Method of estimating filter life

Abstract

There is provided a method of estimating life of a particulate filter installed in a device, the method comprising: retrieving a first set of data indicative of motor performance of a motor for generating airflow through the particulate filter for a time period since the particulate filter is installed in the device; retrieving a second data set indicative of an amount of particulate in the air flow over a time period since the particulate filter is installed in the device; an estimate of life of the particulate filter is generated using the first data set and the second data set.

Description

Technical Field

[0001] The present invention relates to a method of estimating filter life for use with an apparatus having a particle filter. Background Art

[0002] Particle filters are known for use in applications such as air purification. An air purifier may draw in ambient air, direct the air through a suitable particle filter to remove particles carried by the air, and release the filtered air from the air purifier.

[0003] Particle filters have a limited design life, after which filter loading may adversely affect performance. Filter loading refers to the accumulation of particles retained in the particle filter, i.e. particles that the particle filter has filtered out, and which eventually degrades performance. In this case, performance can be restored by replacing the particle filter, e.g. with a clean particle filter or a new replacement filter.

[0004] However, an a priori estimate of when a particle filter will reach the end of its design life can be difficult, as the design life depends on the conditions under which a given particle filter is used. These conditions can vary significantly, especially for domestic applications, due to a wide range of air quality and usage scenarios. This can make a blanket prediction of filter life based on run-time hours unsatisfactorily inaccurate. However, an accurate estimate of filter life will ensure that product operation is optimized and the user can be confident that the air purifier is cleaning the ambient air in a cost-effective and resource-efficient manner.

[0005] The present invention has been designed in view of the above considerations. Summary of the invention

[0006] Increased accuracy in estimating filter life can be achieved using a pressure sensor configured to monitor the pressure drop across the particle filter. However, adding a pressure sensor increases cost and complexity. This may be undesirable because it will mean an increase in the upfront cost of the filter device, thereby at least partially offsetting any reduction in the operating cost of the filter device. In addition, additional maintenance may be required.

[0007] Another alternative to estimating filter life can utilize monitoring of the performance of the motor that generates the air flow through the particle filter. For example, the current draw of the motor can be monitored as part of the control process, and as the particle filter becomes clogged, the current draw will decrease over time for a given setting. However, current draw is also affected by the surrounding atmospheric conditions, such as temperature, humidity, and barometric pressure. Changes in atmospheric conditions may affect motor performance to a similar level as filter loading.

[0008] Another alternative to estimating filter life could be to utilise a particle sensor as this will give a measure of the air quality that the filter is operating in. However, the correlation between particle count and filter life is complex and depends on a range of additional factors such as temperature, humidity, aerosol material (oil, solid), charge state, size distribution (both instantaneous and over time), meaning that data from a particle counter alone may only provide an estimate of the particulate matter removed by a filter.

[0009] The inventors have devised a method for combining data sets from two sensors in an air purification device, which monitor motor performance and particle counts, respectively, to more accurately infer the current state of the filter. By combining the two data sets, a higher accuracy filter life / state prediction can be produced. In addition, this improved method can be implemented without additional hardware costs, as some examples of known air purification devices may already be provided with sensors for monitoring motor performance and for monitoring particle counts.

[0010] According to a first aspect of the invention, there is provided a method for estimating the life of a particle filter installed in a device, the method comprising: retrieving a first data set indicating a motor performance of a motor used to generate an air flow through the particle filter over a period of time since the particle filter was installed in the device; retrieving a second data set indicating an amount of particles in the air flow over a period of time since the particle filter was installed in the device; and using the first data set and the second data set to generate an estimate of the life of the particle filter.

[0011] By utilizing the first data set and the second data set to generate an estimate of the life of the particulate filter, the accuracy of the estimate may be improved without relying on a pressure sensor.

[0012] The method may also include determining that the estimate of filter life has reached the design life; and providing an indication to a user that the life has been reached.

[0013] The time period since the particle filter was installed in the device may include the entire time since the particle filter was installed in the device, but the first data set and the second data set may contain data corresponding to one or more time lengths since the particle filter was installed. For example, during a time period when the device is turned off, no data may be recorded. Similarly, recording of the first data set and the second data set may not begin immediately with the installation of the particle filter because, in some examples, the device is turned off in order to replace the particle filter. However, since there is no air flow through the newly installed particle filter in this case, this time period from installation to the first operation after installation does not affect the life of the particle filter.

