Personal care device user classification

Abstract

Schemes, solutions, concepts, designs, methods, and systems relating to classifying a user of a personal care device that vibrates due to an actuator when in use are presented. In particular, a data set describing operating parameters of a personal care device (during vibration) is used to generate a spectrogram image. The spectrogram image is input to a user classification machine learning algorithm that outputs a predicted user and / or user type of the personal care device. Accordingly, operational settings of the personal care device may be adjusted based on the predicted user and / or user type, thereby improving efficacy and security of the personal care device.

Description

Technical Field

[0001] This invention relates to the classification of users of personal care devices. Background Technology

[0002] Different users will operate personal care devices (e.g., toothbrushes, razors, skin scrubbers, etc.) in different ways. For example, some users apply greater force to their skin with the device, while others apply less force. Furthermore, some users move the device erratically, while others move it carefully and methodically. These differences in how personal care devices are used will affect the effectiveness of the device's personal care functions.

[0003] If the users (or user types) of a personal care device are known, the device's operating settings can be adjusted / adapted to improve its functionality. In other words, if the users or user types are known, their individual usage characteristics can be considered in the device's operating settings. Therefore, the device settings can be personalized to ensure the effectiveness and safety of its operation.

[0004] For example, if the personal care device is a toothbrush and it is known that the user applies too much force when brushing, the toothbrush motor parameters can be adjusted to mitigate the negative effects of excessive force.

[0005] Therefore, there is a need for a device that automatically categorizes users of personal care devices. Summary of the Invention

[0006] According to an example of one aspect of the invention, a method for classifying users of personal care devices is provided. The personal care device includes an actuator that causes the device to vibrate during use. The method includes:

[0007] Obtain a dataset describing the operating parameters of the personal care device during vibration.

[0008] Generate spectrogram images based on at least a portion of the dataset; and

[0009] Spectrogram images are processed using user classification machine learning algorithms to predict users and / or user types of personal care devices.

[0010] A method for classifying users of personal care devices that vibrate due to actuator movement is proposed. Specifically, a dataset describing the operating parameters of the personal care device during vibration is used to generate spectrogram images. These spectrogram images are input into a user classification machine learning algorithm, which outputs a predicted user and / or user type for the personal care device. Therefore, the operating settings of the personal care device can be adjusted based on the user and / or user type. In other words, the operating settings of the personal care device can be personalized for the identified user and / or the predicted user type.

[0011] Personal care devices include actuators that cause the personal care device to vibrate during use. An actuator can be a component of a personal care device configured to affect personal care functions (e.g., a motor that drives the movement of a toothbrush head), or it can simply be a component configured to perform a different function but incidentally causes the personal care device to vibrate. In other words, an actuator can be any component that affects mechanical movement during use, thereby causing the personal care device to vibrate.

[0012] Furthermore, the dataset is acquired in response to vibrations of the personal care device. In other words, the dataset is acquired / obtained / generated when the personal care device vibrates due to the movement of the actuator. Therefore, features indicative of the vibration response of the personal care device can exist in the dataset.

[0013] The dataset for personal care devices describes the operating parameters of the personal care device during vibration. That is, the dataset contains values ​​indicating the system vibration response of the personal care device when the device is in operating mode (i.e., when the actuator moves and thus causes the personal care device to vibrate), measured in terms of, for example, acceleration (in one or more dimensions), motor current, and sound / noise. The vibration response included in the dataset reflects the user and / or user type, as changes in the user and / or user type handling the personal care device will cause changes in the vibration response of the personal care device (towards the movement of the actuator).

[0014] Generate a spectrogram image from (at least a portion of the dataset). That is, process the dataset using known methods to generate the spectrogram image. For example, the dataset can be processed using a Fast Fourier Transform (FFT) to generate the spectrogram image. The spectrogram image contains multiple FFTs, each corresponding to a time point. Each FFT represents the frequency intensity of the vibration response of the personal care device at the time point captured by the dataset.

[0015] Subsequently, the spectrogram image is fed to a user classification machine learning algorithm. The algorithm processes the spectrogram image and can identify various features indicating users and / or user types. Therefore, for example, when multiple different users frequently use personal care devices, predicted user characteristics can be provided. In this case, known (e.g., previous or historical) and user-associated operating settings for the personal care device can be implemented. Furthermore, if the user type is predicted, the operating settings can be adjusted appropriately.

[0016] For clarity, personal care devices can be, for example, electric toothbrushes / oral appliances, electric shavers, or skin scrubbers. Of course, the invention can be applied to a variety of other personal care devices configured to provide personal care functions to a user. Personal care devices can be handheld and / or portable.

[0017] In some embodiments, the user classification machine learning algorithm can be a spectral image classifier. A spectral image classifier classifies features in an image into different categories based on its spectral characteristics. Therefore, this algorithm is well-suited for processing / analyzing spectral images. More specifically, the user classification machine learning algorithm can be a CNN classifier or an FCNN classifier.

[0018] In addition, user classification machine learning algorithms can be CNN classifiers that include 3 to 6 convolutional layers.

[0019] Fewer convolutional layers in a CNN result in lower computational complexity and require less memory and processing resources. However, their capabilities also become less powerful. Three to six convolutional layers can provide a balance between the size and effectiveness of a CNN classifier.

[0020] The dataset may include accelerometer data describing acceleration on at least one axis of a personal care device.

