Detection device for a fluid sample

Abstract

In one aspect, a detection device is described, comprising: at least one sensor configured to detect a signal of a fluid sample comprising at least one analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated from a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors can be the same or different processors. Methods of using detection device for analyzing a fluid sample are also described.

Description

Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025DETECTION DEVICE FOR A FLUID SAMPLERELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Application No. 63 / 698,896, entitled “GAS SENSING TECHNOLOGY FOR INDOOR AIR QUALITY,” filed September 25, 2024, which is incorporated by reference herein in its entirety.GOVERNMENT SUPPORT

[0002] This invention was made with government support under 2344256 awarded by National Science Foundation (NSF). The government has certain rights in this invention.COPYRIGHT NOTICE

[0003] This patent disclosure may contain material that is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure as it appears in the U.S. Patent and Trademark Office patent file or records, but otherwise reserves any and all copyright rights.FIELD OF THE INVENTION

[0004] The present disclosure relates generally to the field of fluid sample analysis. More particularly, the present disclosure relates to the analysis of analytes contained in a fluid sample.BACKGROUND

[0005] Indoor air quality is significantly influenced by the presence of toxic compounds, such as volatile organic compounds (VOCs). VOCs are commonly found in buildings or in the atmosphere. Even at trace concentrations, these toxic compounds can pose serious health risks to humans, contributing to respiratory issues, allergic reactions, and long-term illnesses.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025The impact of VOCs on human health underscores the importance of monitoring and controlling air quality to ensure safe and healthy living and working conditions.

[0006] While traditional air quality sensors offer a platform for the detection of toxic compounds, their performance is notoriously non-specific and lack accuracy. These sensors often analyze the fluid sample by collecting static signals without considering the signal changes over time. Consequently, the results of traditional air quality sensors often have low sensitivity, low reliability, substantial drift, and nonlinearity. These issues are especially troublesome when the sensors respond to multiple toxic compounds at the same time.

[0007] Another option for toxic compound detection is to use high-end analytical instrumentation, such as gas chromatography-mass spectrometry, or commercial plug-and- play VOC detectors that employ narrowly tuned sensor modules. The high-end analytical instrumentation delivers excellent sensitivity, but at prohibitive cost, size, and complexity. The commercial plug-and-play VOC detectors suffer from poor selectivity, calibration instability, and limited interpretability.

[0008] There is increasing interest in advancing the performance capabilities of detection devices to meet the growing demands of modern applications, while maintaining costeffectiveness. As technologies evolve and expectations rise across industries such as environmental monitoring, healthcare, and industrial safety, detection systems must deliver improved functionality without becoming prohibitively expensive. Enhancements in detection of these VOCs are important to ensure these devices remain practical and accessible for widespread use.SUMMARY

[0009] In one aspect, a detecting device is described, comprising at least one sensor configured to detect a signal of a fluid sample comprising at least one analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated from a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; andAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors can be the same or different processors.

[0010] In any one of the embodiments disclosed herein, the at least one sensor comprises an array of sensors.

[0011] In any one of the embodiments disclosed herein, the sensors in the array of sensors are parallelly or sequentially connected.

[0012] In any one of the embodiments disclosed herein, the detection device further comprises a redundant sensor configured to replace the at least one sensor in case of failure.

[0013] In any one of the embodiments disclosed herein, the detection device is configured to connect or disconnect the redundant sensor.

[0014] In any one of the embodiments disclosed herein, the detection device is a handhold gadget, a tabletop station, or an in-line apparatus of an industrial process.

[0015] In any one of the embodiments disclosed herein, the at least one sensor is selected from the group consisting of a chemiresi stive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, and a hybrid multi-modal sensor.

[0016] In any one of the embodiments disclosed herein, the gravimetric sensor is a quartz crystal microbalance sensor.

[0017] In any one of the embodiments disclosed herein, the at least one sensor is a chemiresi stive sensor.

[0018] In any one of the embodiments disclosed herein, the at least one sensor is a chemiresistive metal-oxide sensor, a conductive polymer sensor, a carbon nanotube sensor, or a graphene sensor.

[0019] In any one of the embodiments disclosed herein, the at least one sensor is an optical sensor selected from the group consisting of a fluorescence sensor, a reflection sensor, a surface plasmon resonance sensor, and a Raman sensor.

[0020] In any one of the embodiments disclosed herein, the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, a MQ135 sensor, and a MiCS 5914 sensor.

[0021] In any one of the embodiments disclosed herein, the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, and a MiCS 5914 sensor.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0022] In any one of the embodiments disclosed herein, the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, and a MQ135 sensor.

[0023] In any one of the embodiments disclosed herein, the fluid sample comprises a liquid, a vapor, and / or a solid.

[0024] In any one of the embodiments disclosed herein, the fluid sample is a vapor of a volatile compound mixed in a gaseous sample.

[0025] In any one of the embodiments disclosed herein, the fluid sample is a vapor of a volatile compound, or a water vapor mixed in a gaseous sample.

[0026] In any one of the embodiments disclosed herein, the analyte is a volatile compound.

[0027] In any one of the embodiments disclosed herein, the analyte is selected from the group consisting of benzene, formaldehyde, toluene, xylenes, BTEX, methane, propane, a sulfur gas, ammonia, nitrogen dioxide, hydrogen, an alcohol, smoke, carbon monoxide, carbon dioxide, or a combination thereof.

[0028] In any one of the embodiments disclosed herein, the detection device further comprises a humidity sensor to detect humidity of the fluid sample.

[0029] In any one of the embodiments disclosed herein, the humidity sensor is configured to detect the humidity level from about 0% to about 90%.

[0030] In any one of the embodiments disclosed herein, the detective device is configured to detect at a temperature range from -450E to 3000F.

[0031] In any one of the embodiments disclosed herein, the first or second processor is configured to execute a machine learning model to reduce a humidity -induced detection error of the at least one sensor.

[0032] In any one of the embodiments disclosed herein, the first processor is configured to process the readouts.

[0033] In any one of the embodiments disclosed herein, the second processor is configured to execute a machine learning model.

[0034] In any one of the embodiments disclosed herein, the machine learning model is a linear regression model, a support vector machine model, a support vector classifier model, a random forest model, or a deep neural network model.

[0035] In any one of the embodiments disclosed herein, the memory is configured to be capable of storing additional temporal profile of new known analyte to the database.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0036] In any one of the embodiments disclosed herein, the memory is configured to be trained by a machine learning model to add additional temporal profile of new known analyte to the database.

[0037] In any one of the embodiments disclosed herein, the fluid sample comprises at least two analytes; and the second processor is configured to: deconvolute the plurality of readouts of the fluid sample to generate a temporal profile for each of the at least two analytes; and compare each of the temporal profile with the database to determine the identity and / or concentration of each of the at least two analytes.

[0038] In any one of the embodiments disclosed herein, the detection device further comprises an additional sensor configured to detect an additional signal of the fluid sample at the plurality of time points within the time period to generate an additional plurality of readouts of the additional signal each measured at one of the plurality of time points within the time period; the first processor is configured to generate an additional temporal profile of the additional plurality of readouts; and the second processor is configured to compare the additional temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample.

[0039] In any one of the embodiments disclosed herein, the second processor is configured to determine the identity of the analyte by a linear regression model or a support vector regression model.

[0040] In any one of the embodiments disclosed herein, the second processor is configured to determine the concentration of the analyte by a linear regression model or a support vector regression model.

[0041] In any one of the embodiments disclosed herein, the temporal profile is converted to a leaflet comprising a closed-loop trajectory formed by projecting the temporal signal into a principal component space.

