Enose auto-calibration and auto-learning system and apparatus

Abstract

Disclosed herewith is a system and method for calibrating enose using a calibration apparatus. The calibration apparatus comprises a single analyte sensor, a processing unit and a plurality of enose. The data generated by the single analyte sensor is used as a label for the data generated by the enose, to calibrate the enose. Further, the system generates a training dataset for the AI based enose processor to train on, the training dataset comprise the data generated by the enose and the data generated by the single analyte sensor at the same time, wherein the data generated by the single analyte sensor works as a label for the data generated by the enose.

Description

ENOSE AUTO-CALIBRATION AND AUTO-LEARNING SYSTEM AND APPARATUSFIELD OF THE INVENTION

[0001] This disclosure relates to a system and an apparatus that is used for calibrating and training the enoses in a diverse background environment.BACKGROUND

[0002] An enose comprises an array of sensors / sensing elements and a plurality of Al models that run on the data generated by the enose. To train the Al model on the enose, a dataset generation system is required. As the data comes under a heavy influence of the background environment, therefore, it's imperative that the data is collected in the environment and the Al model is trained to work in those environments. Further, like every other analyte sensor, an enose also undergoes a drift over a period of time and requires a calibration cycle at regular intervals, to keep the measurements correlated to the actual concentration of analyte present in the environment.SUMMARY

[0003] Disclosed herewith is a system that is used for training and calibrating an enose . The system uses a calibrating apparatus that comprises a single analyte sensor, an enose, a clock and a plurality of environment sensors. The system further comprises processors and processing systems. The system generates labeled dataset from the apparatus by combining the data generated by the enose and adding a label to the data using the data provided by the precalibrated single analyte sensor. The dataset thus generated is used for training the Al model to detect the analyte that the single analyte sensor was detecting. Further, the dataset is also used for periodically calibrating the enose, thereby compensating for a drift that it may encounter over a period of time.BRIEF DESCRIPTION OF DRAWINGS

[0004] FIG. 1 illustrates the calibration apparatus that is used by the auto learning and auto calibration system.

[0005] FIG. 2 illustrates the process workflow of the auto learning and auto calibration system as per the embodiments of the invention.DETAILED DESCRIPTION

[0006] The invention discloses a learning and calibrating system for an Enose system wherein the enose is essentially an analyte sensing system, where the analytes are in the form of vapor or gas. The learning and calibrating system comprises a calibration apparatus that includes a single dedicated analyte sensor that detects a single analyte in the gaseous phase. The system also includes an array of sensors that work as an enose and detect multiple analytes from a given exposure or sample. The calibration apparatus includes a cloud communication module that sends time-stamped sensor signals from a single dedicated analyte sensor and array of sensors of the enose to a cloud server. The cloud server generates a label for the sensed value of the single analyte sensor and assigns the label to a fingerprint of sensed data generated from the array of sensors. A plurality of labels may be generated at a plurality of time instances, thereby providing labels that are indicative of the dynamics of analyte / VOC change in an environment.

[0007] The application discloses an auto / shadow-leaming and calibration system and calibration apparatus for an enose wherein the enose is an analyte sensing system that is used to detect a plurality of components present in an analyte and wherein the analytes that are sensed by the enose are in one of the vapour form and gaseous form.

[0008] The enose uses an apparatus that comprises a plurality of analyte sensing systems, processing units, communication modules, and timing units. At least one of the analyte sensing systems of the plurality of sensing systems is a pre-trained sensing system that can detect a component from the analyte when the sensing system is exposed to the analyte. At least one of the sensing systems of the plurality of sensing systems is an environment monitoring sensing system that measures the operating environment of the enose. At least one of the analyte sensing systems of the plurality of sensing systems is an untrained sensing system. The processing units comprise a plurality of processors wherein at least one of the processors of the plurality of processing unit processes the sensed data from the pre-trained sensing systems and at least one of the processors of the plurality of processing unit processes the sensed data from the untrained sensing systems. The timing unit of the apparatus may be a clock that is used to detect the time. The processing unit combines the data from the plurality of sensing systems with the data from the timing unit to make temporal sensed data from the sensing systems data. The communication module of the apparatus is used to communicate the temporal sensed data from at least one of the pluralities of sensing systems to a remote / cloud server.

[0009] The plurality of pre-trained sensing systems of the apparatus comprises sensing elements that generate a response in the form of a signal or raw data when the pre-trained sensing systems are exposed to a certain type of analyte. The certain type of analyte may be a part / component of a plurality of analytes that are present in the form of a mixture of vapors or gases. The pre-trained sensing systems may generate a response against one type of one of analyte or component of the analyte when exposed to the one analyte or mixture of vapour or gas. A processing unit is provided with the pre-trained sensing system that uses the response generated from the sensing element of the sensing system and determines at least one of presence and concentration of a certain type of analyte or component of a plurality of analytes present in a mixture of vapours or gases.