[0021] It has been shown that datasets including accelerometer data in only one dimension / one axis are sufficient to facilitate predictions of users and / or user types of personal care devices. This can be relatively simple data to generate, and the hardware that generates this type of data exists in many existing personal care devices. Furthermore, many personal care devices include inertial motion units (IMUs) that capture acceleration data in three dimensions, where this type of data has the potential to further improve the accuracy of predictions.

[0022] When accelerometer data describes acceleration on three axes, the method may further include calculating norm vector data describing the absolute vector length of the acceleration based on the accelerometer data; and serializing the accelerometer data and norm vector data for each of the three axes into a single data segment.

[0023] Alternatively or additionally, the dataset may include sound data that describes the sounds produced by a personal care device.

[0024] Sound / noise / audio data has been shown to provide sufficient information about the vibration of personal care devices to facilitate user and / or user type prediction.

[0025] Alternatively or additionally, the dataset may include current data that describes the motor current of the actuator.

[0026] Similarly, current data has also been shown to provide sufficient information about the vibration of personal care devices to facilitate user and / or user type prediction. Current data can be obtained from the motor itself or via, for example, a battery to which the motor is connected. The motor can be a component in an actuator that influences the movement of the actuator, thereby causing the personal care device to vibrate.

[0027] Specifically, the obtained dataset may include values ​​describing the operating parameters of personal care devices, spanning up to one second.

[0028] In other words, the dataset includes values ​​spanning up to one second, or signals lasting up to one second. By limiting the time span of the dataset, processing and analysis can be simplified. Of course, the dataset can span more or less time, but approximately one second allows for accurate predictions of users and / or user types while achieving reasonable complexity.

[0029] The obtained dataset may include values ​​describing operating parameters of a personal care device sampled at a rate of at least 800 Hz. In some embodiments, the obtained dataset may include values ​​describing operating parameters of a personal care device sampled at a rate of at least 1.6 kHz.

[0030] Sampling the operating parameters at a rate of at least 800 Hz can be beneficial to ensure that the unique characteristics of the personal care device's vibration are present in the dataset. 1.6 kHz may be preferred to ensure that no unique features are missed, although increasing the dataset size will increase complexity.

[0031] This method may also include at least a normalized portion of the dataset. Therefore, when predicting users and / or user types, variations in the dataset due to differences between personal care devices (material, shape, condition, etc.) or the context in which the personal care devices were stored when the dataset was acquired (e.g., location, environment, etc.) can be considered. This allows for more accurate user classification across different personal care devices.

[0032] An exemplary embodiment of the present invention may provide that a dataset describing operating parameters is obtained in response to a personal care device being operated by a user.

[0033] The personal care device can be considered to be operated by the user when the user physically touches it. Preferably, this would be when the user manipulates the personal care device while performing personal care functions. In this way, the vibration response of the personal care device will depend on the user and will therefore vary with different users and user types.

[0034] More specifically, the dataset describing the operating parameters can be obtained during a single operating mode of the actuator.

[0035] In practice, a single operating mode can correspond to normal use of a personal care device (e.g., a single operating setting of the actuator in an electric toothbrush). This can mean that the user will not notice when the dataset is collected and that it can be collected during typical use of the personal care device. Essentially, this means that the actuator does not need to be controlled to capture the dataset in a way that causes the personal care device to vibrate in a specific manner (e.g., vibration scanning). In other words, the actuator is controlled only in its normal operating manner. Therefore, the present invention provides for the acquisition of datasets when the actuator is not operated outside of normal operating conditions, settings, and / or ranges.

[0036] Alternatively, a dataset describing the operating parameters can be obtained while changing the operating mode of the actuator to change the amplitude and / or duty cycle of the actuated motion (and thus the resulting vibration response of the personal care device).

[0037] In contrast to vibration scans caused by actuators operating outside their typical operating conditions or ranges, amplitude scans can be performed. By acquiring a dataset during an amplitude scan, additional unique vibration response-related features can be present in the dataset, which can be detected to predict users and / or user types. In other words, as an alternative to frequency / vibration scans outside the actuator's normal operating range, amplitude and / or duty cycle scans can be used to further enhance the features present in the dataset.

[0038] Predicting users and / or user types of personal care devices may include: using user classification machine learning algorithms to classify users of personal care devices into one user among multiple users and / or into one user type category among multiple user type categories based on spectrogram images.

[0039] In other words, a user can be predicted to be a specific individual user. Furthermore, a user can be categorized into one of several user type categories (each category describing different user characteristics). For example, based on the force a user exerts on themselves while using a personal care device (reflected in the vibration response held by the dataset), a user can be categorized into a low-force user category or a high-force user category.

[0040] Furthermore, predicting the user type of a personal care device may also include generating probability scores for each of multiple user type categories based on a spectrogram image using a user classification machine learning algorithm, where the probability scores describe the confidence level in predicting a user as a user type category.

[0041] For example, if a user type is predicted as a high-efficiency user with an association probability score of approximately 50%, and a user type is predicted as a low-efficiency user with an association probability score of approximately 50%, then the user type can be predicted as somewhere between a high-efficiency user and a low-efficiency user (i.e., a medium-efficiency user). Therefore, probability scores can be used to determine the level between categories.

[0042] The method may further include generating a user classification machine learning algorithm. Generating a user classification machine learning algorithm may include: generating multiple test datasets that describe the operating parameters of the personal care device during vibration of the personal care device when different, corresponding, known users manipulate it; and

[0043] A user classification machine learning algorithm is trained using a training algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include a test dataset from the test dataset, and the corresponding known outputs include known users.

