WO2021129919A1

WO2021129919A1 - Gaseous fluid monitor and method for monitoring properties of gaseous fluid

Info

Publication number: WO2021129919A1
Application number: PCT/EP2019/086898
Authority: WO
Inventors: Isabelle Raible; Tjalf Pirk; Bee Keen Gan; Chi Trung NGO
Original assignee: Robert Bosch GmbH
Current assignee: Robert Bosch GmbH
Priority date: 2019-12-23
Filing date: 2019-12-23
Publication date: 2021-07-01
Anticipated expiration: 2022-06-23
Also published as: DE112019007992T5

Abstract

An aspect concerns a gaseous fluid monitor for monitoring properties of a gaseous fluid, including particle size. The gaseous fluid monitor may include a detector configured to provide a detection of at least one of the properties of the gaseous fluid, which detector may include a particulate matter sensor configured to detect first particles and second particles suspended in the gaseous fluid and to provide, as part of the detection, a first concentration of the first particles, and a second concentration of the second particles. The gaseous fluid monitor may include a classifier configured to classify the detection into the pre-defined set of classes. Another aspect concerns a method for monitoring the properties of the gaseous fluid. The method may include detecting first particles and second particles, and may further include classifying the detection into a pre-defined set of classes.

Description

GASEOUS FLUID MONITOR AND METHOD FOR MONITORING PROPERTIES

OF GASEOUS FLUID

TECHNICAL FIELD

[0001] This disclosure relates to a gaseous fluid monitor and a method for monitoring properties of a gaseous fluid. This disclosure further relates to a computer program product.

BACKGROUND

[0002] Nowadays smoke detectors are utilized widely for residential safety. However, smoke detectors take rather long time to trigger an alarm when a fire starts for some fires like test fire (TF) smoldering wood and smoldering cotton. Recently, fire detector based on gas sensors have emerged. The principal hypothesis of gas-based fire detectors is that gases and volatile compounds will be released in many types of fire before smoke appears, thereby it is more likely that a fire can be detected earlier. However, existing systems still have a significant rate of false alarms of 35% or more.

[0003] Therefore, it is desirable to provide improved fire detectors with a reduced rate of false alarms.

SUMMARY

[0004] The present disclosure relates to a gaseous fluid monitor and a method for monitoring properties of a gaseous fluid. In some embodiments, the gaseous fluid monitor is a fire detector. [0005] Various embodiments concern a gaseous fluid monitor for monitoring properties of a gaseous fluid. The properties of the gaseous fluid may include at least one of: particle size, particle size concentration, particle velocity, type of gas molecules, concentration of gas molecules. The properties of the gaseous fluid may also include physical properties, for example, at least one of pressure, relative humidity, absolute humidity, and temperature. The gaseous fluid monitor may include a detector and a classifier. The detector may be configured to carry out a detection of at least one of the properties of the gaseous fluid and provide the corresponding detection data. The detector may include a particulate matter sensor configured to detect first particles and second particles suspended in the gaseous fluid. The detector may further be configured to provide, as part of the detection data, a first concentration of the first particles, and a second concentration of the second particles. The first particles may be particles of a first size and the second particles may be particles of a second size different from the first size. The classifier may be configured to classify the detection data into a pre-defmed set of classes. At least two classes of the pre-defmed set of classes may be indicators of respective sources of the gaseous fluid.

[0006] Various embodiments concern a method for monitoring the properties of a gaseous fluid. The method may include detecting first particles and second particles suspended in the gaseous fluid. The first particles may have a first size and the second particles may have a second size different from the first size. The method may further include classifying the detection data into a pre-defmed set of classes. At least two classes of the pre-defmed set of classes may be indicators of respective sources of the gaseous fluid.

[0007] Various embodiments concern a computer program product including instructions to cause a gaseous fluid monitor as described herein to carry out the instructions of the method as described herein.