[0042] In any one of the embodiments disclosed herein, the at least one sensor is configured to allow the adsorption and desorption of the analyte over a surface of the at least one sensor within the time period, and the leaflet is configured to reflect the adsorption and desorption process.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0043] In any one of the embodiments disclosed herein, the leaflet’s orientation is configured to reflect a pathway of the adsorption and desorption of the analyte over the surface of the at least one sensor.

[0044] In any one of the embodiments disclosed herein, the direction of the leaflet reflects an identity of the analyte.

[0045] In any one of the embodiments disclosed herein, a curvature of the leaflet is configured to reflect a hysteresis and recovery kinetics of the adsorption and desorption of the analyte over the surface of the at least one sensor.

[0046] In any one of the embodiments disclosed herein, the geometrical resolution of the leaflet is configured to be higher with increased numbers of readouts at the time points within the time period.

[0047] In any one of the embodiments disclosed herein, the enclosed area of the leaflet is configured to be proportional with a concentration of the analyte.

[0048] In any one of the embodiments disclosed herein, the detection device further comprises an interface for inputting user instructions to control the at least one sensor and / or the first and / or second processors.

[0049] In any one of the embodiments disclosed herein, further comprises a communication module configured to communicate the generated temporal profile and the determined identity and / or concentration of the analyte to a user.

[0050] In any one of the embodiments disclosed herein, the fluid sample is an air sample from a building, a ship, a hotel, a hospital, an industrial site, an industrial process, a farm, a wildfire detection site, a breath sample, or an atmosphere.

[0051] In any one of the embodiments disclosed herein, the time period is about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or 15 seconds.

[0052] In any one of the embodiments disclosed herein, the time period is about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or 15 minutes.

[0053] In any one of the embodiments disclosed herein, the at least one sensor’s frequency ranges from 1 Hz to 1G Hz.

[0054] In any one of the embodiments disclosed herein, the time period comprises 5, 10, 20, 30, 40, 50, 60, 70, 80, 90 or 100 time points.

[0055] In any one of the embodiments disclosed herein, the time period comprises 100, 500, 1000, 1500, 2000, 2500, 3000, 3500, 4000, 4500, 5000, 8000, 10000, 50000, 100000,Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025200000, 500000, 1 million, 1.5 million, 5 million, 10 million, 15 million, or 20 million time points.

[0056] In any one of the embodiments disclosed herein, the detection device further comprises a gas-flow device configured to deliver the fluid sample to the at least one sensor.

[0057] In any one of the embodiments disclosed herein, the detection device further comprises a gas-removal device to remove the fluid sample from the surface of the at least one sensor.

[0058] In another aspect, a detection device is disclosed, comprising an array of sensors configured to detect an analyte in a fluid sample, comprising: a first sensor selected based on a first property in its detection of the analyte; and a second sensor selected based on a second property different from the first property in its detection of the analyte; wherein the first and second properties are each selected from the group consisting of limit of detection, sensitivity, reliability, reproducibility, and linearity in detection of the analyte.

[0059] In any one of the embodiments disclosed herein, the first sensor’s first property is superior to the second sensor’s first property in detecting the analyte.

[0060] In any one of the embodiments disclosed herein, the second sensor’s second property is superior to the first sensor’s second property in detecting the analyte.

[0061] In any one of the embodiments disclosed herein, the detection device further comprises a third sensor selected based on a third property in its detection of the analyte, wherein the third property is selected from the group consisting of limit of detection, sensitivity, reliability, and linearity in detection of the analyte.

[0062] In any one of the embodiments disclosed herein, the limit of detection is from about 0.1 ppb to about 150 ppm.

[0063] In any one of the embodiments disclosed herein, the limit of detection is from about 0.01 ppb to about 50 ppb.

[0064] In any one of the embodiments disclosed herein, the limit of detection is from about 0.01 ppb to about 0.5 ppb.

[0065] In any one of the embodiments disclosed herein, the analyte is benzene, formaldehyde, toluene, or BTEX.

[0066] In any one of the embodiments disclosed herein, the sensitivity is measured by a slope of the calibration curve.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0067] In any one of the embodiments disclosed herein, the slope of the calibration curve is from about 0.05 to 0.99.

[0068] In any one of the embodiments disclosed herein, the reliability is measured by signal-to-noise ratio.

[0069] In any one of the embodiments disclosed herein, the signal-to-noise ratio is from about 0.0001 to about 0.18.

[0070] In any one of the embodiments disclosed herein, the linearity is measured by a coefficient of determination.

[0071] In any one of the embodiments disclosed herein, the coefficient of determination is from about 0.26 to 0.999.

[0072] In any one of the embodiments disclosed herein, the first and / or second sensors are configured to detect the identity or concentration of the analyte.

[0073] In any one of the embodiments disclosed herein, the analyte is one or more volatile organic compounds.

[0074] In any one of the embodiments disclosed herein, the analyte is selected from the group consisting of benzene, formaldehyde, toluene, xylenes, BTEX, methane, propane, a sulfur gas, ammonia, nitrogen dioxide, hydrogen, an alcohol, smoke, carbon monoxide, carbon dioxide, or a combination thereof.

[0075] In any one of the embodiments disclosed herein, the analyte is benzene, toluene, formaldehyde, BTEX, or a combination thereof.

[0076] In any one of the embodiments disclosed herein, the at least one sensor is selected from the group of a chemiresistive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, and a hybrid multi-modal sensor.

[0077] In any one of the embodiments disclosed herein, the first and second sensors are chemiresistive sensors.

[0078] In any one of the embodiments disclosed herein, the first and second sensors are chemiresistive metal-oxide sensors.

[0079] In any one of the embodiments disclosed herein, the first and second sensors are each selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a MQ138 sensor, a MiCS5524 sensor, and a SEN0566 sensor, and a SCD40 sensor.

[0080] In any one of the embodiments disclosed herein, the first sensor is a SGP41 sensor selected for its reproducibility and low limitation of detection of the analyte.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0081] In any one of the embodiments disclosed herein, the first sensor is a ENS160 sensor selected for its heightened sensitivity in detecting the fluid sample.

[0082] In any one of the embodiments disclosed herein, the first sensor is a MQ138 sensor selected for its high linearity in detecting the fluid sample to provide complimentary properties different from the properties from the second sensor.

[0083] In any one of the embodiments disclosed herein, the first sensor is a MiCS5524 sensor selected for its low limit of detection and high linearity in detecting the fluid sample.

[0084] In any one of the embodiments disclosed herein, the first sensor is a SEN0566 sensor selected for its high linearity in detecting the fluid sample.

[0085] In any one of the embodiments disclosed herein, the detection device comprises a SGP41 sensor and a ENS 160 sensor.

[0086] In any one of the embodiments disclosed herein, further comprises a SCD 40 sensor.

[0087] In any one of the embodiments disclosed herein, further comprises a MQ138-B sensor, a SEM0566-B sensor, and / or a MiCS5524-B sensor.

[0088] In any one of the embodiments disclosed herein, further comprises a MiCS5524-B sensor and a MiCS5914 sensor.

[0089] In any one of the embodiments disclosed herein, the array of sensors comprises a SGP41 sensor, a MiCS5524 sensor, a ENS160 R2 sensor, a ENS160 R3 sensor, and a MQ138 sensor.

[0090] In any one of the embodiments disclosed herein, at least one of the first sensor and the second sensor is configured to detect a signal of the analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; and the detection device further comprises a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated by a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; and a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors are the same or different processors.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0091] In yet another aspect, a method of analyzing a fluid sample is described, comprising: providing a detection device of any of the embodiments disclosed herein, detecting the signal of the fluid sample at the plurality of time points within the time period to generate the plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; generating the temporal profile of the readouts by the first processor; and comparing the generated temporal profile of the fluid sample with the database by the second processor to determine the identity and / or concentration of the analyte in the fluid sample.