[0044] Therefore, user classification machine learning algorithms that can accurately predict users and / or user types can be provided.

[0045] Each test dataset can include data generated during user operation of the personal care device. This can be implemented during user testing.

[0046] Alternatively or additionally, each test dataset may include synthetically generated data. This type of data can be valuable for increasing the size of the training dataset on which user classification machine learning algorithms can be trained. As the amount of data increases, the algorithm can become increasingly accurate and robust.

[0047] According to another aspect of the present invention, a computer program is provided, the computer program program including computer program code means, which, when the computer program is run on a computer, is adapted to implement the method of the proposed embodiment.

[0048] According to another aspect of the present invention, a personal care device is provided. The personal care device includes:

[0049] An actuator configured to cause the personal care device to vibrate during use;

[0050] A sensor used to obtain a dataset describing the operating parameters of the personal care device during vibration.

[0051] The processor is configured as follows:

[0052] Generate spectrogram images based on at least a portion of the dataset; and

[0053] Spectrogram images are processed using user classification machine learning algorithms to predict users and / or user types of personal care devices.

[0054] These and other aspects of the invention will become apparent and be elucidated with reference to one or more embodiments described below. Attached Figure Description

[0055] To better understand the invention and to more clearly illustrate how to practice it, reference will now be made to the accompanying drawings by way of example only, in which:

[0056] Figure 1 A flowchart of a method for classifying users of personal care devices according to one embodiment is presented;

[0057] Figure 2 It is a personal care device according to an example embodiment;

[0058] Figure 3 A simplified block diagram of a computer in which one or more portions of an embodiment may be employed is provided. Detailed Implementation

[0059] The invention will be described with reference to the accompanying drawings.

[0060] It should be understood that the accompanying drawings are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used to indicate the same or similar parts in all the accompanying drawings.

[0061] It should also be understood that while the detailed description and specific examples indicate exemplary embodiments of the apparatus, system, and method, they are for illustrative purposes only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, system, and method of the present invention will become more readily understood from the following description, the appended claims, and the accompanying drawings. The fact that certain measures are recited in mutually different dependent claims does not indicate that combinations of these measures cannot be used advantageously.

[0062] This paper proposes schemes, solutions, concepts, designs, methods, and systems related to the classification of users of personal care devices that vibrate due to mechanical actuators during use. Specifically, a dataset describing the operating parameters of the personal care device (during operation) is used to generate spectrogram images. These spectrogram images are input into a user classification machine learning algorithm, which outputs a predicted user and / or user type for the personal care device. Therefore, the operating settings of the personal care device can be adjusted based on the predicted user and / or user type, thereby improving the effectiveness of the personal care device.

[0063] The disclosed embodiments provide a method for user classification based on vibration-related operating parameters, such as a dataset describing these operating parameters obtained from a microphone, accelerometer, actuator motor current, or current from a battery connected to an actuator. Short samples of the dataset are converted into spectrogram images, and a neural network (e.g., a CNN-type network) is used to classify the signal based on features it finds in the spectrogram image, thereby classifying users and / or user types. In other words, processing is simplified by converting the dataset into a spectrogram image, which can be efficiently processed by machine learning algorithms to identify features for user classification (i.e., identification of a specific user, or determination of a user type) for personal care devices. Therefore, the disclosed method provides a sensitive device for user and / or user type prediction.

[0064] Furthermore, the disclosed embodiments can also be implemented using existing personal care device hardware, thus eliminating the need for expensive and complex additional components and sensors. That is, no additional sensors (e.g., NFC sensors) are required to implement the invention. This is because the operating parameter signals that rely on maintaining vibration response information can already be generated by sensors implemented in personal care devices for various other functions. For example, the solution can utilize signals from a 3-axis or 1-axis accelerometer, the motor current of an actuator, microphone signals, or load sensor signals—many of which are typically present in personal care devices.

[0065] User experience can be improved by effectively predicting users and / or user types for personal care devices. In fact, user and / or user type prediction can be used to adjust the operation of personal care devices to enhance device effectiveness by compensating for differences in user habits and operations. Furthermore, user-specific identification can be used to provide personal care feedback to users if needed. In the case of medical-grade personal care devices, this information can also help prevent unsafe situations and provide traceable evidence when necessary.

[0066] Using a toothbrush as an example, the disclosed embodiments can be implemented in existing hardware. This is true even when an inertial motion unit (IMU) sensor is absent, because the invention is capable of working with many vibration-related operating parameters, such as motor current, sound / noise signals, or load sensor signals. Some embodiments specify that approximately one second of operating parameter values ​​are converted into a spectrogram image and fed into a small CNN model to provide user and / or user type predictions. Therefore, due to limited memory and processing requirements, the disclosed embodiments can be implemented as new software installed on existing devices.

[0067] However, it is worth noting that the present invention can be implemented in a wide range of personal care devices, such as oral care mouthpieces, shaving devices, and skin brushing devices. Therefore, it is noteworthy that while toothbrushes and brush heads may particularly benefit from the present invention, it can also provide advantages for many other personal care devices.

[0068] Furthermore, methods for identifying users and / or predicting user types can be implemented in the handle of a personal care device, on an external processor, or even in the cloud or on an edge processing system.