[0010] The plurality of untrained sensing systems of the apparatus comprises sensing elements that generate a response in the form of signal or raw data when the untrained sensor is exposed to at least one of an analyte and component of a plurality of analytes that are present in the form of a mixture of vapour or gases.

[0011] The remote server of the auto-learning and calibration system is a combination of at least one of a processing unit and a memory unit. The memory unit of the remote server stores the temporal sensed data from at least one of the pre-trained sensing systems and untrained sensing systems that are received from the apparatus. The processing unit of the remote server generates a label from the processed data from the pre-trained sensing system and combines the label with the sensed data of the untrained sensing system, thereby generating a training dataset and training the untrained system to detect at least one of an analyte and a part / component of a plurality of analytes that are present in the form of a mixture of vapours or gases.

[0012] At least one of the analyte sensing systems of the enose is a trained sensing system. At least one of the trained sensing systems comprises sensing elements that generate a response in the form of a signal or raw data when the trained sensing system is exposed to a certain type of analyte. The certain type of analyte may be a part / component of a plurality of analytes that are present in the form of a mixture of vapours or gases. The pre-trained sensing systems may generate a response against one type of one of analyte or component of the analyte when exposed to the one of analyte or mixture of vapours or gases. A processing unit is provided with the trained sensing system that uses the response generated from the sensing element of the sensing system and determines at least one of presence and concentration of a certain type of analyte or component of a plurality of analytes present in a mixture of vapours or gases

[0013] The auto-learning and calibrating system of the invention discloses calibrating at least one of the plurality of trained sensing systems of the enose. The process of calibrating the sensing system involves using at least one of the processing units and memory units of the remote server. The memory unit of the remote server stores the temporal sensed data from the trained sensing system. The processing unit of the remote server compares the label from the pre-trained sensing system and the label of the trained sensing system. In the event of observing a difference between the label of the trained sensing system and the pre-trained sensing system by the processing unit, the processing unit changes the label of the trained sensing system by assigning the label of the pretrained sensing system to the trained sensing system.

[0014] The plurality of the trained sensing system and untrained sensing system of the enose may comprise a sensor array wherein the array has a plurality of sensing elements, arranged in a set of configurations. The configuration of the sensing element may be rectangular, row and columnwise, circular, and stepped. The auto-learning and calibrating system and apparatus for enose trains and calibrates at least one of the untrained sensing systems and the trained sensing system of the enose wherein the response from the sensor array of at least one of the trained sensing systems and the untrained sensing system may be generated in the form of multi-dimensional response hereinafter called the fingerprint. The response from the pre-trained sensing system is one of a onedimensional or a two-dimensional response or a multi-dimensional response wherein the number of dimensions of the pre-trained sensing system may be less than the dimension of at least one of trained sensing system and untrained sensing system.

[0015] The processing unit of the remote server assigns the label generated for the low dimensional response of the pre-trained sensing system to the high dimensional response of at least one of the trained sensing systems and untrained sensing systems.

[0016] The auto-learning and auto-calibration system may generate labels for one type of sensing system, such as the pre-trained sensing system, and assign the generated labels to another type of sensing system, such as the trained or untrained sensing systems, and enable cross sensor mapping. The system may be used for loT enabling subsystems using the loT platform. The plurality of sensing systems of the loT platform are trained, retrained, or recalibrated for one of the new types of analytes and a new type of environment using the auto-learning and auto-calibration system. The subsystems of the loT platform may be pollution monitoring systems, industrial emission monitoring systems, natural gas monitoring systems, sewage monitoring systems, etc.

[0017] The pretrained sensing system of the auto training and auto-calibration system comprises a sensing element and a processing unit. The sensing element is one of the single analyte sensing elements wherein the element may be one of the chemiresistive sensing element, chemi-capacitivesensing element, and optical sensing element, wherein the sensing system detects the change in at least one of the physical parameters and chemical parameters of the sensing element when the sensing element is exposed to an analyte. The pre-trained sensing system may be provided with a degradation monitoring system wherein the degradation monitoring system detects the state of the sensing element of the pre-trained sensing system, to determine the accuracy and robustness of the pre-trained system to accurately detect the analyte.

[0018] The untrained and trained sensing system that uses the auto training and auto-calibration system comprises a sensing element and a processing unit. The sensing element may be an array of sensing elements wherein each sensing element may be at least one chemi resistive sensing element, chemi-capacitive sensing element, chemoelectric sensing element, thermochemical sensing element, wherein the sensing system detects the change in at least one of the physical parameters and chemical parameters of the sensing element when the sensing element is exposed to an analyte. Herein, chemi resistive, chemi capacitive and chemoelectric sensing of the sensing systems refer to change in resistance and capacitance across the sensing element, due to the occurrence of a chemical reaction or absorption or adsorption process.