[0092] In any one of the embodiments disclosed herein, the fluid sample comprises at least two analytes; and the method further comprises: deconvolute the generated temporal profile of the fluid sample to generate a temporal profile for each of the at least two analytes; and comparing the temporal profile of each of the at least two analytes with the database to determine the identity and / or concentration of each of the at least two analytes.

[0093] In yet another aspect, a method of detecting an analyte in a fluid sample is described, comprising: providing the detection device of any of the embodiments disclosed herein, detecting the analyte by the first sensor; and detecting the analyte by the second sensor.

[0094] In any one of the embodiments disclosed herein, further comprising determining the limit of detection, sensitivity, reliability, reproducibility, or linearity in detection of the analyte by the first or second sensor.

[0095] In any one of the embodiments disclosed herein, the method further comprises: detecting a signal of the analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal measured at one of the plurality of the time point within the time period by the first or second sensor; generating a temporal profile of the readouts by a first processor; and comparing the generated temporal profile of the fluid sample with the database by a second processor to determine the identify and / or concentration of the analyte in the fluid sample.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0096] While multiple embodiments are disclosed, still other embodiments of the present disclosure will become apparent to those skilled in the art from the following detailed description, which shows and describes illustrative embodiments of the disclosure.Accordingly, the drawings and detailed descriptions are to be regarded as illustrative in nature and not restrictive.BRIEF DESCRIPTION OF THE DRAWINGS

[0097] For a more complete understanding of various embodiments of the disclosed subject matter, reference is now made to the following descriptions taken in connection with the accompanying drawings, in which:

[0098] Fig. 1 A illustrates a detection device comprising an array of sensors, according to one or more embodiments disclosed herein.

[0099] Fig. IB illustrates an experiment setup comprising a gas delivery system to provide the fluid sample and a detection device to detect the fluid sample, according to one or more embodiments disclosed herein.

[0100] Fig. 1C illustrates an experiment setup comprising a gas delivery system with a humidifier to provide the fluid sample with humidity, and a detection device to detect the fluid sample, according to one or more embodiments disclosed herein.

[0101] Fig. 2A-1 illustrates temporal profiles of a SGP41 sensor exposed to formaldehyde in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0102] Fig. 2A-2 illustrates temporal profiles of an ENS160 R2 sensor exposed to formaldehyde in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0103] Fig. 2A-3 illustrates temporal profiles of an MiCS5524 sensor exposed to formaldehyde in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0104] Fig. 2A-4 illustrates temporal profiles of an ENS160 R0 sensor exposed to formaldehyde in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0105] Fig. 2B-1 illustrates temporal profiles of a SGP41 sensor exposed to benzene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0106] Fig. 2B-2 illustrates temporal profiles of an ENS160 R2 sensor exposed to benzene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0107] Fig. 2B-3 illustrates temporal profiles of a MQ138 sensor exposed to benzene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0108] Fig. 2B-4 illustrates temporal profiles of an MiCS5524 sensor exposed to benzene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0109] Fig. 2B-5 illustrates temporal profiles of an ENS160 R0 sensor exposed to benzene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0110] Fig. 2C-1 illustrates temporal profiles of a SGP41 sensor exposed to toluene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0111] Fig. 2C-2 illustrates temporal profiles of an ENS160 R3 sensor exposed to toluene in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0112] Fig. 2D-1 illustrates temporal profiles of a SGP41 sensor exposed to BTEX in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0113] Fig. 2D-2 illustrates temporal profiles of an ENS160 R2 sensor exposed to BTEX in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0114] Fig. 2D-3 illustrates temporal profiles of a MiCS5524 sensor exposed to BTEX in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0115] Fig. 2D-4 illustrates temporal profiles of an ENS160 RO sensor exposed to BTEX in a fluid sample at various concentrations, with lines representing mean and shaded regions indicate standard deviation across multiple trials, according to one or more embodiments disclosed herein.

[0116] Fig. 3A illustrates log-linear calibration curves of sensor response as a function of concentration of formaldehyde using a variety of sensors, according to one or more embodiments disclosed herein.

[0117] Fig. 3B illustrates log-linear calibration curves of sensor response as a function of concentration of benzene using a variety of sensors, according to one or more embodiments disclosed herein.

[0118] Fig. 3C illustrates log-linear calibration curves of sensor response as a function of concentration of toluene using a variety of sensors, according to one or more embodiments disclosed herein.

[0119] Fig. 3D illustrates log-linear calibration curves of sensor response as a function of concentration of BTEX using a variety of sensors, according to one or more embodiments disclosed herein.

[0120] Fig. 4A illustrates temporal profiles of a SGP41 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.

[0121] Fig. 4B illustrates temporal profiles of a ENS160 R0 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.

[0122] Fig. 4C illustrates temporal profiles of a ENS160 R2 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0123] Fig. 4D illustrates temporal profiles of a ENS160 R3 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.

[0124] Fig. 4E illustrates temporal profiles of a MQ138 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.

[0125] Fig. 4F illustrates temporal profiles of a MICS5524 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.

[0126] Fig. 4G illustrates temporal profiles of a SEN0566 sensor exposed to benzene in a fluid sample at a variety of humidity levels, according to one or more embodiments disclosed herein.

[0127] Fig. 4H illustrates the deviation of responses of a variety of sensors to benzene in a fluid sample at an elevated humidity, according to one or more embodiments disclosed herein.

[0128] Fig. 41 illustrates a standard linear regression of the response of a variety of sensors to an analyte in a fluid sample at 10% room humidity, where the top bar graph of Fig. 41 indicates the relative root-mean-square errors (RMSEs) induced by humidity and the bottom of Fig. 41 shows the log scale sensor response at various relative humidity levels, according to one or more embodiments disclosed herein.

[0129] Fig. 4J illustrates a regression using a machine learning inference model of the response of a variety of sensors to an analyte in a fluid sample at 10% room humidity, where the top bar graph of Fig. 4J indicates the relative root-mean-square errors (RMSEs) induced by humidity and the bottom of Fig. 4J shows the log scale sensor response at various relative humidity levels, according to one or more embodiments disclosed herein.

[0130] Fig. 5 A illustrates a temporal profile of a SGP41 sensor response to 25 ppb formaldehyde in a detection device, according to one or more embodiments disclosed herein.

[0131] Fig. 5B illustrates a principal component analysis (“PCA”) graph showing separation of benzene, toluene, and formaldehyde at 0.25 ppm, according to one or more embodiments disclosed herein.

[0132] Fig. 5C illustrates principal component analysis (“PCA”) trajectories showing compound response patterns shift with increasing concentration, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0133] Fig. 5D illustrates temporal profiles of benzene detected by various sensors, highlighting distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0134] Fig. 5E illustrates temporal profiles of benzene detected by various sensors, highlighting distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0135] Fig. 5F illustrates temporal profiles of formaldehyde detected by various sensors, highlighting distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0136] Fig. 5G illustrates temporal profiles of formaldehyde detected by various sensors, highlighting distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0137] Fig. 5H illustrates temporal profiles of toluene detected by various sensors, highlighting distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0138] Fig. 51 illustrates temporal profiles of toluene detected by various sensors, highlighting distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0139] Fig. 5J illustrates a correlation map of sensor-sensor and sensor-compound relationships, indicating response redundancy and specificity, according to one or more embodiments disclosed herein.

[0140] Fig. 5K illustrates a confusion matrix showing 100% classification accuracy using support vector classifier (“SVC”), according to one or more embodiments disclosed herein.