BRIEF DESCRIPTION OF THE DRAWINGS [0008] Embodiments of the invention will now be described, by way of example, and with reference to the following drawings in which:

- FIG. 1 shows a schematic illustration of a gaseous fluid monitor 100, in accordance with various embodiments;

- FIG. 2 shows a schematic illustration of a gaseous fluid monitor 100 including a pre processor 130, in accordance with various embodiments;

- FIG. 3 shows a schematic illustration of a gaseous fluid monitor 100 including a dimensionality reducer 140, in accordance with various embodiments;

- FIG. 4 shows a schematic illustration of a detector 110 including a particulate matter sensor 112 and a gas sensor 114, in accordance with various embodiments;

- FIG. 5 shows a flow chart of a method 200, in accordance with various embodiments;

- FIGS. 6a-6f show a first set of tests, wherein classification is made for ambient data, fires, and nuisances using principal component analysis (PCA) and support vector machine (SVM).

- FIGS. 7a-7f show a second set of tests, wherein classification is made for ambient data, open wood (TF1), Smoldering fire (TF2 and TF3), Flaming fire (TF4, TF5 and TF8), and nuisances.

DETAILED DESCRIPTION [0009] In the description, which follows, the drawing figures are not necessarily to scale and certain features may be shown in generalized or schematic form in the interest of clarity and conciseness or for informational purposes. In addition, although making and using various embodiments are discussed in detail below, it should be appreciated that as described herein are provided many inventive concepts that may be embodied in a wide variety of contexts. Embodiments discussed herein are merely representative and not limiting.

[0010] Various embodiments disclosed herein relate to the various aspects of the disclosure such as a monitor, a method, and a computer program product. Embodiments and explanations thereof disclosed in connection with one embodiment may be applicable to other embodiments. For example, embodiments and explanations to the system may be applicable to the method. [0011] According to various embodiments, the properties of the gaseous fluid may include at least one of: particle size, particle size concentration, particle velocity, type of gas molecules, concentration of gas molecules. The properties of the gaseous fluid may also include physical properties, for example, at least one of: pressure, relative humidity, absolute humidity, and temperature. For example, the gaseous fluid may include particulate matter with a particle size distribution, and wherein the one or more particles sizes and/or one or more particle velocities may be chosen as properties of the gaseous fluid. The gaseous fluid may be air. Further, the properties of the gaseous fluid may include the concentration of a gas component, for example carbon monoxide.

[0012] According to various embodiments, the gaseous fluid monitor may include a detector and a classifier. According to various embodiments the detector may include a particulate matter sensor. The particulate matter sensor may be configured to detect first particles and second particles suspended in the gaseous fluid. The detector may further include a gas sensor, and the detection data may include gas detection data.

[0013] According to various embodiments, the detector may further be configured to provide, as part of the detection and as detection data, a first concentration being the concentration of the first particles in the gaseous fluid, and a second concentration being the concentration of the second particles in the gaseous fluid. The first particles may be particles of a first size and the second particles may be particles of a second size different from the first size. The detector may be configured to provide a concentration and/or particle velocity of further particles of different sizes included in the detection data. For example, the detector may be configured to detect particle of one or more sizes selected from: 0.5 micrometers, 1 micrometer, 2.5 micrometers, 5 micrometers, e.g., with a tolerance of +/- 10%. In another example, the detector may be configured to detect particles of one or more size ranges selected from: substantially equal or smaller than 0.5 micrometers (e.g. PM0.5), substantially equal or smaller than 1 micrometers (e.g. PM1.0), substantially equal or smaller than 2.5 micrometers (e.g. PM2.5), substantially equal or smaller than 5 micrometers (e.g. PM5.0). The particulate matter may be determined, e.g., in accordance to ISO 7708:1995.

[0014] According to various embodiments, the gaseous fluid monitor may be equipped with the detector including one or more particulate matter (PM) sensors. Each PM sensor may detect particles concentration at a pre-determined particle size and/or a particle concentration distribution. As used in the present disclosure and in accordance with various embodiments, the term “concentration” used in connection with particles, may mean particle counts per second for a pre-determined flow rate. Particle concentration may be absolute or relative, thus, as used herein, and in accordance with various embodiments, for a certain type of particles (e.g. particles of a same size) the expression “particle count” when measured over time (e.g. particle rate) may be used to refer to relative particle concentration, which is a value proportional to the absolute particle concentration. Each PM sensor may operate using various principles such as, e.g., optical or mass detection. The gaseous fluid monitor may be equipped with a gaseous fluid flow producer, for example a fan, for controlling gaseous fluid flow (e.g. airflow) inside the detector, and generating the necessary gaseous fluid exchange or flow for each of the PM sensors which is necessary for operation of the PM sensor(s). For example, environmental PM may travel inside the PM sensor from an inlet to an outlet, carried by the airflow.

