example by controlling a switch allowing or preventing the heating device to be supplied with electrical energy. For example, the method may comprise allowing generation of aerosol from authentic aerosol-generating articles or substrates. Conversely, the method may comprise disallowing generation of aerosol from non-authentic aerosol-generating articles or substrates from third parties. However, the method may comprise allowing generation of aerosol from non-authentic aerosol-generating articles or substrates. It may even be provided that the method may comprise allowing generation of aerosol from some types of non-authentic aerosolgenerating articles or substrates while disallowing generation of aerosol from other types of non-authentic aerosol-generating articles or substrates. Additionally or alternatively, the method may comprise disallowing generation of aerosol whenever the presence of a potentially toxic or toxic substance and / or gas, for example above a predetermined threshold, is detected. In this way, safety of the user may be ensured while simultaneously providing the highest quality aerosolpossible and even allowing the user to use non-authentic aerosol-generating articles or substrates if their quality is sufficiently high.

[0056] In general, the method according to the invention may be performed by the aerosolgenerating device itself. Alternatively, the method may also be performed by a device having a data connection to the aerosol-generating device, for example a companion device or a computing device, as will be explained in more detail below. Therefore, it may be enough to simply receive the sensing profile, for example from the aerosol-generating device, to be able to perform the method. However, the sensing profile may be determined, for example by a sensor of the aerosol-generating device. Therefore, receiving the sensing profile may include determining the sensing profile, for example by at least one sensor. In other words, receiving the sensing profile may or may not include the actual measurement of the data values constituting the sensing profile by a sensor. One or more sensors may be used to determine a sensing profile each and the method may comprise using all of the sensing profiles provided by all of the sensors to determine the type of the aerosol-generating article or substrate as explained herein.

[0057] For determining the sensing profile, any type of suitable parameter may be used as long as it may be measured by a sensor provided on the aerosol-generating device. For determining the type of the aerosol-generating article or substrate, a parameter which may change over time during the measurement may be particularly useful for analysis. For example, the value of the parameter may change with the temperature of the aerosol-generating article or substrate during heating. It may therefore be provided that the sensing profile is indicative of the evolution in time of the at least one parameter during heating of the aerosol-generating article or substrate.

[0058] In general, the determination of the type of the aerosol-generating article or substrate used may be performed before or during the usage session. This may mean that the time necessary for the determination may end, and therefore the result of the determination may be available, before or during the usage session. An ongoing usage session in which aerosol is generated from the aerosol-generating article or substrate may be aborted when the type of the aerosolgenerating article or substrate has been determined and indicates that this specific type is flagged for disallowing the generation of aerosol. Also, the heating profile may be adjusted during the usage session in response to the determination result. However, both for reasons of user experience and user security, it may be desirable that the determination of the type of the aerosolgenerating article or substrate is finished before a usage session is started. The sensing profile and / or the type of the aerosol-generating article or substrate may thus be determined before a usage session using the aerosol-generating article or substrate is started. The determination of the sensing profile and / or the type of the aerosol-generating article or substrate may therefore be performed during a preheating phase, in which the aerosol-generating article or substrate is heated from an ambient temperature to a maintenance temperature, particularly before anaerosolization temperature of the aerosol-generating article or substrate is reached. The ambient temperature may be the temperature at which the aerosol-generating article or substrate is inserted into the aerosol-generating device. Before a usage session is started, the aerosolgenerating article or substrate may be heated to the maintenance temperature, which may be a temperature at which the aerosol-generating article or substrate is held throughout the usage session. However, the maintenance temperature may be below an aerosolization temperature of the aerosol-generating article or substrate. The aerosolization temperature may be the temperature at which a vapor or an aerosol is released from the aerosol-generating article or substrate, i.e. at which fine solid particles, vapor or liquid droplets are released from the aerosolgenerating article or substrate and suspended in the surrounding air. There may be devices in which the aerosol is created from a vapor coming from the heated aerosol-generating article or substrate being mixed with air from the outside environment, for example air at ambient temperature, which then leads to the formation of droplets and therefore aerosol from the vapor. In this case, the aerosolization temperature may be the temperature at which vapor for creating the aerosol is released from the aerosol-generating article or substrate. After the start of a usage session, whenever a user puffs on the aerosol-generating device, the temperature of the aerosolgenerating article or substrate is increased from the maintenance temperature to the aerosolization temperature for the provision of aerosol to the user. After the puff ends, the temperature of the aerosol-generating article or substrate may decrease again, but is held at a minimum at the maintenance temperature until the next puff is detected or the usage session ends. Therefore, when the type of the aerosol-generating article or substrate is determined in the preheating phase in which the aerosol-generating article or substrate is heated from the ambient temperature to the maintenance temperature, no usage session has started yet. Simultaneously, the increase of temperature of the aerosol-generating article or substrate during preheating may influence the value of the measured parameter and may therefore improve the determination of the type of aerosol-generating article or substrate.

[0059] For example, the determination of the sensing profile and / or the type of the aerosolgenerating article or substrate may be performed within a maximum of 20 seconds after insertion of the aerosol-generating article or substrate into the aerosol-generating device, for example a heating chamber of the aerosol-generating device, for example within a maximum of 15 seconds or within a maximum of 10 seconds or within a maximum of 8 seconds or within a maximum of 6 seconds or within a maximum of 5 seconds after insertion of the aerosol-generating article or substrate into the aerosol-generating device, for example a heating chamber of the aerosolgenerating device. Typically, the preheating phase may last up to 20 seconds, so that these timespans ensure that the determination of the sensing profile and / or the type of the aerosolgenerating article or substrate is finished before the start of a usage session. The method maythen comprise implementing any adjustments depending on the type of the aerosol-generating article or substrate as determined, for example allowing or disallowing the generation of aerosol and / or using a specific heating profile.