[0141] Fig. 5L illustrates regression results for a variety of compounds using support vector regression (“SVR”), with root-mean-square errors (RMSEs) of 6.1 ppb (benzene), 20.7 ppb (formaldehyde), and 142 ppb (toluene), according to one or more embodiments disclosed herein.

[0142] Fig. 6A illustrates temporal profiles of benzene, toluene, and formaldehyde detected by SGP41 sensor, showing distinct adsorption / desorption dynamics, according to one or more embodiments disclosed herein.

[0143] Fig. 6B illustrates classification accuracy using static data analysis using random forest models, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0144] Fig. 6C illustrates classification accuracy using time-dependent analysis using random forest models, according to one or more embodiments disclosed herein.

[0145] Fig. 6D illustrates regression of benzene concentrations showing accuracy and variance without time-resolved (temporal) features.

[0146] Fig. 6E illustrates regression of benzene concentrations showing major gains in accuracy and variance reduction with time-resolved (temporal) features, according to one or more embodiments disclosed herein.

[0147] Fig. 7A is a schematic illustration of a closed-loop trajectories (leaflet) of the temporal profile projected into principal component space, according to one or more embodiments disclosed herein.

[0148] Fig. 7B-1 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 16 repeat trials at a concentration of 0.056 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0149] Fig. 7B-2 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 16 repeat trials at a concentration of 0.1 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0150] Fig. 7B-3 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 16 repeat trials at a concentration of 0.025 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0151] Fig. 7B-4 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 17 repeat trials at a concentration of 0.5 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0152] Fig. 7B-5 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 17 repeat trials at a concentration of 1.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0153] Fig. 7C-1 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 17 repeat trials at a concentration of 0.28 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0154] Fig. 7C-2 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 13 repeat trials at a concentration of 0.5 ppm projected into principal component space, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0155] Fig. 7C-3 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 13 repeat trials at a concentration of 1.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0156] Fig. 7C-4 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 13 repeat trials at a concentration of 2.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0157] Fig. 7C-5 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 13 repeat trials at a concentration of 5.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0158] Fig. 7D-1 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 12 repeat trials at a concentration of 0.025 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0159] Fig. 7D-2 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 12 repeat trials at a concentration of 0.04 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0160] Fig. 7D-3 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 12 repeat trials at a concentration of 0.07 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0161] Fig. 7D-4 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 12 repeat trials at a concentration of 0.1 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0162] Fig. 7D-5 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 11 repeat trials at a concentration of 0.25 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0163] Fig. 7E-1 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 11 repeat trials at a concentration of 0.28 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0164] Fig. 7E-2 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 12 repeat trials at a concentration of 0.5 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0165] Fig. 7E-3 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 12 repeat trials at a concentration of 1.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0166] Fig. 7E-4 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 11 repeat trials at a concentration of 2.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0167] Fig. 7E-5 illustrates closed-loop trajectories (leaflet) of the temporal profile of toluene from 11 repeat trials at a concentration of 5.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0168] Fig. 7F-1 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 11 repeat trials at a concentration of 0.025 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0169] Fig. 7F-2 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 12 repeat trials at a concentration of 0.04 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0170] Fig. 7F-3 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 12 repeat trials at a concentration of 0.07 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0171] Fig. 7F-4 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 11 repeat trials at a concentration of 0.1 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0172] Fig. 7F-5 illustrates closed-loop trajectories (leaflet) of the temporal profile of formaldehyde from 11 repeat trials at a concentration of 0.25 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0173] Fig. 7G-1 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 14 repeat trials at a concentration of 0.056 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0174] Fig. 7G-2 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 14 repeat trials at a concentration of 0.1 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0175] Fig. 7G-3 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 16 repeat trials at a concentration of 0.25 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0176] Fig. 7G-4 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 15 repeat trials at a concentration of 0.5 ppm projected into principal component space, according to one or more embodiments disclosed herein.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0177] Fig. 7G-5 illustrates closed-loop trajectories (leaflet) of the temporal profile of benzene from 16 repeat trials at a concentration of 1.0 ppm projected into principal component space, according to one or more embodiments disclosed herein.

[0178] Fig. 7H illustrates trajectory area of benzene, formaldehyde, and toluene at various concentrations, according to one or more embodiments disclosed herein.

[0179] Fig. 71 illustrates distribution of trajectory directions (theta) of benzene, formaldehyde, and toluene at various concentrations, according to one or more embodiments disclosed herein.DETAILED DESCRIPTIONA gas detection device

[0180] In one aspect, a detection device is described, comprising at least one sensor configured to detect a signal of a fluid sample comprising at least one analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated from a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; and a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors can be the same or different processors.

[0181] In some embodiments, the detection device is a low-cost electronic device designed for real-time air quality monitoring. In some embodiments, the detection device comprises at least one sensor. In some embodiments, the at least one sensor comprises an array of sensors. In some embodiments, the sensors in the array of sensors are connected in a parallel architecture, a sequential architecture, or a combination of parallel and sequential architecture.

[0182] In some embodiments, the detection device may be a hand-hold detection device, a portable detection device, a tabletop fluid detection station, or a detection device integrated in-line in an industrial process. In some embodiments, the sensor and the first and secondAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 processors of the detection device may be positioned in proximity. In some embodiments, the sensor and the first and second processors of the detection device may be positioned in different locations. In some embodiments, the detective device comprises more than one sensors and the sensors may be positioned in different locations. In some embodiments, the sensor and the first and second processors are positioned within or close to the fluid sample, while the first and second processors are located close to the user. In some embodiments, the sensor and the first and second processors may communicate the detected signal, the analysis, and the user instructions. In some embodiments, the sensor and the first and second processors communicate wirelessly or through physical connection.

[0183] In some embodiments, the detection device is also referred to as electronic nose (e-nose), gas sensor array, or Sniffia.

[0184] In some embodiments, the array of sensors has an architecture comprising a redundancy design where at least one sensor is a duplicated design for another sensor to ensure the functionality of the detection device. In some embodiments, in the case where one of the sensors fails, the detection device uses the remaining functional sensor (e.g., a duplicated or redundant sensor) to detect and analyze the analytes.

[0185] In some embodiments, the redundancy design of the detection device is tunable. In these embodiments, a user of the detection device may select to use or bypass the redundant sensor. In these embodiments, a user of the detection device may select to connect or disconnect the redundant sensor. In these embodiments, some sensors may have more than one duplicated sensor in the redundancy design. In these embodiments, a user of the detection device may select one or more duplicated or redundant sensors according to the importance of the sensor in the detection device.

[0186] In some embodiments, the at least one sensor is selected from the group consisting of a chemiresi stive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, and a hybrid multi-modal sensor. In some embodiments, the sensor is a chemiresi stive sensor, such as a metal oxide sensor, a conductive polymer sensor, or a carbon nanotub e / graphene sensor. In some embodiments, the sensor is an optical sensor. Nonlimiting examples of an optical sensor include a fluorescence sensor, a reflection sensor, a surface plasmon resonance (“SPR”) sensor, or a Raman sensor. In some embodiments, the sensor is a gravimetric sensor. Non-limiting examples of an optical sensor include a quartz crystal microbalance (“QCM”) sensor. In some embodiments, the sensor is an electrochemical sensor or a hybrid transducer (e.g., a hybrid multi-modal sensor).Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0187] In some embodiments, the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a MQ138 sensor, a MiCS5524 sensor, a SEN0566 sensor, and a SCD40 sensor.

[0188] In some embodiments, the at least one sensor is selected from the group consisting of a SGP41 sensor, a MiCS5524 sensor, a ENS160 R2 sensor, a ENS160 R3 sensor, and a MQ138 sensor.