[0015] According to various embodiments, the gaseous fluid monitor may be equipped with the detector including one or more gas sensors. Each of the gas sensors may be selective to a certain gas, e.g., an electrochemical cell for carbon monoxide monitoring, or may be configured to measure the sum of several gases, e.g. metal oxide sensors for volatile organic compounds (VOCs). The gas sensor may be configured to detect a gas component, e.g., carbon monoxide. The detection data may include gas detection data. The gaseous fluid monitor may be equipped with a gaseous fluid flow producer, for example a fan, for controlling gaseous fluid flow (e.g. airflow) inside the detector, and generating the necessary gaseous fluid exchange or flow for each of the one or more gas sensors which is necessary for operation of the gas sensor(s). For example, gaseous fluid may travel inside the sensor from an inlet to an outlet, carried by the airflow. The flow producer may be a same flow producer for the gas sensor(s) and the PM sensor(s).

[0016] According to various embodiments, the gaseous fluid monitor may further include other types of sensors, such as standard environmental sensors, e.g. temperature, pressure, humidity, sound. The detection data may further include one or more of: temperature, pressure, relative humidity, sound data.

[0017] According to various embodiments, the gaseous fluid monitor may be implemented as a fire detector.

[0018] According to various embodiments, the gaseous fluid monitor may further comprise: a computational unit configured to process the detection data and to carry out a classification with the classifier; and a communication circuit configured to transmit information indicating the source of the gaseous fluid based on the set of classes. For example, the communication circuit may be able to send a message for the fire guard, which may include the detection of a fire, and may further include further information about the source of the fire.

[0019] A classification model tries to draw some conclusion from an input, such as observed values (detection data) or training data. Classification may be provided, e.g., with a classification model including a neural network. There are two approaches to build a classification model including a neural network, namely supervised and unsupervised learning. Supervised learning uses a labelled data training set to build the classification model. On the other hand, unsupervised learning uses non-labeled data training set to build clusters to search for similarities, patterns or outliers within the dataset. Other exemplary approaches for classification may be based on statistical models and/or analytical models, which may also be further combined with a neural network.

[0020] According to various embodiments, the classifier may be configured to classify the detection data into a pre-defmed set of classes, for example, by using a pre-trained classifier. The classifier may include a suitable classification model, for example trained with classification data. At least two classes of the pre-defmed set of classes may be indicators of respective sources of the gaseous fluid. The pre-defmed set of classes may include at least two of: ambient data, open wood fire, smoldering fire, flaming fire, nuisance. Indicators of sources of the gaseous fluid may be for example: open wood fire class as indicator of open wood fire as a source; flaming fire class as indicator of flaming fire as a source; smoldering fire class as indicator of smoldering fire as a source. The classifier may further be configured to evaluate the detection data into ambient data. Ambient data refers to ambient environmental data when there are no events of fire or nuisance. The classifier may further be configured to determine if the gaseous fluid is a result of fire, e.g., classifying detection data into fire and non-fire. The classifier may be implemented with machine learning algorithms, for example as selected from: Support Vector Machine (SVM), K-Nearest Neighbour (KNN), Multi-Layer Perceptron (MLP), Long Short Term Memory (LSTM) and variations thereof. The machine classifier may perform two tasks or more, for example detect fire, detect no fire, and to classify the detection data into a class of fire. The algorithms may be trained with data sets obtained from controlled fires of different known types (classes) of fires, for example a dataset of data collected from EN54 standard fires.