[0069] The dataset used to predict users and / or user types describes the operating parameters of personal care devices (when the personal care device vibrates in response to the movement of the actuator during normal operation). The operating parameters that can be used relate to the vibrational response of the personal care device to the movement of the actuator, such as acceleration (in one or more dimensions), sound / noise, and motor / system current.

[0070] In some embodiments, the dataset is obtained in response to user manipulation of the personal care device. For example, the personal care device may be manipulated during operation of the device, or simply when the user physically moves the device. This ensures that features related to user-induced motion / vibration are included in the dataset, and consequently ensures that user classification machine learning algorithms can accurately predict the user and / or user type.

[0071] The dataset can span approximately one second, but it can span longer time periods to improve prediction accuracy, or shorter time periods to reduce the computational complexity of prediction.

[0072] The dataset can be sampled at a rate of at least 800 Hz, preferably 1.6 kHz, to capture a frequency range containing unique characteristics. Note that the rate can depend on the selected operating parameters. For example, it may be preferable to capture acceleration data at 1.6 kHz, while capturing sound and current at 16 kHz. In practice, the operating speed of the personal care device can affect its frequency response, thus affecting the data sampling rate. For example, a toothbrush operating at 200 Hz can sample data at a rate of 1.6 kHz, while a toothbrush operating at 500 Hz may require a higher sampling rate.

[0073] Furthermore, capturing the dataset at different bit depths (i.e., resolutions) can also affect the computational complexity and accuracy of the predictions. For sound data, 16 bits can be captured, acceleration data can be captured at 12 bits, and current data can be captured at 8 bits or 16 bits.

[0074] Of course, the selected operating parameters, data length, rate, and dataset resolution may depend on the chosen personal care device implementing the invention. Those skilled in the art will fully understand the range of selected parameters for the dataset and will be able to adjust these parameters for the chosen application.

[0075] Furthermore, the dataset can be preprocessed before being used to predict users and / or user types. Particularly advantageous is the ability to normalize the dataset so that variations in amplitude (e.g., caused by production variations in personal care devices) are accounted for. This mitigates variability between individual personal care devices.

[0076] Specifically, when the operating parameter is acceleration, the data format of acceleration data (e.g., from a 3-axis accelerometer) can be improved through preprocessing. Acceleration values ​​for each different axis (XYZ) can also include norm (N) calculations as a fourth set of values, either serialized or concatenated. Four sets of values ​​(XYZN) sampled at 1600 Hz over a 2.5-second time span will yield 16,000 data points (similar to a microphone sampling at 16 kHz for 1 second). This 3-axis accelerometer data format can be advantageous compared to, for example, single-axis accelerometer data or data from a microphone or current data, because it also includes vibration direction information within the serialized data format. Normalization can be performed on each value in this set of values ​​(XYZN) to preserve the direction information in the signal.

[0077] Once a dataset describing the operating parameters is obtained, and optionally preprocessed, this dataset is used to generate spectrogram images representing the signal strength over time at various frequencies in the vibration response of the personal care devices contained in the dataset. Thus, low-resolution (e.g., 128×128 pixels) spectrogram images can be generated, which have proven effective in performing user and / or user type classification with very high accuracy. Of course, higher-resolution images can be generated to ensure accuracy in user classification, but this could require significantly more computational resources.

[0078] The generation of spectrograms can be achieved using known methods, such as applying multiple Fast Fourier Transform (FFT) to the dataset. Those skilled in the art will readily understand the various applications and implementations of FFT on datasets.

[0079] Once the spectrogram image is generated, it is fed into a user classification machine learning algorithm (e.g., a trained neural network). The user classification machine learning algorithm processes the spectrogram image to predict the user and / or user type.

[0080] Artificial neural networks (or simply neural networks (NNs)) are inspired by the human brain. A neural network consists of layers, each containing multiple neurons. Each neuron performs a mathematical operation. Specifically, each neuron can include different weighted combinations of a single transformation type (e.g., the same type of transformation, sigmoid, etc., but with different weights). In processing input data, the mathematical operation of each neuron is performed on the input data to produce a numerical output, and the output of each layer in the neural network is sequentially fed to the next layer. The final layer provides the output.

[0081] There are several types of neural networks, such as convolutional neural networks (CNNs), fully connected neural networks (FCNNs), and recurrent neural networks (RNNs). Examples of this invention typically implement user classification machine learning algorithms as spectral image classifiers, such as CNNs or FCNNs.

[0082] Specifically, user classification machine learning algorithms can be classic convolutional neural network (CNN) classifiers with a finite number of layers. It has been shown that CNNs with 3 to 6 convolutional layers provide efficient processing of spectrogram images. Furthermore, these small CNN-based algorithms can be readily used in embedded firmware that can exist in existing personal care devices. Alternatively, cloud processing is also possible due to the small sample size.

[0083] Given a spectrogram image (corresponding to a dataset acquired during vibration while a personal care device is being operated by a user), a user classification machine learning algorithm can predict the user and / or user type. The algorithm is trained on previous spectrograms associated with known users (e.g., known specific individual users, or known user types).

[0084] Methods for training machine learning algorithms are well-known. Typically, such methods involve obtaining a training dataset that includes training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The error between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry. This is often referred to as a supervised learning technique.

[0085] For example, the weights of the mathematical operations for each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation, and other algorithms.