[0189] In some embodiments, the fluid sample is a liquid sample, a gaseous sample, or a mixture thereof. In some embodiments, the fluid sample is an air sample from a building, a ship, a hotel, a hospital, an industrial site, an industry process, a manufacturing plant, a breath sample, an outdoor agriculture site, a farm, a wildfire monitoring site, a food processing facility, or an atmosphere. In some embodiments, the detection device is configured to detect analyte in a fluid sample to analyze indoor air quality, such as air quality in buildings, ships, hotels, industry processes, and hospitals. In other embodiments, the detection device is configured to detect analyte in a fluid sample to analyze industrial safety and leak detection (e.g., in an industrial site). In these embodiments, the analytes comprise methane, sulfur gases, or ammonia. In some embodiments, the detection device is configured to detect analyte in a fluid sample to conduct environmental monitoring, such as greenhouse gas mapping. In some embodiments, the detection device is configured to detect analyte in a fluid sample to conduct medical diagnostics, such as detecting analyte from a person’s breath to conduct breath analysis. In some embodiments, the detection device is configured to detect food spores in a food processing facility.

[0190] In some embodiments, the fluid sample comprises a gaseous sample. In some embodiments, the fluid sample comprises an analyte in carrier gas. In some embodiments, the carrier gas is air, nitrogen, other inert gas, or a combination thereof. In some embodiments, the fluid sample comprises one or more volatile compounds (VCs) in air. In some embodiments, the fluid sample comprises one or more volatile organic compounds (VOCs) in air. In some embodiments, the fluid sample comprises gas, vapor, liquid, solid or a combination thereof. In some embodiments, the fluid sample comprises a vapor of a VC, a VOC, an organic compound, or an inorganic compound. In some embodiments, the fluid sample comprises a mixture of liquid and gas. In some embodiments, the fluid sample is a vapor of a volatile organic compound mixed in a gaseous sample. In some embodiments, the fluid sample is a vapor of a volatile organic compound, or a water vapor mixed in a gaseous sample. In some embodiments, the fluid sample comprises a liquid, such as a liquid state VCAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 or VOC. Non-limiting examples of liquid state VOC are liquid phenol, liquid methylene chloride, liquid toluene, liquid BTEX, and a combination thereof. In some embodiments, the liquid is droplets dispersed in a gaseous sample or a vapor. In some embodiments, the fluid sample comprises droplets dispersed in a gaseous sample. In some embodiments, the fluid sample comprises a solid, such as a spore, a particle, a dust, a fine powder, or a combination thereof. In some embodiments, the fluid sample includes benzene, formaldehyde, and / or toluene in air. In some embodiments, the fluid sample includes benzene, formaldehyde, toluene, and / or water vapor (or water droplets) in air. In some embodiments, the fluid sample includes benzene, formaldehyde, toluene, and / or BTEX in air. In some embodiments, the fluid sample includes benzene, formaldehyde, toluene, BTEX and / or water vapor (or water droplets) in air.

[0191] In some embodiments, the fluid sample comprises at least one analyte. In some embodiments, the analyte may be a chemical compound. In some embodiments, the analyte may be in vapor, liquid (e.g., droplets), or solid form. In some embodiments, the analyte may be one or more volatile compounds (VCs). In some embodiments, the analyte may be one or more volatile organic compounds (VOCs). In some embodiments, the analyte may be benzene, formaldehyde, toluene, xylenes, BTEX, methane, propane, a sulfur gas, ammonia, nitrogen dioxide, hydrogen, an alcohol, smoke, carbon monoxide, carbon dioxide, or a combination thereof. As used herein, the term BTEX refers to a group of VOCs comprising benzene, toluene, ethylbenzene and xylene (e.g., m-xylene and o-xylene). In some specific embodiments, BTEX is a mixture of 1 ppm benzene, 10 ppm toluene, 10 ppm ethylbenzene, 20 ppm m-xylene, and 20 ppm o-xylene.

[0192] In some embodiments, the fluid sample is detected and analyzed at a relative humidity of about 0%, about 10%, about 20%, about 30%, about 40%, about 50%, about 60%, about 70%, about 80%, about 90%, about 100%, or any relative humidity level in a range bounded by any two values disclosed here. In some embodiments, the water in the fluid sample may be in the form of water vapor. In some embodiments, the water in the fluid sample may be in the form of water droplets.

[0193] In some embodiments, the relative humidity may attenuate the sensor response and thereby introduce detection error on the identification and quantification of the analytes in the fluid sample. In some embodiments, the detection device further comprises a humidity sensor to detect humidity of the fluid sample. In some embodiments, the humidity sensor is configured to detect the humidity level at about 0%, about 10%, about 20%, about 30%,Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 about 40%, about 50%, about 60%, about 70%, about 80%, about 90%, about 100%, or any relative humidity level in a range bounded by any two values disclosed here.

[0194] In some embodiments, the fluid sample is detected and analyzed at a temperature of about -459°F, -450F, -400F, -350F, -300F, -150F, -100F, -50F, -40F, -30F, about -20F, about -10F, about OF, about 10F, about 20F, about 30F, about 40F, about 50F, about 60F, about 70F, about 80F, about 90F, about 100F, about 150F, about 300F, about 500F, about 800F, about lOOOF, about 1500F, about 2000F, about 2500F, about 3000F, or about 4000F, or any temperature level in a range bounded by any two values disclosed here. In one specific embodiment, the fluid sample is detected at -454.8F. In one specific embodiment, the fluid sample is detected at 2800F. In some embodiments, the temperature or temperature fluctuation may affect the sensor response and thereby introduce detection error on the identification and quantification of the analytes in the fluid sample. In some embodiments, the detection device further comprises a temperature sensor to detect temperature of the fluid sample.

[0195] In some embodiments, the first or second processor is configured to execute a machine learning model to reduce a humidity-induced and / or temperature-induced detection error of the at least one sensor. In some embodiments, non-limiting suitable machine learning models include linear models, such as linear regression, ridge regression, lasso regression, and elastic net regression. In some embodiments, non-limiting suitable machine learning models include tree-based models, such as decision trees, random forests, gradient boosting, and extreme gradient boosting (XGBoost). In some embodiments, non-limiting suitable machine learning models include kernel-based models, such as support vector regression (SVR), support vector classification (SVC), and other support vector machine (SVM) approaches. In some embodiments, non-limiting suitable machine learning models include probabilistic models, such as Gaussian process regression, Bayesian regression, and naive Bayes classifiers. In some embodiments, non-limiting suitable machine learning models include ensemble models that integrate multiple approaches through bagging, boosting, or stacking. In some embodiments, neural network models may also be employed. In some embodiments, non-limiting examples of neural network models include feed-forward artificial neural networks, convolutional neural networks (CNNs), recurrent neural networks (RNNs), gated recurrent unit (GRU) networks, long short-term memory (LSTM) networks, and transformer-based architectures. In some embodiments, unsupervised learning techniques such as principal component analysis (PCA), independent component analysis (ICA),Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 clustering algorithms, and / or autoencoders can be applied for feature extraction, dimensionality reduction, or denoising. It is noted that the embodiments listed herein are not meant to be limiting. Other statistical or machine learning models may be used for processing the temporal data collected from the sensor(s). In some embodiments, combinations of the afore mentioned models are used, either sequentially or in parallel, to further improve analyte detection accuracy and robustness under varying environmental conditions.