[0021] As used herein and in accordance with various embodiments, the “detection data” may refer to the data provided by the detector. Further, the “detection data” may also refer to detection data provided by the detector which is further processed (processed may also be named as transformed), for example pre-processed by the pre-processor and/or processed by the dimensionality reducer. Thus, in some embodiments the classifier may directly receive detection data provided by the detector as input. In other embodiments, the classifier may receive the further processed detection data as input.

[0022] According to various embodiments, the gaseous fluid monitor may include one or more steps of data transformation to be applied on the detection data before the detection data (in its transformed state) is fed into the classifier. For example, the steps may include pre-processing the detection data, and reducing the pre-processed data, for example, into principal components. Accordingly, the gaseous fluid monitor may include a pre-processor and/or a dimensionality reducer.

According to various embodiments, the gaseous fluid monitor may further include a pre processor. The pre-processor may be configured to pre-process the detection data into pre- processed data as features. The classifier may be configured to receive the pre-processed data as input. The pre-processed data may include a time dependent particle concentration of at least one of the first particles and the second particles, and/or at least one of: an integral, a first derivative, a second derivative. For example, the pre-processed data may include the second derivative of the time dependent particle concentration of the first particles or the second derivative of the time dependent particle concentration of the second particles. Further, the pre- processed data may include a concentration difference between the first particles and the second particles. The detection data may be pre-processed by the pre-processor to remove noise and extract robust information, known as features. The features may include at least one of: integrals, first derivative, second derivatives, difference in particle concentration between different sizes of particles (e.g. first particles and second particles), standard deviation and variance of integral and first derivative of particle, e.g., of particles of size 0.5 mpi and 1.0 mpi. Alternatively or in addition, the features may include one or more of: a difference of particle PM0.5 readings and PM1.0 readings, a difference of PM0.5 readings and PM2.5 readings, a difference between PM0.5 readings and PM5.0 readings, wherein readings may refer to the particles’ concentration or a corresponding or proportional value (e.g. counts per time). The standard deviation of integral and first derivative of particles measures the extent to which integral and slope data of particle counts per second varies from the mean of integral and slope data. The variance is the sum of squares of differences between integral and slope data and their means.

[0023] According to various embodiments, the gaseous fluid monitor may further include a dimensionality reducer configured to transform the pre-processed data into features as principal components. The classifier may be configured to receive the features as the detection data. For example Principal Component Analysis (PCA) may be used for feature reduction and the obtained new principal components may then be fed into the classifier as detection data. Dimensionality reduction refers to the process of reducing the number of random variables in the pre-processed data to obtain a set of principal variables. One approach is feature selection trying to find out a subset of attributes using different strategies like filter, wrapper and embedded strategy. Another approach is feature projection (or feature extraction) which transforms the data from high-dimensional space to a fewer dimensions. The data transformation might be linear or non-linear. Thus, as used herein, the term “reducer” or “reduction” (or other variants thereof) may include feature selection and/or feature extraction. [0024] One described architecture for processing the detection data may include data collection of raw data, the preprocessing of the raw data into prepared data, the dimensionality reduction of the prepared data into features, and the classification of the features into the classes which may be further classified by a classification model. However, the disclosure is not limited thereto, and variations or other architectures may be used in the implementation of the gaseous fluid monitor and also in the method for monitoring the properties of a gaseous fluid.

[0025] According to various embodiments, the detector may further include a gas sensor. The detection may further include gas detection data. For example, the gas sensor may be a sensor of gas that is oxidized, for example carbon monoxide.

[0026] Various embodiments concern a method for monitoring the properties of a gaseous fluid. According to various embodiments, the method may include detecting first particles and second particles suspended in the gaseous fluid and provide detection data. The method may further include to provide, as part of the detection data, a first concentration of the first particles, and a second concentration of the second particles. The method may further include classifying, with the classifier as described in various embodiments, the detection data into the pre-defmed set of classes. At least two classes of the pre-defmed set of classes may be indicators of respective sources of the gaseous fluid.

[0027] According to various embodiments, the method may further include pre-processing the detection into pre-processed data, and wherein the classifier receives the pre-processed data as input, and wherein the classifying uses the pre-processed data as input.