[0014] The method may also include obtaining a subset of the first data set, the subset indicating motor performance with a particulate filter installed in the device; and using the subset indicating motor performance with the particulate filter installed when generating an estimate of life of the particulate filter.

[0015] By recording a subset of the first data set indicative of motor performance when installed, a baseline value of motor performance can be established. Comparison with said baseline value of motor performance can further improve the accuracy of the filter life estimate. For example, the effect of manufacturing tolerances can be reduced thereby.

[0016] The method may also include recording data indicative of motor performance when a replacement filter is installed in the device.

[0017] By recording data indicative of motor performance when a replacement filter is installed, the effect of machine wear on filter life estimates can be reduced because the effect of machine wear on baseline values ​​and subsequent motor performance is comparable throughout the life of a given filter.

[0018] The method may also include calculating an estimate of air flow through the particulate filter during the time interval based on at least a subset of a first data set corresponding to the time interval; calculating an estimate of the number of particles ingested by the device during the time interval based on the estimate of air flow and at least a subset of a second data set corresponding to the time interval; and using the estimate of the number of particles ingested by the device when generating an estimate of the life of the particulate filter.

[0019] By using an estimate of the number of ingested particles in generating an estimate of the life of the particulate filter, the accuracy of the estimate of the life of the particulate filter may be improved.

[0020] The particle population estimate for the time interval may be calculated based on at least a subset of the second data set corresponding to the time interval, for example as a time average or integral.

[0021] The method may further comprise applying a low pass filter to the first data set.

[0022] The first data set can include data indicative of motor performance at a relatively high frequency (e.g., at least once per minute during operation), resulting in the first data set capturing short-term changes in motor performance. Changes in motor performance due to atmospheric conditions may also occur on relatively short time scales, such as days or weeks, and are therefore captured in the first data set. In contrast, filter loading may occur on relatively long time scales, such as months to years. Therefore, applying low pass filtering to the first data set can reduce the impact of short-term trends and help identify long-term trends that are considered more likely to be related to filter loading.

[0023] The method may also include using the first and second data sets to generate an estimate of the life of the particulate filter by using a lookup table to determine the estimate of the life. The lookup table may be generated using previously obtained experimental data, such as experimental data showing how the actual life of the particulate filter varies with the experimentally obtained first and second data sets.

[0024] The method may also include using the first data set and the second data set to generate an estimate of the life of the particulate filter by determining the estimate of the life using a trained machine learning model that has been trained with labeled training data. The labeled training data may include labeled pairs of the first and second data sets (e.g., discussed in more detail below).

[0025] The trained machine learning model can be a Gaussian process regression model.

[0026] The method may also include retrieving a temperature dataset indicating an ambient temperature of the device since the particulate filter was installed in the device; and using the temperature dataset when generating the estimate of the life of the particulate filter.

[0027] The method may also include retrieving a barometric pressure dataset indicating ambient barometric pressure of the device since the particulate filter was installed in the device; and using the barometric pressure dataset when generating the estimate of the life of the particulate filter.

[0028] The method may further include retrieving a humidity dataset indicating ambient humidity of the device since the particulate filter was installed in the device; and using the humidity dataset when generating the estimate of the life of the particulate filter.

[0029] By tracking changes in environmental conditions, the effect of environmental conditions on motor performance can be taken into account when generating an estimate of filter life. Such environmental conditions can include temperature, atmospheric pressure, and humidity.

[0030] The method may also include generating a second data set using a sensor configured to detect particles having a diameter of 4.5 microns or less.

[0031] A sensor that detects particles may be configured to detect particles of 4.5 microns or smaller; may be configured to detect particles of 3.5 microns or smaller; or may be configured to detect particles of 2.5 microns or smaller.

[0032] By utilizing a sensor to detect relatively small sized particles, such as particles having a diameter of 4.5 microns or less, the accuracy of the filter life estimate may be improved. In contrast, a particle sensor configured to detect relatively large particles (e.g., 10 microns or more) may provide a relatively inaccurate estimate of the particles suspended in the ambient air, and thus result in a reduced accuracy of an estimate based thereon. This may be considered surprising, as a particle sensor that detects particles having a diameter of 10 microns or less may be considered to provide a better sample than a particle sensor that detects particles having a diameter of 2.5 microns or less. However, the inventors have discovered that a better estimate of the number of particles can be derived from smaller particle sizes.