[0086] The training input data entries for the user classification machine learning algorithm correspond to example test datasets that describe the operating parameters of the personal care device during vibration when different corresponding known users manipulate the device. For each of the multiple test datasets, the training output data entries correspond to the corresponding known user and / or user type. That is, the machine learning algorithm is trained using a training algorithm configured to receive an array of training inputs and corresponding known outputs, where the training inputs include one of the test datasets, and the corresponding known outputs include known users (and therefore known user types). In this way, the machine learning algorithm is trained to output predicted users and / or user types when provided with a dataset describing the operating parameters during vibration when the personal care device is manipulated by a user.

[0087] In some cases, user classification machine learning algorithms can be trained to classify spectrogram images (and therefore users) into one of multiple user type categories. In one embodiment, a probability score can be provided for each category, indicating the confidence / probability of a correct prediction. The probability score provided for each category can be used to track the user type between two categories. For example, if the probability score associated with the high-intensity user category is 80%, while the probability score associated with the low-intensity user category is 20%, then the user type could be a medium-high-intensity user.

[0088] It will be understood that user classification machine learning algorithms can be (at least partially) pre-trained before deployment for predicting user types. However, calibration and / or retraining can be performed in response to end-user usage to ensure effective user and / or user type predictions. In practice, this can be performed to accurately predict (individual) users of personal care devices.

[0089] Test datasets can be obtained from real-world user testing. That is, test datasets can be obtained by sensing data describing operational parameters (e.g., measuring acceleration, sound, and / or motor current) while known users and / or user types are using the personal care device. Such datasets can be acquired during use, for example, by asking users who they are and what type of user they are. In this way, real-world user workload can be reflected in the test dataset, making machine learning algorithms more likely to provide accurate predictions under unseen conditions.

[0090] Additionally or alternatively, the test dataset can be synthetically generated data. In other words, the test dataset can be obtained by sensing a dataset describing operational parameters (e.g., measuring acceleration, sound, and / or motor current) while using a personal care device in a robotic test setting that mimics known users and / or user types. The robotic test setting can reflect / mimic real-world users based on real user data / loads. This allows for the generation of large test datasets spanning multiple test conditions in a relatively inexpensive and rapid manner (compared to test datasets obtained from user operations). Furthermore, by providing more test datasets, the accuracy and power of machine learning algorithms can be improved. Additionally, computer simulations can be used to generate the test dataset.

[0091] Go to Figure 1 This document presents a flowchart illustrating a method for classifying users of personal care devices. It should be noted that users can be individual users or user types (reflecting how a user operates / manipulates / uses the personal care device). In other words, this method can be used to identify specific users of personal care devices or to classify users based on how they use the devices.

[0092] This method relates to a personal care device including an actuator that causes the personal care device to vibrate during use. The personal care device can be a device including an actuator that performs movement, causing a vibrational response in the personal care device. Furthermore, the personal care device can vibrate, but the vibration may not be part of its function. Essentially, the personal care device includes an actuator that can induce mechanical action, causing (measurable) vibration of the personal care device, whether for the personal care function of the device or for other purposes. Nevertheless, in some cases, the personal care device can be a vibrating personal care device that vibrates to perform a function.

[0093] Personal care devices can be, for example, electric toothbrushes, razors, skin scrubbers, etc.

[0094] In step 110, a dataset describing the operating parameters of the personal care device during vibration is obtained.

[0095] The dataset may include one or more of the following: accelerometer data describing acceleration on one axis of the personal care device, accelerometer data describing acceleration on three axes of the personal care device, acoustic data describing the sound generated by the personal care device, and / or current data describing the motor current of the actuator. Other operating parameters may also be considered for capturing characteristics indicative of the vibration of the personal care device. However, the above-mentioned operating parameters may be particularly well-suited for measuring the vibration of personal care devices.

[0096] In some cases, the obtained dataset may include values ​​describing the operating parameters of the personal care device, spanning up to one second. Of course, values ​​spanning a larger time range can be obtained (and subsequently used to generate spectrogram images), but one second may be sufficient to provide a dataset with extractable vibration-related features for predicting users and / or user types. Furthermore, the obtained dataset may include values ​​describing the operating parameters of the personal care device sampled at a rate of at least 800 Hz to clearly capture vibration-related features.

[0097] Step 110 (obtaining the dataset) can be performed in response to the personal care device being in a specific context. More specifically, step 110 can be performed when the personal care device is in a specific location or is used in a specific manner. For example, a dataset can be obtained when a toothbrush (as a personal care device) comes into contact with a specific tooth. This allows for a more refined / precise classification of user types and a more accurate identification of users.

[0098] Step 110 may also include (optional) sub-steps for preprocessing the dataset. For example, the dataset may be normalized to account for any variations between similar types of personal care devices (e.g., material differences, minor malfunctions, different environmental conditions). This could include, for example, changing all values ​​in the dataset so that they are referenced between 0 and 1 or between -1 and 1.

[0099] Furthermore, in cases where the dataset includes accelerometer data describing acceleration along three axes, preprocessing may involve calculating (using known methods) norm vector data describing the absolute vector length of acceleration using the accelerometer data. This norm vector data can then be concatenated with serialized accelerometer data (in each of the three axes) to generate a dataset consisting of a single data segment containing accelerometer data and norm vector data for each of the three axes.

[0100] Step 110 (obtaining the dataset) can be performed in response to user manipulation of the personal care device. That is, the dataset can be obtained when the personal care device is subjected to load (i.e., is being used to perform its function). This can indicate that the vibration response of the personal care device depends at least in part on the user.