[0028] Various embodiments concern a computer program product including instructions to cause a gaseous fluid monitor to carry out the instructions of the method according to various embodiments.

[0029] FIG. 1 shows a schematic illustration of a gaseous fluid monitor 100 for monitoring properties of a gaseous fluid 30, in accordance with various embodiments. The properties may include particle size. The gaseous fluid monitor 100 may include a detector 110 and a classifier 120. The detector 110 may be configured to carry out a detection of at least one of the properties of the gaseous fluid 30 and provide respective detection data 201. The detector may include a particulate matter sensor 112 configured to detect first particles 32 and second particles 34 suspended in the gaseous fluid 30, for example as an aerosol. The detector may be further configured to provide, as part of the detection data 201, a first concentration of the first particles 32, and a second concentration of the second particles 34. The first particles 32 have a first size and the second particles 34 have a second size different from the first size. For example, the first size may be smaller than the second size. Alternatively, the second size may be smaller than the first size.

[0030] The classifier 120 shown in FIG. 1 may be configured to classify the detection data 201 (which may be from the detector or further processed) into pre-defmed set of classes, shown for illustrative purposes as 41, 42, CL0-CL4. At least two classes of the pre-defmed set of classes may be indicators of respective sources of the gaseous fluid 30 for example. Two classes may be used for identification of fire and non-fire, for example, class 41 for indication of fire and class 42 for indication of non-fire.

[0031] FIG. 2 shows a schematic illustration of a gaseous fluid monitor 100 including a pre processor 130, in accordance with various embodiments. The pre-processor pre-processes the detection as pre-processed data PD1 which is fed as input into the classifier 120.

[0032] FIG. 3 shows a schematic illustration of a gaseous fluid monitor 100 including a pre processor 130, in accordance with various embodiments. The gaseous fluid monitor 100 further includes a dimensionality reducer 140. The pre-processor pre-processes the detection data as pre-processed data PD1 which is fed into the dimensionality reducer 140 which in turn is configured to transform the pre-processed data into features as principal components OF1 which are then fed as input into the classifier 120.

[0033] FIG. 4 shows a schematic illustration of a detector 110 including a particulate matter sensor 112 and a gas sensor 114, in accordance with various embodiments. The detection 202 includes a first concentration of the first particles, and a second concentration of the second particles and gas detection data.

[0034] FIG. 5 shows a flow chart of a method 200 for monitoring the properties of a gaseous fluid 30, wherein the properties include particle size, in accordance with various embodiments. The method 200 includes detecting 210 the first particles 32 and the second particles 34 suspended in the gaseous fluid 30 and providing corresponding detection data 201. As previously described, the first particles 32 have a first size and the second particles 34 have a second size different from the first size. The method 200 further includes classifying 220 the detection data 201 into a pre-defmed set of classes, wherein at least two classes of the pre defined set of classes are indicators of sources of the gaseous fluid 30. The method 200 may further include pre-processing 204 the detection data 201 into pre-processed data (PD1), and wherein the classifier 120 receives the pre-processed data (PD1) as input, and wherein the classifying 220 uses the pre-processed data (PD1) as input. The pre-processed data (PD1) includes a time dependent particle concentration of at least one of the first particles 32 and the second particles 34. The time dependent particle concentration includes at least one of: an integral, a first derivative, a second derivative. The time dependent particle concentration may further include a concentration difference between the first particles and the second particles. [0035] There are several main advantages in using the combination of gas and particulate matter (PM) sensors for fire detection. Firstly, it enables early fire detection, especially in case of smoldering fires where gas is released before smoke evolves. Secondly, it is believed to bring additional benefits for building occupants since based on different released chemical signatures of fires and particle with different particle sizes and pattern recognition algorithms it is possible to identify different types of fires and to offer suitable and timely warning to help building owners and firefighters determine the possible locations where the fire may happen and how to extinguish it and evacuate. Thirdly, the cost of fire detectors is low, considerably due to large availability of sensors for consumer electronics. Fourthly, the size of fire detectors will be smaller when ionization chambers are no longer used. Finally, it is also possible to use the gaseous fluid monitor equipped with gas and PM sensors, including when the monitor is a fire detector, for environmental purposes like measuring volatile organic compounds and particles for indoor air quality monitoring. Thereby, additional benefits may be provided to building occupants, for example by being able to issue warnings about further potential health hazards besides fire.