[0033] An air purification system may be provided, comprising: an air purification device comprising a particle filter, a motor, a first sensor system configured to acquire a first data set, and a second sensor system configured to acquire a second data set; and the air purification system is configured to perform the method as described above.

[0034] The motor may be configured to drive the rotor at a fixed speed, and the first data set indicative of motor performance corresponds to power consumption of the motor.

[0035] The air purification device may be configured to perform the method as described above. Alternatively, the air purification system may include a remote server configured to perform the method as described above.

[0036] According to another aspect of the present invention, a method for training a machine learning model for the method as described above is provided, comprising: obtaining multiple pairs of first and second data sets, wherein each pair of first and second data sets comprises: a first data set indicating motor performance of a motor used to generate an air flow through a particulate filter over a period of time since the particulate filter was installed in the device; a second data set indicating the amount of particles in the air flow over a period of time since the particulate filter was installed in the device; generating labeled training data by labeling each pair of first and second data sets to correlate those first and second data sets with an actual life of the particulate filter determined with respect to the particulate filter used to obtain the pair of first and second data sets; and using the labeled training data to train a machine learning model for calculating an estimated life of the particulate filter based on a pair of first and second data sets.

[0037] The method of training a machine learning model may include obtaining a further labeled data set for training the machine learning model. The further labeled data set may correspond to a data set indicating ambient temperature, humidity, or atmospheric pressure.

[0038] The present invention includes any combination of the described aspects and preferred features unless such a combination is expressly impermissible or explicitly avoided. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Embodiments and experiments illustrating the principles of the present invention will now be discussed with reference to the accompanying drawings, in which:

[0040] Figure 1 It is a schematic diagram of an air purification device.

[0041] Figure 2 is another schematic diagram of an air purification device, also showing certain internal components.

[0042] Figure 3 is a schematic diagram of an air purification system including an air purification device and a remote server.

[0043] Figure 4 A method for estimating the life of a particle filter of an air purification device is shown.

[0044] Figure 5 A schematic diagram for generating a filter life estimate based on motor performance is shown.

[0045] Figure 6 A schematic diagram for generating a filter life estimate based on particle count is shown.

[0046] Figure 7 Shows the combination Figure 5 and Figure 6 Schematic diagram of the overall estimate of the estimate.

[0047] Figure 8 is a schematic diagram of generating a particulate filter life estimate based on a single model.

[0048] Fig. 9 A method for training a machine learning model for estimating particulate filter life is shown. DETAILED DESCRIPTION

[0049] Various aspects and embodiments of the present invention will now be discussed with reference to the accompanying drawings. Further aspects and embodiments will be apparent to those skilled in the art.

[0050] Figure 1 and Figure 2 An exemplary air purification device 100 configured to perform a method according to the present invention is shown. In use, the device 100 inhales ambient air and discharges purified air by generating an air flow 110 through a particle filter 120. The particle filter 120 is configured to retain particles 1000 carried in the air flow 110, such as dust and other pollutants. Suitably, the device 100 includes a motor 130 and a rotor 140 driven by the motor 130 to generate the air flow 110. In combination, the motor 130 and the rotor 140 may also be referred to as a compressor. In this example, the particle filter 120 is located upstream of the motor 130 and the rotor 140.

[0051] Device 100 includes a first sensor system 150 , a second sensor system 160 and a third sensor system 170 .

[0052] The first sensor system 150 is configured to generate a first data set indicating the motor performance of the motor 130 when generating the air flow 110 through the particle filter 120. The first data set is a measurement of the efficiency of generating the air flow 110 over a period of time since the particle filter 120 was installed in the device 100. The first sensor system 150 includes at least one sensor. In this example, the first sensor system 150 is a single sensor arranged to monitor the power consumption, current consumption of the motor 130.

[0053] The second sensor system 160 is configured to generate a second data set indicating the number of particles 1000 in the air flow 110 over a period of time since the particle filter 120 was installed in the device 100. In this example, the second sensor system 160 is a single sensor arranged to sample the particles 1000 in the air flow 110. More specifically, the second sensor system 160 is a PM2.5 sensor configured to detect particles having a diameter of 2.5 microns or less.