[0060] In principle, any suitable sensor measuring any suitable parameter for determining the sensing profile may be implemented. In the following, several suitable sensors and several suitable parameters are described as examples in more detail. The invention may comprise use of any one of the sensors or of any combination of the sensors. The invention may also comprise using two or more sensors of the same type, in or without combination with other types of sensors. Also, the list is not exhaustive, and any other suitable sensors may be implemented in a similar manner in the invention.

[0061] For example, the at least one sensor may be a gas sensor and the sensing profile may comprise information about an effect of an exposure of at least part of the gas sensor to one or more gasses released from the aerosol-generating article or substrate, particularly during heating, more particularly during preheating of the aerosol-generating article or substrate. Before the aerosolization temperature is reached, one or more gases may already be released from the aerosol-generating article or substrate. These gases may be released even without any heating, but even more so during preheating the aerosol-generating article or substrate from the ambient temperature to the maintenance temperature. The gas sensor may therefore be configured to sense gases released from the aerosol-generating article or substrate at ambient temperature and / or between the ambient temperature and the maintenance temperature and / or at the maintenance temperature and / or below the aerosolization temperature. The gas sensor may, for example, be a metal-oxide-semiconductor (MOS) sensor. A MOS sensor may change its electrical resistance depending on different characteristics of gases coming into contact with a sensing part or sensing layer of the MOS sensor. Therefore, the parameter measured by the gas sensor may be its resistance. During preheating, for example, the resistance of a MOS sensor may change as the concentration of one or more gases released from the aerosol-generating article or substrate increases. For example, the gas sensor may be configured to react to the presence of volatile organic compounds (VOC), which play a major role in the composition and characterization of aerosol-generating articles or substrates.

[0062] Additionally or alternatively, the at least one sensor may be a temperature sensor and the sensing profile may comprise information about the temperature of the aerosol-generating article or substrate and / or a heating device configured to heat the aerosol-generating article or substrate during heating. The composition of the aerosol-generating article or substrate, which may be different for each type, may have a strong influence on the temperature curve of the aerosolgenerating article or substrate during heating. For example, the time it takes to reach a specific temperature of the aerosol-generating article or substrate may depend on its composition, andtherefore its type. The temperature may therefore be used as the parameter to derive the type of the aerosol-generating article or substrate. In some aerosol-generating devices, the heating device may be in contact with the aerosol-generating article or substrate and may conduct heat directly to the aerosol-generating article or substrate. The composition of the aerosol-generating article or substrate may influence the amount of heat that may be conducted or the speed of heat conduction from the heating device to the aerosol-generating article or substrate. Therefore, the temperature of the heating device may also be used as the parameter to derive the type of the aerosol-generating article or substrate.

[0063] The heating device may be a resistive heating device and / or may comprise a resistive heating element. Heat may therefore be generated by ohmic losses in the resistive heating element. The electrical resistance of a heating element may be proportional to its temperature. Therefore, the electric resistance of the resistive heating element may similarly be used as the parameter to derive the type of the aerosol-generating article or substrate as already explained for the temperature itself. Thus, additionally or alternatively, the at least one sensor may be a resistance sensor and the sensing profile may comprise information about the electrical resistance of a resistive heating element configured to heat the aerosol-generating article or substrate.

[0064] An alternative method of heating the aerosol-generating article or substrate may be inductive heating. The heating device of the aerosol-generating device may therefore comprise an inductor configured to heat a susceptor arranged in the aerosol-generating article or substrate. For example, the heating device may comprise an inductor coil, for example wherein the heating chamber is arranged in the center of the coil. The aerosol-generating article or substrate may, in turn, comprise a susceptor element, for example made from or comprising electrically conductive materials, namely metals or semi-conductors, such as a piece of metal foil provided inside the aerosol-generating article or substrate, for example in contact with or in close proximity to the aerosol-generating material. Both the inductor and the susceptor may react differently depending on the composition of the aerosol-generating article or substrate. Therefore, the at least one sensor may be an inductance sensor and the sensing profile may comprise information about the inductance of an inductor configured to heat a susceptor arranged in the aerosol-generating article or substrate. Additionally or alternatively, the at least one sensor may be a susceptibility sensor and the sensing profile may comprise information about the susceptibility of a susceptor arranged in the aerosol-generating article or substrate. Either one or both of the inductance of the inductor and / or the susceptibility of the susceptor may therefore be used as the parameter to derive the type of the aerosol-generating article or substrate.

[0065] Additionally or alternatively, the at least one sensor may be a voltage and / or current sensor and the sensing profile may comprise information about the voltage over and / or current througha heating element configured to heat the aerosol-generating article or substrate. For example, the power supplied to the heating element or heating device may be determined. The voltage and / or the current and / or the power supplied to the heating element may vary depending on the composition of the aerosol-generating article or substrate. These values may also be used in conjunction with the temperature of the aerosol-generating article or substrate and / or the heating device. In this way, a relationship may be obtained of how much power is necessary to reach a certain temperature in the aerosol-generating article or substrate, which may allow to derive an information about the composition of the aerosol-generating article or substrate and therefore its type. Thus, one or more or all of voltage over and / or current through and / or the power supplied to a heating element configured to heat the aerosol-generating article or substrate may therefore be used as the parameter to derive the type of the aerosol-generating article or substrate.

