A robust artificial olfactory system

Abstract

The present invention discloses a system and a method for detecting the presence and changes in the concentration of volatile organic compounds (VOCs) within an aroma stream. The system utilizes a sensor array composed of polymer polymer composite sensing elements that generate a time-series of electrical signals in response to VOC interactions with the polymer polymer composite sensing elements. A preprocessing unit filters noise from these signals and incorporates aroma state features into the time series of electrical signals, to generate a preprocessed time series. The system then processes the data using a machine learning model, to detect changes in the preprocessed time series and correlate the changes with the presence or concentration variations of VOCs in the aroma stream.

Description

A ROBUST ARTIFICIAL OLFACTORY SYSTEMFIELD OF THE INVENTION

[0001] This disclosure relates to a system and method that is used for detecting volatile organic compounds present in an aroma irrespective of the background confounding aroma or VOC present in an environment of detection.BACKGROUND

[0002] Various aroma sources such as exhaled breath, skin, food, etc. emanate aroma into the environment. These aromas consist of a plurality of analytes or volatile organic compounds(VOCs). Current aroma or VOC analysis systems are unable to accurately and / or repeatedly identify and / or quantify chemicals such as volatile organic compounds when a confound background is present in the environment of detection.SUMMARY

[0003] Disclosed herewith is that detects the presence and change in concentration of volatile organic compounds present in an aroma while ignoring the background confounding aroma or VOC that may be present in the environment.BRIEF DESCRIPTION OF DRAWINGS

[0004] FIG. 1 illustrates a filtered and normalized electrical signal generated by the preprocessing unit.

[0005] FIG. 2 illustrates a process flow chart to determine the presence and concentration of volatile organic compounds in an aroma using the embodiments of the invention.DETAILED DESCRIPTION

[0006] The system comprises a sensor array for analyzing an analyte or component of an analyte wherein the analyte may be present in one of the liquid phase, gaseous phase, or solid phase. A sensor array is a group of sensor elements wherein each sensor element is a polymer composite thin film that may conduct electric current. The sensor array further comprises a plurality of electrodes that are connected to the sensor elements. The polymercomposite thin film reacts with the compositions of the aroma source and undergoes a property change. The system also comprises a processing unit that may use an artificial intelligence-based system to draw inference from the sensor value, regarding the compositions of the aroma source.

[0007] In some embodiments of the invention, the robust artificial olfaction system may detect an analyte or component of an analyte irrespective of the environment of operation, by using at least some of the components of the system. In some embodiments, the robust artificial olfaction system may detect an analyte or component of an analyte irrespective of the system components functioning degradation over a long period and in the environment of operation, wherein system component functioning degradation may refer to sensor aging, sensor drift, an onboard timing system drift and degradation, etc.

[0008] According to an embodiment of the invention, the polymer composite thin film of the sensor element comprises nanoparticles embedded in a chemical or matrix, wherein the chemical or matrix may belong to a class of compounds that are required for sensing a specific type of analyte or component of the analyte. Upon exposure to the analyte or the component of the analyte, the thin film may undergo one of a physical and chemical change. In some embodiments of the invention, the physical change, associated with exposure to a specific type of component of the analyte, is an increase in distance between the nanoparticles. In yet another embodiment of the invention, the conductivity of the polymer composite thin film is different for each of the plurality of sensing elements present in the system. Further, the conductivity of the plurality of thin films may be dependent on the concentration of nanoparticles per unit volume, present in the polymer composite thin film.

[0009] For the purpose of this invention, an aroma is defined as a mixture of a plurality of volatile organic compounds or analytes that can volatilize rapidly. The term volatile organic compounds, VOCs and analyte may be used interchangeably without deviating from this definition.