[0054] The third sensor system 170 is configured to generate a third data set. The third data set includes a temperature data set indicating the ambient temperature of the device 100 during a period of time since the particle filter 120 was installed in the device 100. Suitably, the third sensor system 170 includes a temperature sensor. In addition, the third data set includes a humidity data set indicating the ambient humidity of the device 100 during a period of time since the particle filter 120 was installed in the device 100. Suitably, the third sensor system 170 includes a humidity sensor.

[0055] Figure 3 An air purification system 10 is shown. The air purification system 10 includes an apparatus 100 and a remote server 200. In this example, a method of estimating the life of a particle filter 120 is performed by the remote server 200 based on a first data set and a second data set generated by the apparatus 100 and transmitted to the remote server 200.

[0056] The remote server 200 also retrieves an atmospheric data set indicating the ambient atmospheric pressure of the device 100 since the installation of the particulate filter 120. In this example, the atmospheric data set is not obtained by an onboard sensor of the device 100, but is obtained using the location information of the device 100 and retrieved by the remote server 200 from a suitable source (e.g., an online repository of weather data). In this example, the location information of the device 100 is provided directly to the remote server 200 by a smart device of a user of the device 100.

[0057] Figure 4 A method of estimating the life of the particulate filter 120 installed in the device is shown.

[0058] The method comprises a first step S110 of retrieving a first data set indicating the motor performance of the motor 130 when generating the air flow 110 through the particle filter 120 over a period of time since the particle filter 120 was installed in the device 100 .

[0059] The method comprises a second step S120 of retrieving a second data set indicating the amount of particles in the air flow 110 over a period of time since the particle filter 120 was installed in the device 100 .

[0060] It should be noted that the first step S110 and the second step S120 are not performed in any particular order, ie, the first step S110 may be performed before the second step S120, or the second step S120 may be performed before the first step S110, or they may be performed simultaneously.

[0061] In this example, the first and second data sets are collected by the apparatus 100 and transmitted to the remote server 200. The remote server 200 stores these data sets temporarily, for example, and retrieves the stored data sets to generate a lifetime estimate for the particle filter 120 as part of the third step S130 of the method.

[0062] The third step S130 of the method is to use the first and second data sets to generate an estimate of the life of the particle filter 120. The estimate of the life of the particle filter 120 can be generated in various ways, examples of which are discussed below, but before doing so, basic considerations are discussed to provide relevant background.

[0063] Filter life depends on the amount of particulate matter removed from the air by the device 100. Over time, particles clog the particle filter 120 and affect the performance of the particle filter 120. This is manifested as an increase in the pressure drop across the particle filter 120, which in turn is reflected as a change in the motor performance when generating the air flow 110. In this example, the motor 130 is configured to drive the rotor 140 at a fixed operating speed for each setting (e.g., "silent" at low speed, "normal" at medium speed, and "turbo" at high speed). As filter loading occurs, it is observed that the power consumed by the motor 130 at a given operating speed gradually decreases. However, the instantaneous current consumption also depends on environmental conditions, such as local temperature and humidity. The instantaneous current consumption also depends on atmospheric pressure. Although the temperature and humidity of the air are measured locally by the third sensor system 170, the device 100 does not include an atmospheric sensor. The local air density can vary, for example due to natural fluctuations in atmospheric pressure or due to differences in local altitude relative to the sea level of the device 100. Therefore, it may not be possible to determine whether the instantaneous measurement of the current drawn by the motor 130 is mainly affected by local changes in filter loading or atmospheric pressure.

[0064] This makes it difficult to accurately determine the instantaneous state of the filter using current draw alone. While filter loading occurs on a time scale of months to years, local changes in atmospheric conditions may occur on a time scale of days or weeks. Therefore, the first data set may contain fluctuations that are not representative of filter loading. In contrast, from a second data set indicating particles 1000 in the air flow 110, a gradually changing profile can be derived, such as by estimating the total number of particles 1000 over a time period by sampling individually at discrete times. Therefore, making estimates based on a combination of the first data set and the second data set can provide valuable synergies. For example, the first data set and the second data set can be cross-referenced to identify any rapid fluctuations in a filter load estimate based on the first data set.

[0065] Likewise, the filter load estimate based on the second data set can easily be biased by assumptions about aerosol composition. Using the first data set, the filter load as a function of time based on the second data set can be compared to the change in power consumption as a function of time based on the first data set. If the two metrics disagree, the filter load estimate can be adjusted based on the second data set to obtain a more accurate estimate of the filter state.