[0196] In some embodiments, the first or second processor of the detection device is configured to execute a machine learning model to reduce a humidity-induced and / or temperature-induced detection error of the at least one sensor. In some embodiments, the machine learning model is trained using known analytes across a range of temperature and humidity levels. In some embodiments, the trained model may subsequently be applied to detect characteristics of unknown analytes, including identity and concentration, within such environmental ranges, provided that the influence of humidity and temperature on sensor response is consistent. In some embodiments, suitable machine learning models include, but are not limited to: linear models such as linear regression, such as ridge regression, lasso regression, and elastic net; tree-based models, such as decision trees, random forests, gradient boosting machines, extreme gradient boosting (XGBoost), LightGBM, and CatBoost; kernelbased models, such as support vector machines (SVM), support vector regressors (SVR), and support vector classifiers (SVC); probabilistic models, such as Gaussian process regression, Bayesian regression, hidden Markov models (HMMs), and naive Bayes classifiers; neural network models, such as feed-forward artificial neural networks, convolutional neural networks (CNNs), recurrent neural networks (RNNs), long short-term memory (LSTM) networks, gated recurrent units (GRUs), and transformer-based architectures; ensemble models, such as bagging, boosting, and stacking methods; and unsupervised learning methods, such as principal component analysis (PCA), independent component analysis (ICA), clustering algorithms (k-means, hierarchical clustering, DBSCAN), and autoencoders. In some embodiments, hybrid approaches that combine multiple models (e.g., PCA or autoencoder feature extraction followed by a random forest or deep neural network) may be employed to further improve robustness. In some embodiments, the foregoing list of models is non-exhaustive, and any statistical or machine learning model suitable for processing sensor data may be used to reduce environmental interference and enhance analyte detection.

[0197] In some embodiments, the first sensor is configured to detect a signal of a fluid sample comprising at least one analyte. In some embodiments, the first sensor generates aAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 plurality of readouts of the signal each measured at a plurality of time points within a time period.

[0198] In some embodiments, the time period may be about 0.01, 0.05, 0.1, 0.5, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 30, 50, or 100 seconds, or any time period in a range bounded by any two values disclosed herein. In some embodiments, the sensor may have a frequency of about 0.01 Hz to about 0.1 kHz, about 0.1 Hz to about 1 kHz, about 1 Hz to about 1 kHz, about 1kHz to about 1MHz, about 1MHz to about 1GHz., or about 1G Hz to about 100GHz. In some embodiments, the time period may be about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 25, 50, 100, 200, 500 minutes, or any time period in a range bounded by any two values disclosed herein. In some embodiments, the time period needed is analyte dependent. In some embodiments, the analyte may have faster kinetic when response to the sensors, which leads to faster absorption and desorption onto the sensors and subsequent shorter time periods.

[0199] In some embodiments, the time period may comprise 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 500, 1000, 5000, 8000, 10000, 50000, 100000, 200000, 500000, 1 million, 1.5 million, 5 million, 10 million, 15 million, or 20 million time points, or any number of time points in a range bounded by any two values disclosed herein. In some embodiments, the time points needed to determine the identity and quantity is analyte dependent. In some embodiments, more time points may lead to more accurate results. In some embodiments, the time points may be evenly distributed during the time period. In other embodiments, the time points may be unevenly distributed. As a non-limiting example, more time points may be used for detecting the signal of the analyte during the adsorption and / or desorption process of the analyte to the sensor, compared to the equilibrium stage of the analyte over the surface of the sensor.

[0200] In some embodiments, the first processor of the detection device is configured to generate a temporal profile of the readouts of the signal detected by the sensor as a response to the analyte. In some embodiments, the temporal profile of an analyte is the response (Y- axis) of the sensor to the analyte over time (x-axis) during adsorption and / or desorption of the analyte over the surface of the sensor. In some embodiments, the temporal profile reveals temporal behaviors of the analyte during interaction with the sensors, which contributes to compound differentiation. In some embodiments, the temporal kinetics, such as rising time, decay slope, peak width, and recovery profile, varied significantly between analytes. Without wishing to be bound by any particular theory, it has been surprisingly found that the detection device of certain embodiments disclosed herein generates a temporal profile of the analyteAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 and results in a superior and more accurate detection of the identity and concentration of the analyte.

[0201] In some embodiments, the first processor is configured to process the plurality of readouts. In some embodiments, the first processor removes noises or errors from detected signals of the sensors, such as the noises caused by humidity and / or temperature. In some embodiments, the first processor removes noises or errors from detected signals caused by fluctuations of humidity and / or temperature within the time period of detection.

[0202] In some embodiments, the memory of the detection device stores temporal profiles of known analytes in a database. In some embodiments, the memory is configured to be trained by a machine learning model to add additional temporal profiles of new known analytes to the database. In some embodiments, the machine learning model is a random forests model, a linear regression model, a support vector machine model, a support vector classifier model, a Random Forest model, or a deep neural network model.

[0203] In some embodiments, the second processor analyzes the signal by executing a machine learning model disclosed herein. In some embodiments, the second processor analyzes the signal by executing a machine learning model disclosed herein to obtain characteristics of the liquid sample, such as identity and concentration. In some embodiments, the second processor of the detection device is configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample.

[0204] In some embodiments, the first processor and the second processor of the detection device are the same processor. In these embodiments, the same processor generates the temporal profile of the readouts and subsequently compares the generated temporal profile with the database in the memory to determine the identity and / or concentration of the analyte in the fluid sample. In other embodiments, the first processor and the second processor of the detection device are different processors. In some embodiments, the first processor generates the temporal profile of the readouts. In these embodiments, the temporal profile generated by the first processor is then processed by the second processor. In these embodiments, the second processor compares the temporal profile generated by the first processor with the known temporal profiles stored in the database of the memory. In some embodiments, the second processor determines the characteristics (e.g., identity and / or concentration) of the analyte if a matching temporal profile is found in the database.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0205] In some embodiments, the fluid sample comprises at least two analytes. In some embodiments, the at least one sensor is capable of detecting more than one analyte. In some embodiments, one sensor may be configured to detect two or more analytes. In some embodiments, the first processor converts the readouts of the signals into temporal profiles. In some embodiments, the second processor compares the temporal profiles with the database. In other embodiments, the first processor may generate one temporal profile from the readouts containing information of more than one analyte. In these embodiments, the one temporal profile is then deconvoluted by the second processor into separate temporal profiles, each corresponding to a single analyte, and compared with the database to determine the identity and concentration of the analytes.

[0206] In some embodiments, the deconvolution is a process to separate, clarify and resolve complex and tangled data set into its simpler and constituent form. In some embodiments, the deconvolution of analytes is to separate the collective signals detected by multiple sensors into separate signals, each separated signal reflects a single signal corresponding to a single analyte. In some embodiments, the deconvolution is done by dimensionality reduction through principal component analysis (PCA). In some embodiments, the deconvolution is done by one of the machine learning models or statistical models disclosed herein.

[0207] In some embodiments, the detection device further comprises an additional sensor (a second sensor) in addition to the at least one sensor (the first sensor). In some embodiments, the additional sensor is configured to detect an additional signal of the fluid sample at the plurality of time points within the time period to generate an additional plurality of readouts of the additional signal each measured at one of the plurality of time points within the time period; the first processor is configured to generate an additional temporal profile of the additional plurality of readouts; and the second processor is configured to compare the additional temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample.

[0208] In some embodiments, the first and the second sensors both detect a signal from the same fluid sample. In some embodiments, the first sensor may be superior in detecting some properties of the fluid sample, while the second sensor may be superior in detecting a complimentary set of properties compared to the first sensor. In some embodiments, the firstAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 sensor may have a lower limit of detection (LOD) than the limit of detection of the second sensor for the analyte. In some embodiments, the second sensor may have a smaller noise-to- signal ratio compared to that of the first sensor.