[0066] Additionally or alternatively, the at least one sensor may be a puff sensor and the sensing profile may comprise information about a pressure change in the puff sensor during a puff of a user. A puff sensor may be configured to detect when a user takes a puff on the aerosolgenerating device. For example, the puff sensor may be a differential pressure sensor. The puff sensor may be used to determine when, during the usage session, the aerosol-generating article or substrate needs to be heated from the maintenance temperature to the aerosolization temperature. The aerosol-generating article or substrate may only be heated from the maintenance temperature to the aerosolization temperature during a puff of the user. The composition of the aerosol-generating article or substrate may influence the resistance of the aerosol-generating article or substrate to airflow through the aerosol-generating article or substrate. Therefore, the composition of the aerosol-generating article or substrate may influence the pressure or pressure change induced by the puff and determined by the puff sensor. Therefore, the pressure and / or pressure change determined by the puff sensor during a puff of a user may be used as the parameter to derive the type of the aerosol-generating article or substrate.

[0067] As already mentioned, any combination of the sensors and / or parameters as mentioned above may be implemented in the method and / or the aerosol-generating device.

[0068] According to another aspect of the present invention, there is provided an aerosolgenerating device configured to perform the method as described herein. Particularly, an aspect of the present invention may be an aerosol-generating device comprising control circuitry comprising at least one controller and / or processor, wherein the control circuitry may be configured to perform the method according to the present disclosure. All of the functions, features and advantages of the method as described herein are also applicable to the aerosol-generating device and vice versa.According to another aspect of the present invention, there is provided an aerosolgenerating system comprising an aerosol-generating device and at least one of a companion device configured to charge the aerosol-generating device with electrical energy, a computing device, for example a smartphone, and a server device, for example a web server or a cloud server, wherein the aerosol-generating system is configured to perform the method according the present disclosure. All of the features, functions and advantages of the method and / or the aerosolgenerating device as described herein are also applicable to the aerosol-generating system and vice versa.

[0069] In particular, the step of determining the type of the aerosol-generating article or substrate may be performed by the at least one of the companion device configured to charge the aerosolgenerating device with electrical energy, the computing device, and the server device. In other words, the method according to the present disclosure may be performed by either one of the devices which may also be part of the aerosol-generating system, namely the aerosol-generating device, the companion device, and the computing device.

[0070] The aerosol-generating device as described herein and / or the aerosol-generating system as described herein may further comprise an aerosol-generating article or substrate. The aerosolgenerating device may be configured to generate aerosol from the aerosol-generating substrate or article. Also, the aerosol-generating device and / or the aerosol-generating system may be configured to determine the type of the aerosol-generating article or substrate, for example by performing the method as described herein.

[0071] According to another aspect of the present invention, there is provided a computer program, which when executed by processing circuitry of an aerosol-generating device, for example an aerosol-generating device according to the present invention, or a companion device configured to charge an aerosol-generating device with electrical energy, or a computing device, causes the aerosol-generating device or the companion device or the computing device to perform the steps of the method according to the present invention. All of the features, functions and advantages described herein for the method and / or the aerosol-generating device and / or the aerosolgenerating system are also applicable to the computer program and vice versa.

[0072] According to another aspect of the present invention, there is provided a non-transitory computer-readable medium storing a computer program according to the present invention. All of the features, functions and advantages described herein for the method and / or the aerosolgenerating device and / or the aerosol-generating system and / or the computer program are also applicable to the computer-readable medium and vice versa.

[0073] The invention is defined in the claims. However, below there is provided a non-exhaustive list of non-limiting examples. Any one or more of the features of these examples may be combined with any one or more features of another example, embodiment, or aspect described herein.Example 1. A computer-implemented method of determining a type of an aerosolgenerating article or substrate couplable to an aerosol-generating device, the method comprising:

[0074] receiving a sensing profile by at least one sensor, wherein the sensing profile comprises information about the evolution in time of a parameter indicative of and / or influenced by the composition of the aerosol-generating article or substrate, and

[0075] determining the type of the aerosol-generating article or substrate from the sensing profile based on evaluating the sensing profile with an artificial intelligence-based, Al-based, model.

[0076] Example 2. The method according to any one of the previous Examples, further comprising

[0077] classifying the sensing profile into one of a plurality of classes, each class being representative of and / or associated with a particular type of aerosol-forming article or substrate.

[0078] Example 3. The method according to the previous Example,

[0079] wherein the Al-based model is configured and / or trained to classify the sensing profile into one of the plurality of classes.

[0080] Example 4. The method according to any one of the previous Examples, wherein the Al-based model is configured to provide, based on the sensing profile, a classification result as output data, the classification result being indicative of the type of aerosolgenerating article or substrate.

[0081] Example 5. The method according to any one of the previous Examples, further comprising

[0082] extracting feature information from the sensing profile.

[0083] Example 6. The method according to the previous Example,

[0084] wherein the feature information comprises one or more of mathematical feature information, one or more mathematical features, one or more derivatives, one or more time derivatives, one or more statistical features, statistical feature information, and a standard deviation of the sensing profile.

[0085] Example 7. The method according to any one of Examples 5-6,

[0086] wherein the feature information comprises one or more of a time series complexity, an autocorrelation, a maximum derivative, a number of peaks, a permutation entropy, a forward Sharpe ratio, a backward Sharpe ratio, a minimum derivative, a logarithmic sum, an integral, a difference measure, a peak-to-peak value, a root mean square, a mean slope, a kurtosis, a skewness, a median, a mean absolute deviation, an interquartile range, a zero-crossing rate, an entropy, an energy, a wavelet energy, and an instantaneous frequency.

[0087] Example 8. The method according to any one of Examples 5-7, further comprising: generating input data for the Al-based model based on the extracted feature information of the sensing profile.Example 9. The method according to any one of Examples 5-8,

[0088] wherein the feature information comprises a difference measure in comparison to one or more reference profiles.

[0089] Example 10. The method according to any one of Examples 5-9,

[0090] wherein the feature information is extracted from one or more parts of the sensing profile, particularly a first part of the sensing profile at the beginning of the sensing profile and / or a second part of the sensing profile at the end of the sensing profile.