[0019] In an embodiment of the invention, an analyte or a group of analytes, herein after also referred to as aroma, may be subjected to the calibration apparatus to generate a calibration curve for the analyte.

[0020] In an embodiment of the invention, the sensing element of the trained and untrained sensing system is a chemiresistive type sensing element wherein the sensing element may be made from a polymeric thin film. The polymeric thin film may be provided with fillers, to increase the conductivity across the polymeric thin film of the sensing element.

[0021] The auto training and auto-calibration system of the enose comprise a plurality of pretrained sensing systems, a plurality of untrained sensing systems, a plurality of trained sensing systems, timing units, and processing units. The relative location of the pre-trained sensing system and un-trained sensing system is selected, such that, the analyte exposure to the pre-trained sensing system is the same as the analyte exposure to the un-trained sensing system. Further, the relative location of the pre-trained and the trained sensing system is selected, such that, the analyte exposure to the pre-trained sensing system is the same as the analyte exposure to the trained sensing system. Further, a timer may be provided for each of the pre-trained sensing systems, untrained sensing systems, and trained sensing systems wherein the timer is a clock that records the time at which the sensing element interacts with the analytes that it is being exposed to. The pretrained and untrained sensing system or enose comprise a processor that hosts a plurality of Al models or expert systems. The pretraining or training referees to the process of training those models.

[0022] In an embodiment of the invention, a single pre-trained sensing system is provided with a dedicated clock and a plurality of untrained sensing systems are provided with a dedicated clock for each plurality of untrained sensing systems. Each of the clocks of the pre-trained sensing system and the untrained sensing system is calibrated to record the same time. An enclosure is provided for the pre-trained sensing system and the untrained sensing system. The enclosure is provided with an input port for the analyte and a flow path that is taken by the analyte to reach from the input port to the sensing element of the pre-trained sensing system and the untrained sensing system. In another embodiment of the invention, the flow path length for the untrained sensing system and the pre -trained sensing system is the same. In yet another embodiment of the invention, the spatial configuration of the flow path for the untrained sensing system and the pre -trained sensing system are the same.

[0023] An onboard power unit may be provided with the auto training and auto-calibration system wherein the power unit may be one of a Li-ion battery, alkaline battery, aluminum-air battery, chromic acid battery, etc.

[0024] For training the sub-systems of the loT platform, the auto-learning and auto-calibration systems may use at least one of timing unit, an onboard pre-trained sensing system, a processing system, and a server that may be present at a remote location and communicates with the autolearning and auto-calibration system using a communication module. In an embodiment of the invention, a second communication module may be provided with the auto-learning and autocalibration system. The second communication module is used for communicating with loT subsystems and to synchronize the loT subsystems with the auto-learning and auto-calibration systems, for example, the clocks of the two systems.

[0025] In an embodiment of the invention, the auto-learning and auto-calibration system, untrained sensing system, trained sensing system, and the loT subsystems are provided with a GPS module and the coordinates of all the mentioned systems are provided along with recorded data from the sensing units for tuning the at least of leaning and calibrating activity through the autolearning and auto-calibrating system.

[0026] The auto-learning and auto-calibrating systems may be provided with at least one of a temperature sensor, humidity sensor, particulate monitor, pressure sensor, and air velocity monitoring system. At least one of a temperature sensor, humidity sensor, particulate monitor, pressure sensor, and air velocity monitoring system may be provided with the auto-sensing and auto-calibrating system, inside the enclosure. Further, at least one of a temperature sensor, humiditysensor, particulate monitor, pressure sensor, and air velocity monitoring system may be provided outside the enclosure and in communication with the auto-learning and auto-calibrating system.

[0027] In an embodiment of the invention, the communication module sends sensed information from at least one of a temperature sensor, humidity sensor, particulate monitor, pressure sensor, and air velocity monitoring system along with the plurality of the pre-trained sensing system, untrained sensing system, and the trained sensing system. The processing system of the remote server uses the data of at least one of a temperature sensor, humidity sensor, particulate monitor, pressure sensor, and air velocity monitoring system to modify the multidimensional response of the untrained and the trained sensing system, before assigning labels to the fingerprint of the response.

[0028] The communication module of the auto training and auto-calibration system is a wireless communication unit, used to communicate with a server that is placed at a remote location, such as a cloud server and edge server. The wireless communication unit of the communication module may use one of WiFi, Bluetooth, Zigbee, and NFC 3G, LTE, 4G, and 5G communication protocols to communicate with the remote server.