[0209] In some embodiments, the first and second processors deconvolute the signals to take advantage of the benefits of the first and second sensors in detecting the analytes, such as identity and / or concentration of the analytes. In some embodiments, the first processor and / or the second processor may process the signal from the first sensor to detect the presence of the analyte (e.g., when the first sensor has a lower LOD) and process the signal from the second sensor to reduce the noise-to-signal ratio, thereby providing more accurate detection compared to using signals from a single sensor.

[0210] In some embodiments, the detection device may comprise multiple sensors to provide signals for multiple analytes. In some embodiments, each sensor provides unique signals of the analytes based on the interaction between the analytes and the sensors. In some embodiments, the first and / or second processors analyze multiple signals from the sensors and deconvolute based on the properties (LOD, linearity, noise-to signal ratio, etc.) detected by the multiple sensors to achieve improved accuracy.

[0211] In some embodiments, the second processor of the detection device is configured to determine the characteristics (e.g., identity and / or concentration) of the analyte. In some embodiments, the detecting capabilities of the detection device are not limited to the identity and concentration of the analyte. In some embodiments, the detection device may detect an identity of a mixture of analytes comprising more than one analyte. In some embodiments, the detection device may detect density, temperature, molecular structure, other suitable characteristics of the analytes, and combinations thereof.

[0212] In some embodiments, the second processor of the detection device determines the identity and / or concentration of the analyte by a linear regression model or a support vector regression model. In some embodiments, the support vector regression model generates superior results in terms of sensitivity and / or reliability.

[0213] In some embodiments, the temporal profile is projected into principal component space to form closed-loop trajectories. In some embodiments, the closed-loop trajectories are referred to as leaflets. In some embodiments, the leaflets encode analyte identity and concentration in a more interpretable way. In some embodiments, the at least one sensor is configured to allow the adsorption and desorption of the analyte over a surface of the at leastAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 one sensor within the time period, and the leaflet is configured to reflect the adsorption and desorption process.

[0214] In some embodiments, the leaflet’s orientation is configured to reflect a pathway of the adsorption and desorption of the analyte over the surface of the at least one sensor. In some embodiments, a curvature of the leaflet is configured to reflect a hysteresis and recovery kinetics of the adsorption and desorption of the analyte over the surface of the at least one sensor. In some embodiments, the enclosed area scales (e.g., proportional with) with analyte quantity. Thus, in some embodiments, the leaflet geometry provides a direct mapping between sensor dynamics and molecular processes, bridging the gap between raw kinetics and machine learning inference.

[0215] In some embodiments, the detection device further comprises an interface for inputting user instructions to control the at least one sensor and / or the first and / or second processors. In some embodiments, the user instructions may include adding or omitting a duplicate sensor, bypassing a signal from a failed sensor, selecting signals from a subset of sensors, selecting regression methods, other reasonable user instructions, and / or combinations thereof.

[0216] In some embodiments, the detection device further comprises a communication module configured to communicate with a user of the detection device regarding the generated temporal profile, the determined identity and / or concentration of the analyte, other characteristics of the analytes generated by the detection device, and / or combinations thereof. In some embodiments, the communication module may communicate the generated temporal profile to the user for manual inspection of the detection process, or to notify the user of a generated temporal profile that does not have a match in the database. In some embodiments, the communication module provides identity and concentration of the analyte to the user for the user to respond accordingly, such as to trigger emergency protocol in the presence of high-level toxic analyte.

[0217] In some embodiments, the detection device further comprises a gas-flow device configured to deliver the fluid sample to the at least one sensor. In some embodiments, the gas-flow device actively moves the fluid sample to the surface of the sensors for detection. In some embodiments, the gas-flow device may be a pump, a fan, a pressurized gas cylinder, a vacuum, other suitable device for moving the fluid sample over the surface of the sensor, and a combination thereof.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0218] In some embodiments, the detection device further comprises a gas-removal device to remove the fluid sample from the at least one sensor. In some embodiments, the gas-removal device may be the same as the gas-flow device, where the same set of device functions both to move the fluid sample towards and away from the surface of the sensors. In other embodiments, the gas-removal device may be different from the gas-removal device. As a non-limiting example, the gas-flow device may be a fan positioned upstream from the sensor(s), while the gas-removal device may be a pump downstream from the sensor(s).

[0219] In some embodiments, it is disclosed herein a detection device collecting temporal profiles of an array of sensors to detect toxic compounds. In some embodiments, the temporal profiles are mapped into principal component spaces as closed-loop manifolds. In some embodiments, by using such detection device, chemically similar toxic compounds are distinguished with low limit of detection, high sensitivity, good reliability, and high signal-to- noise ratios. In some embodiments, the detection device remains robust under variable humidity by using machine learning algorithms.

[0220] In another aspect, a detection device is described, comprising an array of sensors configured to detect an analyte in a fluid sample, comprising: a first sensor selected based on a first property in its detection of the analyte; and a second sensor selected based on a second property different from the first property in its detection of the analyte; wherein the first and second properties are each selected from the group consisting of limit of detection, sensitivity, reliability, reproducibility, and linearity in detection of the analyte.

[0221] In some embodiments, the first sensor’s first property is superior to the second sensor’s first property in detecting the analyte. In some embodiments, the second sensor’s second property is superior to the first sensor’s second property in detecting the analyte. In some embodiments, the first and second properties are selected from limit of detection (LOD), sensitivity, reliability, linearity, other suitable sensor properties in detection of the analyte, other suitable properties, and a combination thereof.

[0222] In some embodiments, the detection device comprises multiple sensors selected based on the unique properties of the sensor in its detection of the analyte. While all of the sensors provide readouts of the analytes, in some embodiments, not all of the collected readout data are the same quality or precision. In some embodiments, the multiple sensors may provide complimentary properties to each other. As a non-limiting example, the firstAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 sensor may provide lower LOD in detecting the analyte, a second sensor may provide high reliability in detecting the analyte, and / or an additional sensor may provide better linearity in detecting the analyte. In some embodiments, the detection device may then deconvolute the signals from multiple sensors and process desirable data from each sensor using regression tools to determine the analyte characteristics (e.g., identity and concentration).

[0223] In some embodiments, the deconvolution is a process to separate, clarify and resolve complex and tangled data set into its simpler and constituent form. In some embodiments, the deconvolution of analytes is to separate the collective signals detected by multiple sensors into separate signals, each separated signal reflects a single signal corresponding to a single analyte. In some embodiments, the deconvolution is done by dimensionality reduction through principal component analysis (PCA). In some embodiments, the deconvolution is done by one of the machine learning models disclosed herein.

[0224] In some embodiments, the limit of detection for the analyte is about 0.00001, 0.0001, 0.0005, 0.001, 0.01, 0.05, 0.1, 0.15, 0.5, 5, 10, 20, 50, 100, 500, 1000, 5000 ppm, or in the range bounded by any two values disclosed herein. In some embodiments, the limit of detection is from about 0.1 ppb to about 150 ppm. In some embodiments, the limit of detection is from about 0.01 ppb to about 50 ppb. In a further embodiment, the limit of detection is from about 0.01 ppb to about 0.5 ppb. In some embodiments, the LOD of detecting benzene using a SGP41 sensor is about 0.1 ppb. In some embodiments, the LOD of detecting benzene using an ENS 162 R2 sensor ranges from about 20.5 ppm to about 37.4 ppm.

[0225] In some embodiments, the sensitivity in detecting the analyte is measured by a slope of the calibration curve. In some embodiments, the calibration curve is the log-linear scale of signal intensities as Y-axis plotted against concentration of the analytes in the fluid sample. In some embodiments, the sensitivity may be about 0.01, 0.05, 0.1, 0.2, 0.05, 0.5, 0.8, 0.9, 0.99, or in the range bounded by any two values disclosed herein. In some embodiments, the sensitivity is from about 0.05 to about 0.99.