[0091] Example 11. The method according to any one of Examples 5-10,

[0092] wherein the feature information is combined into a linear combination, for example wherein each feature information is multiplied with a weighting factor.

[0093] Example 12. The method according to the previous Example, further comprising optimizing one or more linear combinations of the feature information for differentiating between different types of the aerosol-generating article or substrate, for example by maximizing a distance between mean values associated with the different types, for example using linear discriminant analysis, LDA, and / or maximizing a margin between the values associated with the different types, for example using a support vector machine, SVM.

[0094] Example 13. The method according to any one of Examples 5-12,

[0095] wherein the feature information is grouped into clusters based on similarity, for example using a k-means model.

[0096] Example 14. The method according to any one of Examples 5-13,

[0097] wherein the Al-based model is configured and / or trained to evaluate or analyze the sensing profile in terms of the feature information.

[0098] Example 15. The method according to any one of the previous Examples, wherein determining the type of the aerosol-generating article or substrate from the sensing profile comprises correlating the sensing profile, for example through the linear combinations of feature information of the sensing profile, to the type of the aerosol-generating article or substrate.

[0099] Example 16. The method according to any one of the previous Examples, wherein the Al-based model is a machine learning model.

[0100] Example 17. The method according to any one of the previous Examples, wherein the Al-based model is or comprises a support vector machine, SVM, an XGBoost classifier, a linear discriminant analysis, LDA, classifier, a decision tree classifier, a random forest classifier, or a k-nearest neighbors, k-NN, algorithm classifier.

[0101] Example 18. The method according to any one of the previous Examples, wherein the Al-based model is or comprises an ensemble learning model for making the final classification decision, for example a random forest model.

[0102] Example 19. The method according to any one of the previous Examples,wherein the Al-based model is trained on a training data set comprising sensing profiles of known types of aerosol-generating articles or substrates.

[0103] Example 20. The method according to any one of the previous Examples, wherein determining the type of the aerosol-generating article or substrate comprises determining whether the aerosol-generating article or substrate is an authentic aerosol-generating article or substrate provided by the manufacturer of the aerosol-generating device or whether the aerosol-generating article or substrate is provided by a third party.

[0104] Example 21. The method according to any one of the previous Examples, further comprising

[0105] controlling a heating device of the aerosol-generating device to provide or apply a specific heating profile depending upon the determined type of the aerosol-generating article or substrate.

[0106] Example 22. The method according to any one of the previous Examples, further comprising

[0107] allowing or disallowing generation of aerosol from the aerosol-generating article or substrate depending upon the determined type of the aerosol-generating article or substrate.

[0108] Example 23. The method according to any one of the previous Examples, wherein receiving the sensing profile includes determining the sensing profile.

[0109] Example 24. The method according to any one of the previous Examples, wherein receiving the sensing profile includes determining the sensing profile by at least one sensor.

[0110] Example 25. The method according to any one of the previous Examples, wherein the sensing profile is indicative of the evolution in time of the at least one parameter during heating of the aerosol-generating article or substrate.

[0111] Example 26. The method according to any one of the previous Examples, wherein the sensing profile and / or the type of the aerosol-generating article or substrate are determined during a preheating phase, in which the aerosol-generating article or substrate is heated from an ambient temperature to a maintenance temperature, particularly before an aerosolization temperature of the aerosol-generating article or substrate is reached.

[0112] Example 27. The method according to any one of the previous Examples, wherein the sensing profile and / or the type of the aerosol-generating article or substrate are determined within a maximum of 20 seconds after insertion of the aerosol-generating article or substrate into the aerosol-generating device, for example within a maximum of 15 seconds or within a maximum of 10 seconds or within a maximum of 8 seconds or within a maximum of 6 seconds or within a maximum of 5 seconds after insertion of the aerosol-generating article or substrate into the aerosol-generating device.

[0113] Example 28. The method according to any one of the previous Examples,wherein the sensing profile and / or the type of the aerosol-generating article or substrate are determined before a usage session using the aerosol-generating article or substrate is started.

[0114] Example 29. The method according to any one of the previous Examples, wherein the at least one sensor is a gas sensor and the sensing profile comprises information about an effect of an exposure of at least part of the gas sensor to one or more gasses released from the aerosol-generating article or substrate, particularly during heating.

[0115] Example 30. The method according to any one of the previous Examples, wherein the at least one sensor is a temperature sensor and the sensing profile comprises information about the temperature of the aerosol-generating article or substrate and / or a heating device configured to heat the aerosol-generating article or substrate during heating.

[0116] Example 31. The method according to any one of the previous Examples, wherein the at least one sensor is a resistance sensor and the sensing profile comprises information about the electrical resistance of a resistive heating element configured to heat the aerosol-generating article or substrate.

[0117] Example 32. The method according to any one of the previous Examples, wherein the at least one sensor is an inductance sensor and the sensing profile comprises information about the inductance of an inductor configured to heat a susceptor arranged in the aerosol-generating article or substrate.

[0118] Example 33. The method according to any one of the previous Examples, wherein the at least one sensor is a susceptibility sensor and the sensing profile comprises information about the susceptibility of a susceptor arranged in the aerosol-generating article or substrate.

[0119] Example 34. The method according to any one of the previous Examples, wherein the at least one sensor is a voltage and / or current sensor and the sensing profile comprises information about the voltage over and / or current through a heating element configured to heat the aerosol-generating article or substrate.

[0120] Example 35. The method according to any one of the previous Examples, wherein the at least one sensor is a puff sensor and the sensing profile comprises information about a pressure change in the puff sensor during a puff of a user.

[0121] Example 36. An aerosol-generating device comprising control circuitry comprising at least one controller and / or processor, wherein the control circuitry is configured to perform the method according to any one of the previous Examples.