[0101] Furthermore, step 110 can be performed during a single operating mode of the actuator. In other words, the dataset can be obtained in response to / during the normal / regular vibration of the personal care device. Normal vibration can correspond to the vibration patterns that the personal care device typically experiences during the performance of its personal care functions. Unlike frequency offset methods, this eliminates the need for additional complex actuation to the actuator of the personal care device.

[0102] Conversely, step 110 can be performed simultaneously by changing the operating mode of the actuator to alter the amplitude and / or duty cycle of the actuated vibration. This can be considered a vibration amplitude scan, which can enrich the dataset with additional unique features for predicting users and / or user types.

[0103] Then, in step 120, a spectrogram image is generated based on / using at least a portion of the dataset. A variety of known methods (both analog and digital) can be used to generate the spectrogram image. Most suitable may be to generate the spectrogram by applying multiple FFTs to the dataset.

[0104] More specifically, each FFT is applied to the dataset at different time points captured by the dataset to generate spectrogram data. Then, based on the spectrogram data, an image representing the spectrogram data can be generated.

[0105] Therefore, a spectrogram image containing the spectrum present in the dataset over the time spanned by the dataset was obtained.

[0106] In step 130, the spectrogram image is provided to a user classification machine learning algorithm. The user classification machine learning algorithm can be any AI-based algorithm (e.g., a neural network) configured to process the spectrogram image to identify and classify features in order to identify a user or determine the user type of a user.

[0107] The spectrogram image is processed using a user classification machine learning algorithm. Accordingly, the predicted user and / or user type is determined.

[0108] Predicting the user and / or user type of a personal care device may include: using a user classification machine learning algorithm to classify the user of the personal care device into one of a plurality of users and / or into one of a plurality of user type categories based on a spectrogram image. Essentially, the user classification machine learning algorithm may identify features in the spectrogram that identify a user and / or indicate that the user belongs to one of the categories, and classify the user based on said features.

[0109] The user classification machine learning algorithm can be a spectral image classifier, such as a CNN classifier or an FCNN classifier. In the case of a CNN classifier, it can include 3 to 6 convolutional layers.

[0110] In some cases, the method may also include generating a user classification machine learning algorithm. This may also include generating multiple test datasets that describe the operating parameters of the personal care device during vibration when different, corresponding, known users manipulate the device. Each test dataset may be generated during user use or may be synthesized. Synthetically generated data may include data generated using a robotic test setup or computer-simulated vibration response data.

[0111] Then, a user classification machine learning algorithm is trained using a training algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include a test dataset from the test dataset, and the corresponding known outputs include known users.

[0112] continue, Figure 2 A personal care device 200 according to an example embodiment is depicted. In this case, an electric toothbrush is shown. However, it is worth noting that this is merely for illustrative purposes, and the personal care device 200 can be any type of device for personal care (not just oral care).

[0113] The personal care device 200 includes an actuator 210, a sensor 220, and a processor 230, the actuator being configured to cause the personal care device 200 to vibrate during use.

[0114] Sensor 220 acquires a dataset describing the operating parameters of the personal care device 200 during vibration. The dataset can be any of the datasets described above and acquired using any of the methods described above. While sensor 220 is described herein as being provided as part of the personal care device 200, sensor 220 may alternatively be provided as part of an external device. For example, sensor 220 may be a microphone mounted on a charger base / stand. Sensor 220 may also be a load sensor that measures the force exerted by the surface of the personal care device on the user's surface. Alternatively or additionally, sensor 220 may retrieve previously stored datasets from memory and provide said datasets to processor 230.

[0115] Processor 230 is configured to generate spectrogram images based on at least a portion of a dataset, and to process the spectrogram images using a user classification machine learning algorithm to predict the user and / or user type of the personal care device 200. Processor 230 can be configured to perform the above-mentioned... Figure 1 Any method steps described.

[0116] Similar to a sensor, although processor 230 is depicted as part of the personal care device 200, it can also be located externally to the personal care device 200. For example, processor 230 can wirelessly connect to sensor 220 to receive data sets. Processor 230 can be provided in the form of a smartphone, or processor 230 can be implemented in the cloud or other external processing services.

[0117] The resulting predicted user and / or user type can be used to provide outputs indicating the user and / or user type. Such outputs can be used to correlate other data generated during the use of the personal care device with a specific user. Alternatively or additionally, the predicted user and / or user type can be used to modify the operating parameters of the personal care device 200 to improve the effectiveness of the personal care device. Other uses of the predicted user and / or user type are obvious to technicians.

[0118] Therefore, the present invention can promote an improved user experience and the provision of improved personal care functions.

[0119] Figure 3An example of a computer 300 in which one or more portions of an embodiment may be employed is illustrated. The various operations discussed above can utilize the capabilities of computer 300. For example, one or more portions of a system for controlling a handheld device can be incorporated into any of the elements, modules, applications, and / or components discussed herein. In this regard, it should be understood that system functional blocks can run on a single computer or can be distributed across several computers and locations (e.g., via an Internet connection), such as cloud-based computing infrastructure.

[0120] Computer 300 includes, but is not limited to, PCs, workstations, laptops, PDAs, handheld devices, servers, memory, microcontroller units, integrated processors, accelerators, etc. Generally, in terms of hardware architecture, computer 300 may include one or more processors 310, memory 320, and one or more I / O devices 330 communicatively coupled via a local interface (not shown). The local interface may be, for example, but not limited to, one or more buses or other wired or wireless connections, as known in the art. The local interface may have additional elements, such as controllers, buffers (caches), drivers, repeaters, and receivers, to enable communication. Furthermore, the local interface may include address, control, and / or data connections to enable appropriate communication between the aforementioned components.