[0010] In the event of exposure to a composition of an aroma source or the analyte, the polymer composite thin film of the sensor element undergoes an expansion phase, wherein the volume of the thin film increases at the time of exposure. The increase in the volume of the thin film increases the distance between the nanoparticles embedded in the thin film.The increase in distance between the nanoparticles changes the conduct! vity / resistivity of the thin film. The change in conductivity / resistivity / chemical resistance / impedance (electrical signal) of the thin film is identified and inferred with a processing system to determine the composition of the aroma source, analyte, or component of the analyte. As the interaction is continuous over a period of time, being directly in correlation to aroma introduction to the sensor, a time series of electrical signals is generated by the sensor.

[0011] In an embodiment of the invention, the capacitance value of the polymer composite thin film is analyzed and the change in capacitance of the thin film due to exposure to the composition of the aroma source is identified and inferred to determine the composition of the aroma source. In another embodiment of the invention, the voltage fluctuation across the thin film is analyzed at the time of exposure of the thin film to the analyte or aroma source, to determine the composition of the aroma source or component of the analyte. In another embodiment of the invention, the temperature change in the thin film, during the exposure period, is analyzed to determine the composition of the aroma source, the analyte, or component of the analyte. In yet another embodiment of the example, the luminescence of the thin film is analyzed at the time of exposure to the composition, to determine the composition of the aroma source, the analyte, or component of the analyte, at the time of exposure.

[0012] During the event of exposure of the plurality of sensing elements to the analyte or the compositions of the aroma source, the change in physical and chemical parameters of the sensing elements is communicated to the processing unit using electrodes. The electrodes communicate the change in the resistance value of the sensing element, at the time of exposure to the analyte or the aroma source, to the processing unit which in turn determines the composition of the aroma source, analyte, or component of the analyte.

[0013] In an embodiment of the invention, the electrodes communicate the change in capacitance value of the sensing element at the time of exposure to the compositions of the aroma source or the analyte. In yet another embodiment of the invention, at least one of change in voltage, temperature, and luminescence of the sensing element is communicated via the electrodes to determine the composition of the aroma source, the analyte, or the component of the analyte.

[0014] In an embodiment of the invention, a preprocessing unit is provided between the electrodes of the sensing element and the processing system. The electrodes transmit the change in physical and chemical parameters to the preprocessing unit to filter and normalize the change in parameters and scale the change in parameters, according to the limits of the processing system. The preprocessing unit thus generates a preprocessed time series, as depicted in figure 1, that may then be processed by the processing system downstream, as depicted in figure 2. The preprocessing unit comprises a filtering unit and a normalizing unit, the filtering unit being configured to remove noise from the generated electrical signal, the noise being characterized as the signal that does not directly correspond to interaction of sensing elements with the volatile organic compounds preset in the aroma, such as thermal noise, shot noise, flicker noise, environmental noise, quantization noise, burst noise and avalanche noise; and the normalizing unit being configured to add aroma state features to the time series corresponding to a stream of aroma that was subjected to the sensor at a time instance that immediately precedes the instance of measurement. The features may be baseline, exposure, rate of exposure, recovery, rate of recovery etc. A communication path is provided between the preprocessing unit and the processing system, through which the pre-processing system transmits the pre-processed data to the processing system to identify the composition of the aroma source, the analyte, or component of the analyte.

[0015] The processing system uses the filtered and normalized data provided by the preprocessing unit of the olfaction system to draw inferences about the composition of the aroma source or components of the analyte, which in turn leads to the identification of the aroma, the analyte, or component of the analyte. The processing system uses an artificial intelligence unit to draw inferences from the filtered and normalized data to identify the composition of the aroma source, the analyte, or the component of the analyte.

[0016] According to an embodiment of the invention, at least one of deep learning method, such as artificial neural networks, deep neural networks, convolutional neural networks, and recurrent neural networks is being used by the processing system to draw inference from the filtered and normalized data and identify the composition of the aroma source and hence the aroma.

[0017] In yet another embodiment of the invention, at least one of long short-term memory, gated recurrent unit, time series forecasting, transformers, autoregressive integrated moving average, and seasonal autoregressive integrated moving average is used by the processing system to draw inference from the filtered and normalized data and identify the composition of the aroma source and hence the aroma.