[0122] Example 37. An aerosol-generating system comprising

[0123] an aerosol-generating device and

[0124] at least one of a companion device configured to charge the aerosol-generating device with electrical energy, a computing device, for example a smartphone, and a server device,wherein the aerosol-generating system is configured to perform the method according to any one of Examples 1-35,

[0125] particularly wherein the step of determining the type of the aerosol-generating article or substrate is performed by the at least one of the companion device configured to charge the aerosol-generating device with electrical energy, the computing device, and the server device.

[0126] Example 38. The aerosol-generating device according to Example 36 or the aerosolgenerating system according to Example 37, further comprising

[0127] an aerosol-generating article or substrate, preferably wherein the aerosol-generating device is configured to generate aerosol from the aerosol-generating substrate or article.

[0128] Example 39. A computer program, which, when executed by processing circuitry of an aerosol-generating system or an aerosol-generating device or a computing device, causes the aerosol-generating system or the aerosol-generating device to perform the steps of the method according to any one of Examples 1-35.

[0129] Example 40. A non-transitory computer-readable medium storing a computer program according to the previous Example.

[0130] Examples will now be further described with reference to the figures in which:

[0131] Figure 1 shows two examples of aerosol-generating devices;

[0132] Figure 2 shows two examples of aerosol-generating articles or substrates;

[0133] Figure 3 shows example sensing profiles;

[0134] Figure 4 shows a flowchart of the method;

[0135] Figure 5 shows a flowchart of the step of determining the type of aerosol-generating article or substrate from the sensing profile;

[0136] Figure 6 shows a visualization of a dimensionality reduced feature space; and

[0137] Figure 7 shows an overview of an aerosol-generating system.

[0138] The figures are schematic only and not to scale.

[0139] Figure 1 shows two examples of aerosol-forming or aerosol-generating devices 1 for forming or generating aerosol, for example for consumption or inhalation by a user in one or more usage sessions. On the left side of Figure 1 , an aerosol-generating device 1 is shown which may be configured to at least partly receive an aerosol-forming article or substrate 2 into an insertion opening 3, which may lead into the heating chamber of the aerosol-generating device 1. Part of the aerosol-forming article or substrate 2 may protrude from the aerosol-generating device 1 and may be directly used as mouthpiece for a user to puff on. Typically, the aerosol-forming article or substrate 2 used in this type of aerosol-generating device 1 may be formed like cylindrical sticks, as exemplarily shown, but other shapes and / or forms may also be used. The aerosol-generating article 2 may comprise an aerosol-generating substrate, such as a tobacco containing substrate, and / or a cartridge comprising a liquid, for example a liquid that can be aerosolized for inhalation.On the right side of Figure 1, another type of aerosol-generating device 1 is shown. The aerosolgenerating device 1 may comprise a mouthpiece 4, which may be part of the housing of the device, through which a user may inhale aerosol provided by the aerosol-generating device 1 for consumption during a usage session. The aerosol may be provided from an aerosol-generating article or substrate 2 provided inside the aerosol-generating device 1 and therefore not visible in Figure 1.

[0140] The aerosol-generating device 1 may further include processing circuitry or control circuitry 18 with at least one controller 5 and one or more processors 6. For generating the aerosol during use or consumption of the aerosol-generating article 2, the aerosol-generating device 1 may comprise at least one heating element 7 or heater device for applying heat to at least a portion of the aerosol-generating article 2. Instead of the heating element 7, an ultrasonic device (not shown) may also be used to generate aerosol from the aerosol-generating article. The processing circuitry 18 and / or the controller 5 and / or the processor 6 may be configured to control actuation, activation and / or deactivation of at least one heating element 7 or ultrasonic device.

[0141] For powering the at least one heating element 7 with electrical power, the aerosolgenerating device 1 may further comprise the at least one energy storage 15, for example in the form of a battery, for storing electrical energy or power. Energy storage 15 may be removably couplable to the aerosol-generating device 1. In other words, energy storage 15 may be a replaceable energy storage or battery. The connection between the energy storage 15 and the aerosol-generating device 1 may be configured so that the device 1 may be run by electrical energy provided by the energy storage 15.

[0142] The aerosol-generating device 1 may further comprise at least one electrical connector 12 for coupling to a corresponding electrical connector on a companion device (see Figure 7) for the aerosol-generating device 1 and / or an electrical connector of an external power supply (not shown), e.g., a USB charger.

[0143] The aerosol-generating device 1 may further comprise a communications arrangement 9 or communication circuitry 9 with one or more communications interfaces 10 for communicatively coupling the aerosol-generating device 1 with a companion device or other devices, for example, via an Internet connection, a wireless LAN connection, a WiFi connection, a Bluetooth connection, a mobile phone network, a mobile data connection for example but not limited to a 3G / 4G / 5G connection, an edge connection, an LTE connection, a BUS connection, a wireless connection, a wired connection, an optical data connection such as but not limited to IrDa, a radio connection, a near field connection, and / or an loT connection.

[0144] The aerosol-generating device 1 may further comprise a data storage 11 or memory for storing information, program code or data. Data storage 11 may also store collected values of sensors and / or one or more mathematical functions or formulas, software and computerinstructions that can be executed by the processing circuitry 18, particularly controller 5 and / or processor 6.

[0145] The aerosol-generating device 1 may further comprise user interface components, for example comprising an input element or input device 8, for example in the form of a pushbutton or a capacitive button. The input device 8 may be used as a power button to activate or deactivate the heating element 7 or ultrasonic device for aerosol generation thereby to activate or deactivate the aerosol-generating device 1. Upon activation of the aerosol-generating device 1, the heating element 7 may be activated and heat may be applied to at least a part of the aerosol-generating article or substrate 2, such that aerosol can be generated for consumption or inhalation by the user, for example in a usage session. The aerosol-generating device 1 may comprise one or more output elements, such as a display device 17 and / or one or more LEDs, for outputting a signal and / or displaying information to a user, for example a user interface such as a GUI, or haptic and acoustic data output devices. The display device 17 may be, for example, a touchscreen and may therefore be configured as both an output and an input element.