[0121] Processor 310 is a hardware device for executing software that can be stored in memory 320. Processor 310 can actually be any custom or commercially available processor, central processing unit (CPU), digital signal processor (DSP), tensor processing unit (TSP) designed specifically for neural processing, dedicated AI accelerator / processing unit, or auxiliary processor among several processors associated with computer 300, and processor 310 can be a semiconductor-based microprocessor (in the form of a microchip) or microprocessor.

[0122] Memory 320 may include any or a combination of volatile memory elements (e.g., random access memory (RAM), such as dynamic random access memory (DRAM), static random access memory (SRAM), etc.) and non-volatile memory elements (e.g., ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic tape, optical disc read-only memory (CD-ROM), magnetic disk, floppy disk, cassette tape, etc.). Furthermore, memory 320 may incorporate electronic, magnetic, optical, and / or other types of storage media. Note that memory 320 may have a distributed architecture, where various components are geographically separated but accessible by processor 310.

[0123] The software in memory 320 may include one or more individual programs, each including an ordered list of executable instructions for implementing logical functions. According to an exemplary embodiment, the software in memory 320 includes a suitable operating system (O / S) 340, a compiler 360, source code 350, and one or more application programs 370. As shown, the application program 370 includes numerous functional components for implementing the features and operations of the exemplary embodiment. According to the exemplary embodiment, the application program 370 of computer 300 may represent various applications, computing units, logic, functional units, processes, operations, virtual entities, and / or modules, but the application program 370 is not intended to be limiting.

[0124] Operating system 340 controls the execution of other computer programs and provides scheduling, input / output control, file and data management, memory management, communication control, and related services. The inventors anticipate that application program 370 for implementing exemplary embodiments can be applied to all commercially available operating systems.

[0125] Application 370 can be a source program, an executable program (object code), a script, or any other entity including a set of instructions to be executed. When it is a source program, the program is typically transformed by a compiler (e.g., compiler 360), assembler, interpreter, etc., which may or may not be included in memory 320 to operate correctly in conjunction with O / S 340. Furthermore, application 370 can be written in an object-oriented programming language (which has classes of data and methods) or a procedural programming language (which has routines, subroutines, and / or functions), such as, but not limited to, C, C++, C#, Pascal, BASIC, API calls, HTML, XHTML, XML, ASP scripts, JavaScript, FORTRAN, COBOL, Perl, Java, ADA, .NET, etc.

[0126] I / O device 330 may include input devices, such as, but not limited to, a mouse, keyboard, scanner, microphone, camera, etc. Furthermore, I / O device 330 may also include output devices, such as, but not limited to, a printer, monitor, etc. Finally, I / O device 330 may also include devices that serve as both communication input and communication output, such as, but not limited to, a NIC or modulator / demodulator (for accessing remote devices, other files, devices, systems, or networks), a radio frequency (RF) transceiver or other transceiver, a telephone interface, a bridge, router, etc. I / O device 330 also includes components for communication over various networks, such as the Internet or intranets.

[0127] If the computer 300 is a PC, workstation, intelligent device, etc., the software in the memory 320 may also include a Basic Input / Output System (BIOS) (omitted for simplicity). The BIOS is a set of basic software routines that initialize and test the hardware at startup, boot the O / S 340, and support data transfer between hardware devices. The BIOS is stored in some type of read-only memory, such as ROM, PROM, EPROM, EEPROM, etc., so that the BIOS can be executed when the computer 300 is activated.

[0128] When the computer 300 is running, the processor 310 is configured to execute software stored in the memory 320, transfer data to and from the memory 320, and generally control the operation of the computer 300 according to the software. Application programs 370 and operating systems 340 are read, in whole or in part, by the processor 310, may be buffered within the processor 310, and then executed.

[0129] When application 370 is implemented in software, it should be noted that application 370 can be stored on virtually any computer-readable medium for use by or in connection with any computer-related system or method. In the context of this document, a computer-readable medium can be an electronic, magnetic, optical, or other physical device or apparatus that can contain or store computer programs for use by or in connection with a computer-related system or method.

[0130] Application 370 may be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a processor-containing system, or other system that can fetch and execute instructions from and from the instruction execution system, apparatus, or device. In the context of this document, “computer-readable medium” can be any means that can store, communicate, propagate, or transmit a program for use by or in connection with that instruction execution system, apparatus, or device. Computer-readable media can be, for example, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, devices, or propagation media.

[0131] Figure 1 The proposed control methods (one or more) shown and Figure 2The systems (one or more) shown can be implemented in hardware or software, or a combination of both (e.g., as firmware running on a hardware device). With respect to the embodiment being implemented partially or entirely in software, the functional steps shown in the process flowchart can be executed by appropriately programmed physical computing devices, such as one or more central processing units (CPUs) or graphics processing units (GPUs). Each process—and its constituent steps shown in the flowchart—can be executed by the same or different computing devices. According to an embodiment, a computer-readable storage medium stores a computer program comprising computer program code configured to cause one or more physical computing devices to perform the control methods described above when the program is run on one or more physical computing devices.