[0018] The pre-processing unit of the olfaction system is designed to prepare the output of the sensor elements for the processing unit. To prepare the out of the sensor element, the pre-processing unit filters the output of the sensor elements using filters that are present in the pre-processing unit. Post the filtering process, a normalization process is carried out wherein the filtered data is morphed and scaled according to the input specifications of the processing system.

[0019] According to an embodiment of the invention, the normalization process is performed before the filtering process. In yet another embodiment of the invention, a plurality of filtering and normalizing processes are performed and the order of each filtering and normalizing process is determined based on the output of each process. For each process, the output obtained is compared with the response of the sensor, and the difference between the output of each process and the sensor response is then used to determine the number of filtering and normalizing processes required and the order in which the filtering and normalizing processes are to be performed.

[0020] The filtering process of the pre-processing unit involves removing the noise / unwanted data from the sensor data by partially or fully suppressing the unwanted data obtained from the sensing unit. Families of linear continuous-time filters may be used for filtering the data obtained from each sensor of the sensor array. In some embodiments, to get a maximally flat response, filters of the Butterworth filter family may be used for the data obtained from each of the sensors of the sensory array.

[0021] In an embodiment of the invention, at least one of the filters from the family of Chebyshev filer, Bessel filter, Elliptic filter Constant K filter, or m-derived filter may be used for removing the noise from the data obtained from each of the sensors of the sensor array. In yet another embodiment of the invention, a combination of a plurality of filters in the mentioned families may be used serially or in parallel to remove the noise from thedata. The combination is chosen based on the difference between the output of each of the combinations of filter processes and the sensor response.

[0022] For the normalization process, the gradient of the data obtained from the filtering process is determined. The gradient of the filtered data is determined for the data point. In an embodiment of the invention, the gradient of the filtered data is determined for time.

[0023] In yet another embodiment of the invention, the normalization process is performed before the filtering process and the gradient of the data for one of the data points or time, obtained from the plurality of sensors, is fed to the filtering process for removing the noise. In yet another embodiment of the invention, a plurality of normalization and filtering processes are performed in series or parallel, based on the output difference, and for each normalization process, the data from the preceding process is used to obtain the gradient.

[0024] After obtaining the gradient of the filtered response, the ratio of the gradient and the filtered response is obtained during the normalization process. The gradient calculated for one of the data points and with respect to time is divided with the filtered response at the same data point or the same time, to get the ratio of the gradient of the filtered response and the filtered response.

[0025] In an embodiment of the invention, the gradient of the data obtained from the sensor is divided by the date and time and the ratio thus obtained is then fed to the filter to obtain the filtered data. In yet another embodiment of the invention, for the process with multiple normalization and filtering steps, for each of the normalization steps, the ratio of gradient and data points is obtained and fed to the subsequent systems for processing of the signal.

[0026] The process of normalization further comprises a step of scaling the normalized data into a range that is compatible with the processing unit of the olfaction system. The scaling process involves the step of multiplying the filtered and normalized data with a scale factor, that is compatible with the processing unit. A scale factor of 1000 or 10000 is used for responses of each sensor of the plurality of sensor units to make it compatible with the Al-based systems present in the olfaction system.

[0027] In some embodiments of the invention, the olfaction system may be exposed to the analyte or the aroma source for a long period. One of the long periods of exposure and environmental parameters may create shifts in sensed data / response, hereinafter referred to as sensor drift. In the event of sensor drift or long exposure periods, the sensor datarecording period may be divided into smaller instances, such as instances of the length of 1 second, 0.1 seconds, 0.01 second, 0.001 seconds, and the normalization step is carried out based on the previous instance. In some embodiments of the invention, the ratio of the gradient of one of sensed and filtered data, obtained with respect to the data for the previous instance, and one of gradient and the absolute value of sensed and filtered data at the previous instance may be calculated during the normalization process.

[0028] In some embodiments of the invention, the normalization process may use the ratio of two gradients of data at two consecutive instances. In yet some other embodiments, the normalization process may use the ratio of the gradient of data in one instance and absolute or filtered data preceding or succeeding instances.