[0146] One or more sensors 16 may be arranged on, at or in the aerosol-generating device 1 to collect data. One or more of the sensors 16 may for example be temperature sensors, strain sensors, accelerometers, gas sensors, inductance sensors, susceptibility sensors, resistance sensors, puff sensors, voltage sensors, current sensors, or any other suitable sensors. Any one or more of the sensors 16 may be used for determining the type of the aerosol-generating article or substrate 2 inserted into the device 1 by a user as explained herein.

[0147] Figure 2 exemplarily shows two different shapes of aerosol-generating article or substrate 2 which may both be classified according to the present disclosure. The left side of Figure 2 shows a cylindrical rod- or stick-shaped aerosol-generating article or substrate 2. The right side of Figure 2 shows a plate- or cuboid- or parallelepiped-shaped aerosol-generating article or substrate 2. Both may comprise a frame or outer cover or housing made from or comprising a cellulose material, like paper and / or cardboard, and / or a plastic material. An aerosol-generating substrate or aerosol-generating material may be arranged inside the aerosol-generating article 2. The shapes as shown in Figure 2 are merely exemplary, and other shapes of aerosol-generating articles or substrates 2 may be used in the invention. The aerosol-generating device 1 may be configured for use with any one of these particular shapes. However, despite the shape being the same, different types of aerosol-generating articles or substrates 2, each having a different composition, particularly with regard to the aerosol-generating substrate or aerosol-generating material, may be available. The present invention relates to differentiating between these different types by using sensing profiles determined by at least one sensor 16 of the aerosol-generating device 1.Exemplary sensing profiles which may be used for this purpose are shown in Figure 3. Particularly, Figure 3 shows exemplary sensing profiles S1, S2, S3 measured for an exemplary type of aerosol-generating article or substrate 2. Each of the sensing profiles S1, S2, S3 comprises information about the evolution in time of a parameter indicative of and / or influenced by the composition of the aerosol-generating article or substrate 2. In the example shown in Figure 3, the sensing profiles S1, S2, S3 were measured by three gas sensors 16, in particular three MOS sensors. The abscissa or x-axis shows the time t elapsed during the measurement in seconds, whereas the ordinate ory-axis shows the measured resistance R in ohm. The sensing profile determined by the first MOS sensor is denoted by S1, the sensing profile determined by the second MOS sensor is denoted by S2 and the sensing profile determined by the third MOS sensor is denoted by S3. The three exemplary MOS sensors may each comprise a different sensing layers or sensing material, which may result in the slightly different sensing profiles S1, S2, S3 and which may thus increase the accuracy of the determination of the type of aerosolgenerating article or substrate 2. At around second 40 of the measurement, an aerosol-generating article or substrate 2 was inserted into the aerosol-generating device 1. At around second 60, a usage session was started. Therefore, the period of 20 seconds between second 40 and second 60 comprises the preheating phase in which the temperature of the aerosol-generating article or substrate 2 was increased from ambient temperature to the maintenance temperature. As can be seen from the three sensing profiles S1, S2, S3, the resistance of the different MOS sensors changes differently, but distinctively, in this period of time. The changes of the resistance as measured in the sensing profiles S1, S2, S3 represent the effects of the exposure of part of the gas sensor, specifically the sensing layer, to one or more gases released from the aerosolgenerating article or substrate 2. The differences in the three sensing profiles S1, S2, S3 are due to the different materials of the sensing layers 22 implemented in the specific MOS sensors. From the combination of these three sensing profiles S1, S2, S3, the type of an aerosol-generating article or substrate 2 may be determined, for example by the Al-based model as described herein. However, it is noted that the sensing profiles S1, S2, S3 as shown in Figure 3 are merely exemplary and that other sensors 16 may be used and other parameters may be measured. Also, the number of sensors 16 used is merely exemplary and the method may be implemented with only one sensor 16 or with more than three sensors 16.

[0148] Figure 4 shows a flowchart of method 60 according to the present disclosure. Method 60 may start in step 61 by receiving a sensing profile by at least one sensor 16, wherein the sensing profile may comprise information about the evolution in time of a parameter indicative of and / or influenced by the composition of the aerosol-generating article or substrate 2. For example, two or more sensing profiles may be determined by two or more sensors 16. In step 62, the method 60 may comprise determining the type of the aerosol-generating article or substrate 2 from thesensing profile, for example from all sensing profiles determined by the sensors 16. This may be based on evaluating the sensing profile with an Al-based model as described herein. The Al-based model may be employed to determine the type of aerosol-generating article or substrate 2 inserted into the aerosol-generating device 1.

[0149] Optionally, method 60 may comprise step 63 of detecting the insertion of an aerosolgenerating article or substrate 2 into the aerosol-generating device 1. For example, a detected insertion may trigger the start of the determination of one or more sensing profiles and / or the type of the aerosol-generating article or substrate 2 inserted into the aerosol-generating device 1 as described herein. Also, the method 60 may comprise step 64 of heating the aerosol-generating article or substrate 2, for example of pre-heating the aerosol-generating article or substrate 2 from an ambient temperature to the maintenance temperature. This may cause changes, i.e. an evolution in time, of the measured parameter, which in turn may be used to reliably determine the type of the aerosol-generating article or substrate 2 inserted into the aerosol-generating device 1.