[0066] As part of the motor control process, a first sensor system 150 monitors the power consumption of the motor 130. A second sensor system 160 samples the particles 1000 to provide feedback to the user on the room air quality and to automatically control the motor settings (e.g., silent / normal / turbo). In this way, these sensor systems and the data collected by the sensor systems are independent of each other. Each of these can be used to infer changes in filter status: as the filter becomes clogged, the current consumption will decrease over time for a given setting; the total number of particles removed from the room can be estimated by the machine on time, settings, and measured particle counts. However, neither of these data sets alone can provide a sufficiently accurate indicator of filter load / status. As mentioned above, the current consumption measurement is affected by environmental conditions (temperature, humidity, barometric pressure) and machine wear. Similarly, particle counts provide a low resolution measurement and are merely an estimate of the particles removed by the particle filter 120. In addition, the relationship between filter pressure rise and particle size / number is complex and affected by environmental conditions.

[0067] The use of PM2.5 measurements can also estimate the amount of material that the filter has removed from the air. This is achieved by integrating the product of the instantaneous particle readings with the air flow 110 through the device 100. In this way, the amount of the total amount of pollutants that the particle filter 120 has been exposed to during the operating time of the particle filter 120 can be estimated. This method is limited by the resolution of the PM2.5 sensor of the second sensor system 160, and the specific assumption that the PM2.5 concentration is related to the overall concentration of all pollutants in the local environment of the device 100. Some household aerosols may be generated, which may not be recognized by the PM2.5 sensor, but still cause filter loading (such as cooking smoke or burnt bread), or distort the overall distribution of particles. In addition, changes in local atmospheric pressure will also affect the filter loading rate and the mass flow through the machine. Filter loading is sensitive to the surface velocity of the aerosol. When the local density changes due to fluctuations in atmospheric pressure, the air velocity through the device 100 changes, thereby adjusting the surface velocity of the aerosol impinging on the particle filter 120. The total amount of air ingested by the device 100 also varies with local variations in atmospheric pressure and corresponding air density. This further increases the uncertainty in filter load predictions based solely on PM2.5 readings and the operating point of the device 100. Therefore, the output of the second sensor system 160 alone is considered insufficient to accurately quantify the actual particulate matter challenging the particle filter 120.

[0068] Combining these two filter life estimation techniques creates a beneficial synergy. The integration method used in the PM2.5 estimation provides a naturally smooth and gradually changing profile. This can be cross-referenced with any rapid fluctuations in the filter load estimates provided by the current loading methods. In this way, fluctuations due to changes in air pressure can be easily identified. Using appropriate statistical methods, the confidence level of the instantaneous readings can be determined, which means that the lag or delay time associated with low-pass filtering can be significantly reduced. If the location of the device changes - the owner moves to a new location at a different altitude - it will be possible to compare the different filter load indicators and identify inconsistencies. Since altitude changes cause continuous changes in local operating conditions, these can be identified by corresponding changes in the two filter load estimates.

[0069] In this example, the method further comprises applying a low pass filtering to the first data set representing the instantaneous current draw to separate the effects of local variations in atmospheric pressure from the long term trends introduced by the particles 1000 loading the particle filter 120. Note that the application of the low pass filtering results in a lag time (or "delay") in obtaining a reliable estimate of the filter load based on the first data set after the low pass filtering is applied. Therefore, an estimate of the filter life based on the first data set after said application may not be accurately extrapolated over a period of days or even months after the particle filter 120 becomes fully loaded. Using appropriate statistical methods, the confidence in the instantaneous readings can be determined, which means that the lag time associated with the low pass filtering can be significantly reduced.

[0070] In this example, the method of estimating the life of the particle filter 120 also includes obtaining a subset of the first data set that represents the motor performance when the particle filter is installed in the device. That is, the first data set includes a subset related to when the particle filter 120 is newly installed and has experienced no or little filter load. Therefore, when generating an estimate of the life of the particle filter, this subset of the first data set can be used as a baseline value or reference value.

[0071] In this example, the method also includes using a lookup table (e.g., trained using experimental data) to calculate an estimate of air flow 110 through the particle filter 120 during the time period based on at least a subset of the first data set corresponding to the time period, and calculating an estimate of the number of particles ingested by the device during the time period based on the estimate of the air flow and at least a subset of the second data set corresponding to the time period. The estimate of the number of particles ingested by the device 100 is used when generating the life estimate of the particle filter 120.