[0029] The olfaction system is provided with a plurality of environmental parameters sensing units that are used to detect the environmental parameters around an operating olfaction system. Further, in an embodiment of the invention, the olfaction system obtains the environmental parameters data from sensors present outside the olfaction system, in the near vicinity of the olfaction system. Further, the system may include a wireless or wired communication unit to take environmental parameters data from sensors present outside the olfaction system.

[0030] The scaled data is combined with the environmental parameters before providing the data to the processing system. The environmental parameters control the operation range of the sensing element along with the sensitization, functionalization, and rate of reaction of the compositions of the aroma source and the sensor element.

[0031] In an embodiment of the invention, the environmental parameters are combined with the sensed data before the normalization process and the processes of filtering (preprocessing) and normalization are carried out on the combined environmental parameters data and sensed data.

[0032] The environmental parameters that are combined with one of the scaled data and preprocessed data are at least one of humidity in the environment of operation of the olfaction system, the temperature of the environment of operation of the olfaction system, the atmospheric pressure of the environment of operation and the velocity of air around the olfaction system. Further, the combination of the weighted environmental parameters iscombined with one of the scaled and pre-processed data to obtain data for one of preprocessing and processing systems.

[0033] Further, the olfaction system or nearby sensors are used to detect the environmental parameters and the change in environmental parameters due to the introduction of a aroma source or analyte in an environment. Further, the system may detect the change in environmental parameters for the time for which the aroma source’s composition or analyte is sensed by the olfaction system.

[0034] Based on the environmental parameters data obtained by one of the onboard or off- board environmental parameters monitoring sensors, the plurality of preprocessed data is grouped in a plurality of subgroups based on the parameters in which the pre-processed data was obtained by the olfaction system.

[0035] The environmental parameters are also subdivided into a plurality of groups based on their ability to alter the response from the plurality of sensor elements of the olfaction system. The subgroup is divided based on at least one humidity range, temperature, velocity, and pressure around the sensing element of the olfaction system.

[0036] The preprocessed data sub-groups and the environmental parameters groups are combined according to the operation condition of the olfaction system.

[0037] For each of the groups and subgroups, hyperparameters are identified for training the Al engine. The Al engine-based processing system is trained for each of the hyperparameters for each of the sub-groups and groups and confidence level in the learning of the Al engine is provided, for improving the capability of the processing of the olfaction system in determining the composition of the aroma source or the analyte

[0038] In an embodiment of the invention, a penalty is assigned to the learning process based on the comparison of the determined composition of the aroma source by the olfaction system and the real composition of the aroma source or analyte, wherein the real composition of the aroma source or the analyte is determined by pre-fed industrial standard data for a aroma source the analyte or component of the analyte. The pre-fed industrial standard data is assigned a label, that is used for categorizing a response from the olfaction system.

[0039] After the training of the Al engine, a probability distribution may be created for the response of each of the sensors from the group of sensors present in the olfaction system.In an embodiment of the invention, a clustering of distribution may be created by t- distributed stochastic neighbor embedding, and the divergence of the olfaction system sensor data distribution may be analyzed, distinguished, and improved for analyzing the aroma source or the analyte.

[0040] The system further comprises the step of determining the relation between the probability distribution and the labels. The processing unit may assign the labels to the detection composition based on the cluster of the combined probability distribution data of each of the plurality of sensors and environment parameters around the olfaction system. The learning system that provides a confidence level to the prediction done by the processing system relates the confidence level with the probability distribution and assigns the label to one of the compositions of the aroma source, analyte, and component of the analyte.

[0041] The olfaction system allows back-to-back aroma source or analyte detection using the same plurality of sensing elements by using the modified preprocessing stage in which the normalization processes and filtering processes are carried out without requiring any information about a baseline for a sensed response from the plurality of sensing elements.