[0029] In an embodiment of the invention, the communication module may be connected to the remote server through a wired communication unit, wherein the wired communication unit may connect with one of the nodes of the access network, aggregate network, and core network to communicate with the remote server. The wired communication enablers may be one of the electrical conductors and optical conductors.

[0030] The remote server used by the auto training and auto-calibration system is provided with a communication module alongside the processing unit and memory unit, to communicate with the auto training and auto-calibration system. The remote server takes in the information provided by the plurality of sensing systems, sensors, and monitors from the communication module and may use the communication module to send the labels to the untrained and trained sensing systems.

[0031] The processing unit of the remote server may comprise an Al model to process information / data provided by the plurality of sensing systems, sensors, and monitors to generate labels for single or low -dimensional responses and assign the label to a multi-dimensional response. The Al model may use one or deep learning techniques, neural network models, or one of machine learning algorithms to generate and assign the labels.

[0032] The processing unit of the remote server may generate a label for the single or lowdimensional response of the pre-trained sensing system using prestored threshold values that are stored in the memory unit of the remote server, before starting the auto training and auto-calibration process, wherein each of the labels is a measure of the presence of a certain type of analyte and the concentration of the analyte.

[0033] The processing unit of the remote server may generate a fingerprint for the multidimensional response of a plurality of sensing systems and sensor units present in the auto training and auto-calibration system, other than the pre-trained sensing systems. The processing unit may generate the fingerprint by determining the response of each of the plurality of sensing systems and the sensor units at a certain period. The fingerprint thus generated is a manifold that is spanned over multiple dimensions.

[0034] For each period, the processing unit may generate a label from the single or low dimensional response and assign the label to the multi dimensional manifold, thus training the untrained and calibrating the trained sensing system in a plurality of operating scenarios. Thus, the learning process may be referred to as shadow learning from the pre-trained sensing system.

[0035] Also, the calibrating system may assign the label to a manifold that is generated, in situations when few of the sensors of the sensor array of the trained sensing system fail or degrade, thereby increasing the operating life of the sensing system.

Claims

CLAIMS1. A system for calibrating an enose, the system comprising:A plurality of calibrating apparatus, the calibrating apparatus including a single analyte sensors, an enose, a clock and a plurality of environment sensors; wherein the plurality of calibrating apparatus are deployed in different background environments a memory unit configured to store the response of the single analyte sensor, the enose, the clock and the plurality of environment sensor; a processing unit configured to calibrate the enose, wherein the processing unit uses the data generated by the single analyte sensor from the plurality of calibrating apparatus, deployed in different background environment, as a label and assign it to the response of the enose, thereby calibrating the enose for the analyte.

2. The system of claim 1, wherein the single analyte sensor is another enose that is configured to detect presence and concentration of one analyte in an environment.

3. The system of claim 2, wherein the enose may be configured to detect the presence and concentration of an aroma, wherein an aroma is a mixture of a plurality of volatile organic compounds.

4. The system of claim 1, wherein the processing unit generates a plurality of labels from the single analyte sensor at a plurality of time instances and assigns it to the response of the enose.

5. The system of claim 1, wherein a plurality of concentration of a single analyte is exposed to both the single analyte sensor and the enose in the calibration apparatus and the processing unit generates a calibration curve for the response of the enose, using the labels of the single analyte sensor.

6. A system for training an Al model running on the processor of an enose, the system comprising a plurality of calibration apparatus, the calibrating apparatus including a single analyte sensors, an enose, a clock and a plurality of environment sensors; a memory unit configured to store the response of the single analyte sensor, the enose, the clock and the plurality of environment sensors; a processing unit configured to generate training dataset to the the Al model running on the enose processor, wherein the processing unit uses the data generated by the single analyte sensor from the plurality of calibration apparatus as a label for the data generated by the enose on the calibration apparatus and wherein the Al model is trained on the labeled dataset to identify the analyte in an unseen environment.

7. The system of claim 6, wherein the single analyte sensor is another enose that is configured to detect the presence and concentration of an analyte in an environment.The system of claim 7, wherein the enose may be configured to detect the presence and concentration of an aroma, wherein an aroma is a mixture of a plurality of volatile organic compounds.

8. The system of claim 6, wherein the Al model is trained on a dataset generated by the combination of labels generated by the single analyte sensor and the enose, wherein the labeled dataset is generated at a plurality of time instances during analyte exposure to the analyte sensor and the enose, processing unit generates a labeled dataset from the single analyte sensor at a plurality of time instances.

9. The system of claim 9, wherein the single analyte sensor is another enose that is configured to detect the presence and concentration of an analyte in an environment.

10. The system of claim 10, wherein the enose may be configured to detect the presence and concentration of an aroma, wherein an aroma is a mixture of a plurality of volatile organic compounds.

Images