[0150] After the type of the aerosol-generating article or substrate 2 has been determined, method 60 may comprise step 65 of allowing, disallowing or adjusting the operation of the heater device or heating elements 7 of the aerosol-generating device 1. For example, heating may be allowed and a suitable heating profile may be selected for a known and authorized or original type of aerosol-generating article or substrate 2. Similarly, heating may be allowed and a suitable heating profile may be selected for a known and unauthorized or non-original type of aerosol-generating article or substrate 2, for example for an aerosol-generating article or substrate 2 provided by a third party. A suitable heating profile for an unauthorized or non-original type of aerosol-generating article or substrate 2 may comprise lower heating temperatures than a heating profile for an authorized or original type of aerosol-generating article or substrate 2. Alternatively, heating may be disallowed for known and / or unknown unauthorized or non-original types of aerosol-generating article or substrate 2. Heating may also be disallowed when the presence of a potentially toxic constituent is detected in the sensing profiles, particularly when a gas sensor is used, for example in a concentration above a predetermined threshold.

[0151] Figure 5 shows some optional steps comprised in step 62 of determining the type of the aerosol-generating article or substrate 2 from the sensing profile using the Al-based model. For example, the data constituting the sensing profile may be preprocessed. One possible reprocessing of the sensing profile may comprise step 66 of extracting features or feature information from the sensing profile. A plurality of features or feature information may be extracted from the sensing profile, for example mathematical or statistical feature information as explained herein. The feature information may illustrate or indicate characteristics of the sensing profiles which may not have been obvious from the profiles themselves. The features of featureinformation may be or may represent numerical values which may be used for further analysis by the Al-based model. In step 67, further preprocessing may comprise reducing the dimensionality of the extracted features of feature information. As a specific example, LDA may be used to reduce the dimensionality by finding and / or computing linear combinations of the features or feature information which separates the different classes or types of aerosol-generating articles or substrates 2. However, other methods or algorithms for dimensionality reduction may alternatively be used. The separation of classes and / or types of aerosol-generating articles or substrates 2 in the linear combinations may further be enhanced by SVM or other suitable models and / or algorithms. Step 68, which may also be comprised in step 62 of the method 60, may then pertain to classifying the aerosol-generating article or substrate 2 into the different types by using the reduced dimensionality features or feature information. In otherwords, the type of the aerosolgenerating article or substrate 2 may be determined from the reduced dimensionality feature information.

[0152] The steps of the method 60 as explained herein may be used both in training the Al-based model and in employing the Al-based model to determine the type of an aerosol-generating article or substrate 2 used by a user. In training, the type of the aerosol-generating article or substrate 2 is known, so that a training data set may be generated. This training data set may then be used to train the Al-based model so that the Al-based model may correctly determine the type of an unknown aerosol-generating article or substrate 2 inserted into the aerosol-generating device 1 by a user. Exactly which features or feature information to extract from the sensing profile and which linear combinations of features or feature information to use may be determined during training, for example by optimizing the Al-based model in terms of class separation and classification result. In use, then, the same features or feature information are extracted from the sensing profile and the same linear combinations using the same weighting factors are computed. These are then mapped to the classes or types that the Al-based model was trained on during training, providing the desired result.

[0153] Figure 6 shows a diagram visualizing how dimensionality reduction may be used to differentiate between different types of aerosol-generating articles or substrates 2. The diagram shows a 3-dimensional feature space, in which the three axes labeled comp-1, comp-2, and comp-3 each represent a linear combination of one or more or a plurality of features or feature information. Graphs as shown in Figure 6 may be produced using LDA. The different volumes in feature space may represent the results of the LDA. These volumes help visualize how the classes or types of aerosol-generating articles or substrates 2 are separated in the reduced dimensional space after applying LDA. Each small letter a-j in the diagram represents one aerosol-generating article or substrate 2, wherein the same letters represent aerosol-generating articles or substrates 2 of the same type. As can be seen in Figure 6, aerosol-generating articlesor substrates 2 of the same type are grouped together in the same volumes of the feature space as indicated by the oval borders. Without dimensionality reduction, the feature space would have as many dimensions as feature information or features are used. As a plurality of features or feature information may be extracted from the sensing profiles, this may primarily increase complexity of the problem to be solved. As LDA weights the features to optimize class separation, the dimensionality reduction may simplify the underlying problem without losing valuable information.

[0154] Figure 7 shows an aerosol-generating system which may perform method 60. The aerosolgenerating system may comprise an aerosol-generating device 1 as explained above. The system may also comprise a companion device 14 configured to charge the aerosol-generating device 1 with electrical energy. For example, the companion device 14 may comprise an opening 19 into which the aerosol-generating device 1 may at least partly be inserted. By inserting the aerosolgenerating device 1 into the companion device 14, an electrical connection may be established between the devices so that the companion device 14 may charge the aerosol-generating device 1 with electrical energy. The aerosol-generating system may also comprise a computing device 20, which may, for example, be a smartphone, a tablet computer, a smartwatch, a personal computer or a similar device. The aerosol-generating device 1, the companion device 14 and / or the computing device 20 may transmit data between them, for example through a data connection 22, which may be a wired or a wireless data connection. For example, data may be transmitted between the aerosol-generating device 1 and the companion device 14 when the aerosolgenerating device 1 is at least partly inserted into the companion device 14. The data connection 22 may comprise a Bluetooth data connection, for example a Bluetooth low energy (BLE) connection, or may comprise a USB connection, for example a USB-C connection. However, any other data connection 22 facilitating data exchange between the mentioned devices may also be used. The aerosol-generating system may further comprise a server device 21. The server device 21 may, for example, be a Web server or a cloud server, for example provided and / or maintained and / or run by the manufacturer of the aerosol-generating device 1 and / or the companion device 14. The aerosol-generating device 1 and / or the companion device 14 may exchange data with the server device 21 through an Internet connection 23, which may, for example, be established through the computing device 20.