[0072] Figure 5 and Figure 6 Schematic diagrams showing the generation of independent lifetime estimates based on a first data set of motor performance and a second data set of particle counts, respectively.

[0073] The device 100 does not have a differential pressure sensor from which the air flow 110 through the particle filter 120 can be obtained directly. However, by viewing the history generated by the second sensor system 160, the amount of particles 1000 that have been removed can be inferred. In addition, the current drawn by the motor 130 can be used to infer the pressure drop across the particle filter 120. By benchmarking the current consumption of the device 100 when the particle filter 120 is installed, for example, a "clean filter state" is detected. Monitoring current consumption allows the algorithm to estimate the life of the particle filter 120. This may also optionally include altitude, temperature and humidity sensors. Combining this with the PM2.5 log will allow a more accurate measurement of system performance and therefore filter life without the need to directly measure the pressure drop across the filter.

[0074] Figure 7 The overall model that provides the overall filter status is shown based on Figure 5 and Figure 6 The output of produces a schematic diagram of the lifetime estimate. Figure 7 The overall filter status is determined by Figure 5 and Figure 6 The filter state estimation technology provides a life estimate to provide a more accurate inference of filter life.

[0075] The overall filter state can be an instantaneous estimate, utilizing only the most recent outputs from two independent filter state models. Alternatively, the overall filter state can contain a storage block to record a window of historical filter state estimates. In this way, temporal patterns and changes can be used to further enhance the capabilities of the filter state estimate. The selection of an appropriate algorithm or machine learning model for this purpose can be based on the required accuracy, computational cost, and memory footprint. Candidate machine learning models include decision trees, random forests, XGBoost, Gaussian process regression, and neural network methods. For neural architectures, historical information can be explicitly included through multiple input channels, or by implementing a convolutional input layer to encode historical input vectors. Alternatively, a recursive element (RNN) can be used, where LSTM (Long Short-Term Memory) or GRU (Gated Recursive Unit) can be used to enable historical data to influence the instantaneous filter state estimate.

[0076] Figure 8 Schematic diagram of generating a life estimate for the particulate filter 120 based on a single model that takes as input: atmospheric conditions (e.g., temperature and relative humidity); motor performance (e.g., current draw, voltage, and coil temperature); a second data set (e.g., PM2.5 sampling data); a target operating speed for the rotor 140 and a current operating speed for the rotor 140; and any other measurements performed by the device's onboard sensor system; and optionally information provided from an external server.

[0077] Figure 8 A single model can provide a comprehensive estimate of the filter state, establishing a relationship between each data input to provide an optimized function model for the filter state estimate. The selection of an appropriate algorithm or selected machine learning model can be based on the required accuracy, computational cost, and memory footprint. Candidate machine learning models include decision trees, random forests, XGBoost, Gaussian process regression, and neural network methods. For neural architectures, historical information can be explicitly included through multiple input channels, or by implementing a convolutional input layer to encode historical input vectors. Alternatively, a recursive element (RNN) can be used, where LSTM or GRU can be used to enable historical data to influence the instantaneous filter state estimate.

[0078] In some examples, machine learning can be utilized. In such examples utilizing machine learning, generating a life estimate for the particulate filter using the first data set and the second data set can include determining the life estimate using a trained machine learning model that has been trained using labeled pairs of the first data set and the second data set. In some examples, the trained machine learning model can be a Gaussian process regression model, although any suitable machine learning model can be utilized.

[0079] Training a machine learning model can use a dataset of labeled examples. The data can be generated by a series of methods. The particle filter 120 can be loaded over time under representative conditions, with accurate pressure drop measurements taken at regular intervals to obtain accurate filter limit values. Accelerated testing can also be used to generate data, in which the particle filter 120 is challenged with an abnormally high level of filter load, and the actual pressure rise on the particle filter 120 as a function of the filter load and PM2.5 ingested by the device 100 is accurately measured. Another method of collecting data can be based on using a masking of the filter area equivalent to a certain filter restriction level to artificially block the particle filter 120 to a known degree. Collecting experimental data can be very expensive and will limit the amount of data available in a short time frame. Synthetic data can be generated to increase the number of examples and expand the coverage of the data set to include extreme examples of, for example, atmospheric pressure changes. For a given machine tolerance distribution, atmospheric conditions, and filter restrictions, a computer simulation model can be used to simulate the current consumption of the motor 130 to generate synthetic data.