[0036] Comprehensive tests were performed to determine which feature dominates the results of the pattern recognition and to evaluate the performance of the gaseous fluid monitor and of the method in accordance with various embodiments. A first data set was provided as training data and a second set was provided for test data. For these tests, a supervised Machine Learning algorithm - Support Vector Machine (SVM) and a neural network - Long Short-Term Memory (LSTM) were used. The features considered for the following tests are integral, first derivative, second derivative, and concentration difference between different particle sizes, these features were extracted by preprocessing. Every feature and feature combination was varied to evaluate the performance of the classification model.

[0037] FIGS. 6a-6f show a first set of tests, wherein classification is made for ambient data (labelled as CL0), fires (labelled as CL1) and nuisances (labelled as CL2) using PCA and SVM. [0038] FIG. 6a shows the confusion matrix with integral using the (principal components (PC) = 3).

[0039] FIG. 6b shows the confusion matrix with first derivatives (PC=3).

[0040] FIG. 6c shows the confusion matrix with second derivatives (PC=3).

[0041] FIG. 6d shows the confusion matrix with integral, first derivatives, and second derivatives (PC=10).

[0042] FIG. 6e shows the confusion matrix with integral, first derivatives, second derivatives, and particle count difference between particles of different sizes (PC=12).

[0043] FIG. 6f shows the confusion matrix with integral, first derivatives, second derivatives, and particle count difference between particles of different sizes, standard deviation, and covariance of integral of PM (PC=13).

[0044] As shown in FIGS. 6a, 6b and 6c, the classification accuracy of the model is worse when using solely integral, first or second derivatives, around 38%. This could be because only one feature is not sufficient for ML algorithm to discriminate three different classes. However, the combination of those three features increases the accuracy up to 77% as displayed in FIG. 6d. Excellent results were achieved when the concentration difference between different sizes of particles was taken into account, as shown in FIG. 6e, where the accuracy is about 96%.

[0045] FIGS. 7a-7f show a second set of tests in which fires were grouped in following classes: open wood (TF1) CL1, smoldering fire (TF2 and TF3), flaming fire (TF4, TF5 and TF8). The fires were classified with ambient data CLO and nuisances CL3 using PCA and Long-Shost Term Memory.

[0046] FIG. 7a shows the confusion matrix with integral (PC=3).

[0047] FIG. 7b shows the confusion matrix with first derivative (PC=3).

[0048] FIG. 7c shows the confusion matrix with second derivative (PC=3).

[0049] FIG. 7d shows the confusion matrix with integral, first derivative, second derivative (PC=10).

[0050] FIG. 7e shows the confusion matrix with integral, first and second derivatives, and particle count difference between different particle sizes (PC=12).

[0051] FIG. 7f shows the confusion matrix with integral, first and second derivatives, particle count difference between different particle sizes, and standard deviation and covariance of integral of principal component (PCM 3).

[0052] As shown in FIGS. 7a, 7b and 7c, the classification accuracy of the model is improved when more parameters, than solely integral and first derivatives, are used. By combining three parameters (integral, first derivative, second derivative) with the concentration difference between different sizes of particle, the accuracy increases to 90% or above as displayed in FIGS. 7d and 7e. Excellent results are achieved when taking into account the standard deviation and covariance of integral of particle size 0.5 and 1.0 mpi as in FIG. 7f where the accuracy is about 98% for TF1 and 91% for group of TF2+TF3.

[0053] The gaseous fluid monitor allows in a first step detecting an event, e.g., large deviations from typical ambient concentration, with signal above a certain threshold. In a second step, it allows gaining information, e.g. fire against nuisances, or type of fire. As shown, various embodiments provide a low rate of false alarms and fire classification. In addition, fire classification may enable first responders to be better prepared to combat a fire, for example by preparing proper extinguishers to extinguish the fire.