[0132] Storage media can include volatile and non-volatile computer memories, such as RAM, PROM, EPROM and EEPROM, optical discs (such as CD, DVD, BD), and magnetic storage media (such as hard disks and magnetic tapes). Various storage media can be fixed within a computing device or can be removable, allowing one or more programs stored thereon to be loaded into a processor.

[0133] Regarding the implementation of the embodiments in part or in whole in hardware, Figure 2 Some blocks shown in the block diagram may be individual physical components or logical partitions of a single physical component, or they may all be implemented in an integrated manner within a single physical component. The functionality of a block shown in the figures may be divided among multiple components in an implementation, or the functionality of multiple blocks shown in the figures may be combined within a single component in an implementation. Hardware components suitable for embodiments of the present invention include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs). One or more blocks may be implemented as a combination of dedicated hardware performing some functions and one or more programmable microprocessors and associated circuitry performing other functions.

[0134] Based on a study of the accompanying drawings, the disclosure, and the appended claims, those skilled in the art can understand and implement variations of the disclosed embodiments in practicing the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude multiple. A single processor or other unit can perform the functions of several items recited in the claims. The fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used advantageously. If a computer program is discussed above, it can be stored / distributed on a suitable medium, such as an optical storage medium or solid-state medium provided with or as part of other hardware, but it can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. If the term "suitable" is used in the claims or description, it should be noted that the term "suitable" is intended to be equivalent to the term "configured as." No reference numerals in the claims should be construed as limiting the scope.

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a portion of a module, segment, or instruction, comprising one or more executable instructions for implementing a specified logical function(s). In some alternative implementations, the functions marked in the blocks may occur in a different order than indicated in the drawings. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order, depending on the functionality involved. It will also be noted that each block in the block diagram and / or flowchart illustrations, and combinations of blocks in the block diagram and / or flowchart illustrations, may be implemented by a system based on dedicated hardware that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

Claims

1. A method for classifying users of a personal care device, the personal care device including an actuator that causes the personal care device to vibrate during use, the method comprising: (110) Obtain a dataset describing the operating parameters of the personal care device during vibration of the personal care device when the user manipulates the personal care device; Generate (120) spectrogram images based on at least a portion of the dataset; as well as The spectrogram image is processed using a user classification machine learning algorithm to predict (130) the user and / or user type of the personal care device.

2. The method according to claim 1, wherein the user classification machine learning algorithm is a spectral image classifier, and optionally, wherein the user classification machine learning algorithm is a CNN classifier or an FCNN classifier.

3. The method of claim 2, wherein the user classification machine learning algorithm is a CNN classifier comprising 3 to 6 convolutional layers.

4. The method according to any one of the preceding claims, wherein the dataset comprises accelerometer data describing acceleration on at least one axis of the personal care device.

5. The method of claim 4, wherein the accelerometer data describes acceleration on three axes, and wherein the method further comprises: Based on the accelerometer data, a norm vector data describing the absolute vector length of acceleration is calculated; as well as The accelerometer data and norm vector data for each of the three axes are serialized into a single data segment.

6. The method according to any one of the preceding claims, wherein the dataset includes sound data describing sounds generated by the personal care device.

7. The method according to any one of the preceding claims, wherein the dataset includes current data describing the motor current of the actuator.

8. The method according to any one of the preceding claims, wherein the obtained dataset includes values ​​describing the operating parameters of the personal care device, the values ​​spanning up to one second, and wherein the obtained dataset includes values ​​describing the operating parameters of the personal care device sampled at a rate of at least 800 Hz and optionally at a rate of at least 1.6 kHz.

9. The method according to any one of the preceding claims further includes normalizing at least a portion of the dataset.

10. The method according to any one of the preceding claims, wherein the dataset describing the operating parameters is obtained in response to the personal care device being operated by the user (110).

11. The method of claim 10, wherein the dataset describing the operating parameters is obtained during a single operating mode of the actuator (110), or wherein the dataset describing the operating parameters is obtained when the operating mode of the actuator is changed to change the amplitude and / or duty cycle of the actuated vibration (110).

12. The method according to any one of the preceding claims, wherein predicting (130) the user and / or user type comprises: The user classification machine learning algorithm is used to classify the user of the personal care device into one user among multiple users and / or into one user type category among multiple user type categories based on the spectrogram image; and / or The user classification machine learning algorithm generates probability scores based on the spectrogram image, wherein the probability scores describe the predicted user of the personal care device among multiple user categories.

13. The method according to any one of the preceding claims further includes generating the user classification machine learning algorithm, comprising: Multiple test datasets are generated, which describe the operating parameters of the personal care device during vibration of the personal care device when different corresponding known users manipulate the personal care device. The user classification machine learning algorithm is trained using a training algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training inputs include a test dataset from the test dataset, and wherein the corresponding known outputs include the known users. Optionally, each test dataset includes data generated when the user manipulates the personal care device, and / or synthetically generated data.

14. A computer program comprising computer program code means, wherein when the computer program is run on a computer, the computer program code means is adapted to perform the method according to any one of claims 1 to 13.

15. A personal care device, comprising: An actuator (210) is configured to cause the personal care device to vibrate during use; Sensor (220), the sensor being used to obtain a dataset describing the operating parameters of the personal care device during vibration of the personal care device when the user manipulates the personal care device; Processor (230), the processor being configured to: Generate a spectrogram image based on at least a portion of the dataset; as well as The spectrogram image is processed using a user classification machine learning algorithm to predict the user and / or user type of the personal care device.

Images