[0080] The model is trained to minimize a given loss function. The loss function is chosen to ensure that at convergence, the algorithm performance is maximized in terms of accuracy in predicting the current filter state and remaining lifetime. When optimizing a machine learning model, different model hyperparameters can be adjusted, as well as evaluating the importance of different input signals to the model.

[0081] The method for training a machine learning model may include obtaining a plurality of pairs of first and second data sets. The pair of data sets is obtained under the same conditions, such as the same device 100. More specifically, each pair of first and second data sets includes: a first data set indicating the motor performance of a motor for generating an air flow through a particle filter during a period of time since the particle filter was installed in the device; and a second data set indicating the amount of particles in the air flow during a period of time since the particle filter was installed in the device.

[0082] The method of training a machine learning model may also include generating labeled training data by labeling each pair of first and second data sets to relate those first and second data sets to an actual life of the particulate filter determined with respect to the particulate filter used to obtain the pair of first and second data sets.

[0083] The method of training a machine learning model also includes training the machine learning model using labeled training data for calculating an estimated life of a particulate filter based on a pair of first and second data sets.

[0084] The first sensor system 150 described above utilizes a single sensor to monitor motor performance. In other examples, multiple sensors may be utilized. These sensors may monitor the same physical quantity, such as the power or current consumed by the motor 130. Alternatively, the sensors may monitor different physical quantities, each indicating motor performance, such as a combination of power and current drawn by the motor 130, the rotational frequency of the rotor 140, etc.

[0085] The second sensor system 160 described above samples the air flow 110 using a single sensor. In other examples, the sensor 160 is not located in the air flow 110, but is located in or at the device 100, so that air indicating particles in the air flow 110 is sampled. In addition, multiple sensors may be provided. Some of these sensors may be located within the device 110, and others may be arranged outside or even external.

[0086] The air purification system 10 described above utilizes the remote server 200 to determine the life of the particle filter 120. According to other examples, this is done at the device 100.

[0087] In the above example, the location information associated with the apparatus 100 is transmitted directly by the user's smart device to the remote server 200. In other examples, the location information may be provided to the apparatus 100 and transmitted by the apparatus 100 to the remote server 200. Furthermore, the smart device is merely an example of a suitable means for obtaining the location information, but alternatively, other means may be utilized, such as an onboard location sensor of the apparatus 100, or the location information may be inferred based on the details of the internet connection of the apparatus 100.

[0088] When filter loading occurs, the above-mentioned device 100 sees a decrease in the power required to maintain the mass air flow. In other examples, different rotor arrangements may result in increased power consumption. In other examples, the air flow may be generated according to different parameters, such as at a fixed power. In this case, since the filter load and the rotational frequency of the rotor may change, the mass air flow may decrease over time. Therefore, in this case, the motor performance can be monitored by the rotational frequency (e.g., revolutions per minute) rather than the power consumption.

[0089] The method of estimating filter life as described above allows the operation of the necessary control methods and algorithms to determine an improved estimate of filter life. For example, this can be achieved by calculating the pressure drop across the filter and the stage in the filter life. Once the filter has removed contaminants representative of its design life, the device can highlight to the user the need to replace or maintain the filter.

[0090] The features disclosed in the preceding description, or in the following claims, or in the accompanying drawings, expressed in their specific form or as means for performing the disclosed functions, or as methods or processes for obtaining the disclosed results, may, where appropriate, be used alone or in any combination of these features to implement the invention in its different forms.

[0091] Although the present invention has been described in conjunction with the above exemplary embodiments, many equivalent modifications and variations will be apparent to those skilled in the art when this disclosure is given. Therefore, the above exemplary embodiments of the present invention are considered to be illustrative rather than restrictive. Various changes may be made to the described embodiments without departing from the spirit and scope of the present invention.

[0092] For the avoidance of any doubt, any theoretical explanations provided herein are intended to enhance the reader's understanding. The inventors do not wish to be bound by any of these theoretical explanations.