Claims

1. A gaseous fluid monitor (100) for monitoring properties of a gaseous fluid (30), the gaseous fluid monitor (100) comprising: a detector (110) configured to provide detection data (201) of at least one of the properties of the gaseous fluid (30), wherein the detector comprises a particulate matter sensor (112) configured to detect first particles (32) and second particles (34) suspended in the gaseous fluid (30), and further configured to provide, as part of the detection data (201), a first concentration of the first particles (32), and a second concentration of the second particles (34), wherein the first particles (32) have a first size and the second particles (34) have a second size different from the first size; and a classifier (120) configured to classify the detection data (201) into pre-defmed set of classes, wherein at least two classes of the pre-defmed set of classes are indicators of respective sources of the gaseous fluid (30).

2. The gaseous fluid monitor (100) of claim 1, wherein the classifier (120) is further configured to evaluate the detection data (201) into ambient data.

3. The gaseous fluid monitor (100) of claim 1 or claim 2, wherein the classifier (120) is further configured to determine if the gaseous fluid (30) is a result of fire.

4. The gaseous fluid monitor (100) of any one of the previous claims, wherein the pre- defmed set of classes comprises at least two of: ambient data, open wood fire, smoldering fire, flaming fire, nuisances.

5. The gaseous fluid monitor (100) of any one of the previous claims, further comprising a pre-processor (130) to pre-process the detection data (201) into pre-processed data (PD1) and wherein the classifier (120) receives the pre-processed data (PD1) as detection data (201).

6 The gaseous fluid monitor (100) of claim 5, wherein the pre-processed data (PD1) comprises a time dependent particle concentration of at least one of the first particles (32) and the second particles (34), and wherein the time dependent particle concentration comprises at least one ofan integral, a first derivative, a second derivative.

7. The gaseous fluid monitor (100) of claim 6, wherein the pre-processed data (PD1) comprises at least one of the second derivative of the time dependent particle concentration of the first particles (32), the second derivative of the time dependent particle concentration of the second particles (34).

8. The gaseous fluid monitor (100) of claim 6 or claim 7, wherein the pre-processed data (PD1) comprises a concentration difference between the first particles (32) and the second particles (34).

9. The gaseous fluid monitor (100) of any one of the claims 5 to 8, further comprising a dimensionality reducer (140) configured to transform the pre-processed data (PD1) into features (OF1) as principal components, and wherein the classifier (120) receives the features (OF1) as the detection data (201).

10. The gaseous fluid monitor (100) of any one of the previous claims, wherein the detector (110) further includes a gas sensor (114) and wherein the detection data (201) further includes gas detection data.

11. The gaseous fluid monitor (100) of any one of the previous claims, further comprising: a computational unit configured to process the detection data and to carry out a classification with the classifier; and a communication circuit configured to transmit information indicating the source of the gaseous fluid (30) based on the set of classes.

12. A method (200) for monitoring the properties of a gaseous fluid (30), the properties comprising particle size, the method (200) comprising: detecting (202) first particles (32) and second particles (34) suspended in the gaseous fluid (30) and providing detection data (201), wherein the first particles (32) have a first size and the second particles (34) have a second size different from the first size; and classifying (208) the detection data (201) into a pre-defmed set of classes, wherein at least two classes of the pre-defmed set of classes are indicators of sources of the gaseous fluid (30).

13. The method (200) of claim 12, further comprising: pre-processing (204) the detection data (201) into pre-processed data (PD1), and wherein the classifier (120) receives the pre-processed data (PD1) as detection data (201), and wherein the classifying (208) uses the pre-processed data (PD1) as the detection data (201).

14. The method (200) of claim 13, wherein the pre-processed data (PD1) comprises a time dependent particle concentration of at least one of the first particles (32) and the second particles (34), and wherein the time dependent particle concentration comprises at least one of:, an integral, a first derivative, a second derivative.

15. The method (200) of claim 13 or claim 14, wherein the pre-processed data (PD1) comprises a concentration difference between the first particles (32) and the second particles (34).

16. A computer program product comprising instructions to cause a gaseous fluid monitor to carry out the instructions of the method according to one of the claims 12 to 15.