[0042] Further, the olfaction system may be provided with an embedded heater unit along with the sensor unit that is used for clearing the surfacing of the sensing element post a prediction, to ensure repeated use of the sensing unit. Further, the heater unit may also be used to alter the reaction rate, between the sensing element and the composition of the aroma source or the analyte, to control the rate of aroma source or analyte determination, and compensate for the effects of one of the environmental parameters, sensor degradation and sensor drifts on the sensing capabilities of the olfaction system.

[0043] The use of the modified pre-processing stages, which comprise the normalization process and the filtering process allows accurate sensing of the source of aroma or the analyte, without complete cleaning of the plurality of sensor elements, as the processing stage may be dependent on the ratio of the gradient of the filtered data point to the filtered data point, that in turns removes the need of knowing the baseline for sensing and comparison and hence the olfaction system becomes robust. Further, in some embodiments, the processing, specifically normalization, may be dependent on the filtereddata at two consecutive instances and the gradient of filtered data at two consecutive instances, which eliminates the dependence of the sensing system on the baseline.

Claims

CLAIMS1. A system for detecting the presence or change concentration of a plurality of volatile organic compounds present in a stream of aroma, the system comprises; a sensor including a plurality of polymeric polymer composite based sensing elements, each of the sensing elements being configured to generate a time series of electrical signal in response to an interaction with the plurality of volatile organic compounds present in the stream of aroma; a preprocessing unit comprising a filtering unit and a normalizing unit, the filtering unit being configured to remove noise from the generated electrical signal, the noise being characterized as the signal that does not directly correspond to interaction of sensing elements with the volatile organic compounds preset in the aroma; and the normalizing unit being configured to add aroma state features to the time series corresponding to a stream of aroma that was subjected to the sensor at a time instance that immediately precedes the instance of measurement; a processing system, configured to execute a machine learning model, that is trained to detect a change in a time series of electrical signal corresponding to the interaction between the plurality of volatile organic compounds and the plurality of sensing elements over a period of time and correlate it with the presence or change in concentration of a plurality of volatile organic compound present in the stream of aroma.

2. The system of claim 1, wherein the filtering unit includes at least one type of noise filter, to filter out at least one of thermal noise, shot noise, flicker noise, environmental noise, quantization noise, burst noise and avalanche noise.

3. The system of claim 1 , wherein the normalizing unit, configured to add aroma state features corresponding to the stream of aroma, replaces the time series of electrical signals with the gradient of the time series at each instance of the generated electrical signal.

4. The system of claim 1, wherein the normalizing unit replaces the times series of the electrical signal with the ratio of gradient of time series at each instance of the generated electrical signal and the electrical signal at the corresponding instance.

5. The system of claim 1, wherein the filtering unit removes the noise from the generated electrical signal and the normalizing unit uses the filtered electrical signal to add aroma state features to the time series of electrical signals.

6. The system of claim 1, wherein the normalizing unit adds aroma state features to the time series of electrical signals and the filtering unit filters the noise from the normalized electrical signal generated by the normalizing unit.

7. The system of claim 1 wherein the preprocessing unit generates a preprocessed time series of electrical signals after filtering the noise and normalizing the electrical signal using the filtering unit and the normalizing unit.

8. The system of claim 1 and 7, wherein the machine learning model of the processing system is trained on and is applied to the preprocessed time series, for determining the presence or change in concentration of a plurality of volatile organic compounds present in the stream of aroma.

9. A method for detecting the presence or change concentration of a plurality of volatile organic compounds present in a stream of aroma, the method includes; generating a time series of electrical signal corresponding to the interacting of plurality of plurality of sensing elements of a sensor with a plurality of volatile organic compounds present in the stream of aroma; preprocessing the electrical signal by filtering the electrical signal to remove noise generated during the process of electrical signal generation and normalizing the signal by adding aroma state features to the time series corresponding to a stream of aroma that was subjected to the sensor at a time instance that immediately precedes the instance of measurement; executing a machine learning model, that is trained to detect a change in a time series of electrical signal corresponding to the interaction between the plurality of volatile organic compounds and the plurality of sensing elements over a period of time; wherein the machine learning model is applied on the time series generated after preprocessing the electrical signal.

Images