[0093] Any section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0094] Throughout the specification, including the following claims, unless the context requires otherwise, the words "comprises" and "comprising" and variations will be understood to imply the inclusion of stated integers or steps or groups of integers or steps but not the exclusion of any other integers or steps or groups of integers or steps.

[0095] It must be noted that, as used in the specification and the appended claims, the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from "about" one particular value, and / or to "about" another particular value. When such a range is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when a value is expressed as an approximation by using the antecedent "about," it will be understood that the particular value forms another embodiment. The term "about" in relation to a value is optional and refers to, for example, + / - 10%.

Claims

1. A method for estimating the life of a particle filter installed in a device, the method include: retrieving a first data set indicating motor performance of a motor for generating air flow through a particulate filter over a period of time since the particulate filter was installed in the device; retrieving a second data set indicating an amount of particulates in the air stream over a period of time since a particulate filter was installed in the device; An estimate of the life of the particulate filter is generated using the first data set and the second data set.

2. The method according to claim 1, further comprising: include: determining that the estimate of the life of the filter has reached the design life; and Provides an indication to the user that end of life has been reached.

3. A method according to any preceding claim, further comprising: include: obtaining a subset of the first data set, the subset indicating motor performance when the particulate filter is installed in the device; and When generating an estimate of the life of the particulate filter, a subset indicative of the motor performance when the particulate filter is installed is used.

4. The method according to claim 3, further comprising: include: Data indicative of motor performance with a replacement filter installed in the device is recorded.

5. A method according to any preceding claim, further comprising: include: calculating an estimate of air flow through the particulate filter during the time period based on at least a subset of the first data set corresponding to the time period; calculating an estimate of the number of particles ingested by the device during the time period based on the estimate of air flow and at least a subset of the second data set corresponding to the time period; and When generating an estimate of the life of a particulate filter, an estimate of the number of particles ingested by the device is used.

6. A method according to any preceding claim, in, The method further comprises applying a low pass filter to the first data set.

7. A method according to any preceding claim, in, Using the first data set and the second data set to generate an estimate of the life of the particulate filter includes: An estimate of lifespan is determined using a trained machine learning model that has been trained using the labeled first data set and the labeled second data set.

8. The method according to claim 7, in, The trained machine learning model is a Gaussian process regression model.

9. A method according to any preceding claim, further comprising: include: retrieving a temperature data set indicating an ambient temperature of the device since the particulate filter was installed in the device; and The temperature data set is used when generating an estimate of the life of the particulate filter.

10. The method according to claim 1, further comprising: include: retrieving an atmospheric pressure data set indicating ambient atmospheric pressure of the device since the particulate filter was installed in the device; and When generating an estimate of the life of the particulate filter, the atmospheric pressure data set is used.

11. A method according to any preceding claim, further comprising: include: retrieving a humidity data set indicating an ambient humidity of the device since the particulate filter was installed in the device; and The humidity data set is used when generating an estimate of the life of the particulate filter.

12. A method according to any preceding claim, further comprising: include: The second data set was generated using a sensor configured to detect particles having a diameter of 4.5 microns or less.

13. An air purification system, include: An air purification device comprising a particle filter, a motor, a first sensor system configured to acquire a first data set, and a second sensor system configured to acquire a second data set; and The air purification system is configured to perform a method according to any preceding claim.

14. The air purification system according to claim 13, in, The air purification device is configured to perform a method according to any preceding claim.

15. The air purification system according to claim 13 or 14, in, The motor is configured to drive a rotor at a fixed speed, and the first data set indicative of motor performance corresponds to power consumption of the motor.

16. The air purification system according to claim 13, further comprising a remote server, wherein the remote server is configured to perform the method according to any one of claims 1 to 12.

17. A method of training a machine learning model for use in the method according to any one of claims 1 to 12, in, The method includes: A plurality of pairs of first data sets and second data sets are obtained, wherein each pair of first and second data sets comprises: a first data set indicating motor performance of a motor for generating air flow through a particulate filter over a period of time since the particulate filter was installed in the device; a second data set indicating an amount of particulates in the air stream over a period of time since a particulate filter was installed in the device; generating labeled training data by labeling each pair of first and second data sets to relate those first and second data sets to an actual life of a particulate filter determined with respect to the particulate filter used to obtain the pair of first and second data sets; A machine learning model is trained using the labeled training data for calculating an estimated life of a particulate filter based on a pair of the first and second data sets.

Images