[0155] The sensor 16 for determining the sensing profile may be arranged on the aerosolgenerating device 1. However, the sensing profile as established by sensors 16 may be transmitted to the companion device 14 and / or the computing device 20 through data connection 22. The sensing profile may also be transmitted to the server device 21 through the Internet connection 23. Therefore, any of the aerosol-generating device 1, the companion device 14, the computing device 20 or the server device 21 may receive the sensing profile and may thereforeperform the method 60 as described herein. The result of the method 16, for example the type of the aerosol-generating article or substrate 2, may then be communicated to the aerosolgenerating device 1, and may be used for controlling the aerosol-generating device 1 in reaction to the determined type. For example, a heating profile specific for the determined type of aerosolgenerating article or substrate 2 may be employed. Alternatively, heating of the aerosolgenerating article or substrate 2 for the generation of aerosol may be allowed or disallowed.

[0156] In summary, therefore, the present disclosure provides for the determination of a type of the aerosol-generating article or substrate 2 used in the aerosol-generating device 1 without the need for specifically marking the aerosol-generating article or substrate 2. In this way, control of the aerosol-generating device 1 may be adjusted to the specific type as determined, improving both user safety and user experience. The production costs of the aerosol-generating articles or substrates 2 are also kept low, as no special markings are necessary.

[0157] For the purpose of the present description and of the appended claims, except where otherwise indicated, all numbers expressing amounts, quantities, percentages, and so forth, are to be understood as being modified in all instances by the term "about". Also, all ranges include the maximum and minimum points disclosed and include any intermediate ranges therein, which may or may not be specifically enumerated herein. In this context, therefore, a number A is understood as A ± 10 % of A. Within this context, a number A may be considered to include numerical values that are within general standard error for the measurement of the property that the number A modifies. The number A, in some instances as used in the appended claims, may deviate by the percentages enumerated above provided that the amount by which A deviates does not materially affect the basic and novel characteristic(s) of the claimed invention. Also, all ranges include the maximum and minimum points disclosed and include any intermediate ranges therein, which may or may not be specifically enumerated herein.

Claims

34 / 37CLAIMS1. A computer-implemented method of determining a type of an aerosol-generating article or substrate couplable to an aerosol-generating device, the method comprising:receiving a sensing profile by at least one sensor, wherein the sensing profile comprises information about the evolution in time of a parameter indicative of and / or influenced by the composition of the aerosol-generating article or substrate, anddetermining the type of the aerosol-generating article or substrate from the sensing profile based on evaluating the sensing profile with an artificial intelligence-based, Al-based, model.

2. The method according to any one of the previous claims,wherein the Al-based model is configured to provide, based on the sensing profile, a classification result as output data, the classification result being indicative of the type of aerosolgenerating article or substrate.

3. The method according to any one of the previous claims, further comprising extracting feature information from the sensing profile;reducing the dimensionality of the feature information; andgenerating input data for the artificial intelligence-based, Al-based, model based on the extracted feature information of the sensing profile,wherein the type of the aerosol-generating article or substrate is determined from the reduced dimensionality feature information.

4. The method according to the previous claim,wherein the feature information comprises one or more of mathematical feature information, one or more mathematical features, one or more derivatives, one or more time derivatives, one or more statistical features, statistical feature information, and a standard deviation of the sensing profile.

5. The method according to any one of claims 3 and 4,wherein the feature information extracted from the sensing profile at least comprises one or more of a time series complexity, an autocorrelation, and a maximum derivative.

6. The method according to any one of the previous claims, further comprising35 / 37extracting feature information from the sensing profile, wherein the feature information is combined into a linear combination, for example wherein each feature information is multiplied with a weighting factor; andoptimizing one or more linear combinations of the feature information for differentiating between different types of the aerosol-generating article or substrate bymaximizing a distance between mean values associated with the different types using linear discriminant analysis, LDA, andmaximizing a margin between the values associated with the different types using a support vector machine, SVM.

7. The method according to any one of the previous claims, further comprising extracting feature information from the sensing profile;wherein at least five features are derived from the sensing profile as feature information.

8. The method according to any one of the previous claims,wherein the Al-based model is or comprises a support vector machine, SVM, an XGBoost classifier, a linear discriminant analysis, LDA, classifier, a decision tree classifier, a random forest classifier, or a k-nearest neighbors, k-NN, algorithm classifier.

9. The method according to any one of the previous claims,wherein the Al-based model is or comprises an ensemble learning model for making the final classification decision, for example a random forest model.

10. The method according to any one of the previous claims,wherein the at least one sensor is a gas sensor and the sensing profile comprises information about an effect of an exposure of at least part of the gas sensor to one or more gasses released from the aerosol-generating article or substrate, particularly during heating.

11. An aerosol-generating device comprising control circuitry comprising at least one controller and / or processor, wherein the control circuitry is configured to perform the method according to any one of the previous claims.

12. An aerosol-generating system comprisingan aerosol-generating device andat least one of a companion device configured to charge the aerosol-generating device with electrical energy, a computing device, for example a smartphone, and a server device, wherein the aerosol-generating system is configured to perform the method according to any one of claims 1 to 10,particularly wherein the step of determining the type of the aerosol-generating article or substrate is performed by the at least one of the companion device configured to charge the aerosol-generating device with electrical energy, the computing device, and the server device.

13. The aerosol-generating device according to claim 11 or the aerosol-generating system according to claim 12, further comprisingan aerosol-generating article or substrate, preferably wherein the aerosol-generating device is configured to generate aerosol from the aerosol-generating substrate or article.

14. A computer program, which, when executed by processing circuitry of an aerosolgenerating system or an aerosol-generating device or a computing device, causes the aerosolgenerating system or the aerosol-generating device to perform the steps of the method according to any one of claims 1-10.

15. A non-transitory computer-readable medium storing a computer program according to the previous claim.

Images