[0226] In some embodiments, the reliability in detecting the analyte is measured by signal-to-noise ratio. In some embodiments, the signal-to-noise ratio is 0.00001, 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 0.8, 0.9, or in the range bounded by any two values disclosed herein. In some embodiments, the signal-to-noise ratio is from about 0.0001 to about 0.1. In some embodiments, the signal-to-noise ratio is from about 0.01 to about 0.5. InAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 some embodiments, the signal-to-noise ratio is from about 0.1 to about 0.2. In some embodiments, the signal-to-noise ratio is from about 0.15 to about 0.2. In some embodiments, the signal -to noise ratio is 0.18

[0227] In some embodiments, the linearity in detecting the analyte is measured by a coefficient of determination. In some embodiments, the linearity is about 0.01, 0.1, 0.2, 0.5, 0.7, 0.8, 0.9, 0.99, 0.999, or in the range bounded by any two values disclosed herein. In some specific embodiments, the linearity is from about 0.2 to about 0.999. In some specific embodiments, the linearity is from about 0.1 to about 0.5. In some specific embodiments, the linearity is from about 0.2 to about 0.3. In another specific embodiment, the linearity is 0.26.

[0228] In some embodiments, the analyte is one or more volatile organic compounds. In some embodiments, the analyte is selected from the group consisting of benzene, formaldehyde, toluene, xylenes, BTEX, methane, propane, a sulfur gas, ammonia, nitrogen dioxide, hydrogen, an alcohol, smoke, carbon monoxide, carbon dioxide, or a combination thereof. In some embodiments, the analyte is benzene, toluene, formaldehyde, BTEX, or a combination thereof. In some embodiments, the fluid sample is a group of analytes including benzene, toluene, formaldehyde, and BTEX in air. In some embodiments, the fluid sample is a group of analytes including benzene, toluene, and formaldehyde in air. In some embodiments, the air is dry air with 0% relative humidity. In some embodiments, the relative humidity of the air is about 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, or 90%, or in the range bounded by any two values disclosed herein.

[0229] In some embodiments, the fluid sample is an air sample from a building, a ship, a hotel, a hospital, an industrial site, a breath sample, or the atmosphere. In some embodiments, the detection device is configured to detect analyte in a fluid sample to analyze indoor air qualities, such as air qualities in buildings, ships, hotels, and hospitals. In other embodiments, the detection device is configured to detect analyte in a fluid sample to analyze industrial safety and leak detection (e.g., in an industrial site). In these embodiments, the analytes comprise methane, sulfur gases, or ammonia. In some embodiments, the detection device is configured to detect the analyte in a fluid sample to conduct environmental monitoring, such as greenhouse gas mapping. In some embodiments, the detection device is configured to detect the analyte in a fluid sample to conduct medical diagnostics, such as detecting the analyte from a person’s breath to conduct breath analysis.

[0230] In some embodiments, the at least one sensor is selected from the group of a chemiresi stive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, andAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 a hybrid multi-modal sensor. In one example, the first and second sensors are chemiresi stive sensors, such as metal-oxide sensors.

[0231] In some embodiments, the at least one sensor is selected from the group consisting of a chemiresi stive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, and a hybrid multi-modal sensor. In some embodiments, the sensor is a chemiresistive sensor, such as a metal oxide sensor, a conductive polymer sensor, or a carbon nanotub e / graphene sensor. In some embodiments, the sensor is an optical sensor. Non-limiting examples of an optical sensor include a fluorescence sensor, a reflection sensor, a surface plasmon resonance (“SPR”) sensor, or a Raman sensor. In some embodiments, the sensor is a gravimetric sensor. Non-limiting examples of an optical sensor include a quartz crystal microbalance (“QCM”) sensor. In some embodiments, the sensor is an electrochemical sensor or a hybrid transducer (e.g., a hybrid multi-modal sensor).

[0232] In some embodiments the detective device consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, a MQ135 sensor, and a MiCS 5914 sensor.

[0233] In some embodiments the detective device consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138-B sensor, a SEN0566-B sensor, a MiCS5524-B sensor, a MQ135-B sensor, and a MiCS 5914 sensor.

[0234] In some embodiments the detective device consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, and a MiCS 5914 sensor.

[0235] In some embodiments the detective device consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138-B sensor, a SEN0566-B sensor, a MiCS5524-B sensor, and a MiCS 5914 sensor.

[0236] In some embodiments the detective device consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, and a MQ135 sensor.

[0237] In some embodiments the detective device consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138-B sensor, a SEN0566-B sensor, a MiCS5524-B sensor, and a MQ135-B sensor.

[0238] In some embodiments, the first sensor is selected based on one or more properties in detecting analytes in the fluid samples. In some embodiments, the first sensor is a SGP41 sensor selected for its reproducibility and low limitation of detection of the analyte. In someAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 embodiments, the first sensor is an ENS160 sensor selected for its heightened sensitivity in detecting the fluid sample. In some embodiments, the first sensor is an ENS160 sensor selected for its consistent results for detecting both benzene and toluene, e.g., concentrations and identities. In some embodiments, the first sensor is a MQ138 sensor selected for its high linearity in detecting the fluid sample to provide complimentary properties different from the properties of the second sensor. In some embodiments, the first sensor is a MiCS5524 sensor selected for its low limit of detection and high linearity in detecting the fluid sample. In some embodiments, the first sensor is a SEN0566 sensor selected for its high linearity in detecting the fluid sample.

[0239] In some embodiments, the detection device comprises a SGP41 sensor and a ENS160 sensor. In some embodiments, the detection device further comprises a SCD 40 sensor. In some embodiments, the detection device further comprises a MQ138-B sensor, a SEM0566-B sensor, and / or a MiCS5524-B sensor. In some embodiments, the detection device further comprises a MiCS5524-B sensor and a MiCS5914 sensor. In some embodiments, the array of sensors comprises a SGP41 sensor, a MiCS5524 sensor, a ENS160 R2 sensor, a ENS 160 R3 sensor, and a MQ138 sensor.

[0240] In some embodiments, at least one of the first sensor and the second sensor is configured to detect a signal of the analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; and wherein the detection device further comprises a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated by a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; and a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors are the same or different processors.

[0241] In some embodiments, the detection device integrates multiplexed sensing (e.g., chemiresistive) with time-series analysis to address key challenges in analyte (e.g., volatile compounds or volatile organic compound) detection. In some embodiments, the detection device demonstrates many advances, including reliability on (1) identifying analytes in a fluid sample comprising multiple analytes, (2) quantifying the influence of ambient humidity on identifying and quantifying analytes in the fluid sample, (3) complete classification ofAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 analytes using temporal profiles of the analytes, and (4) quantitative prediction of structurally similar analytes.

[0242] In some embodiments, reliable detection of benzene, formaldehyde, toluene, xylenes, and BTEX mixture is achieved at concentrations as low as 25 ppb with high signal- to-noise ratios, surpassing established regulatory thresholds.

[0243] In some embodiments, the influence of ambient humidity on sensor behavior was systematically quantified. In some embodiments, the results show that machine learning inference reduces humidity-induced error by more than 50%, while linear regression provides partial or complete correction.

[0244] In some embodiments, time-series principal component trajectories form distinctive leaflet geometries were revealed, which enables complete classification of benzene, toluene, and formaldehyde, with even simple random forest models reaching 100% accuracy across independent trials.

[0245] In some embodiments, quantitative prediction of formaldehyde, toluene, and benzene was demonstrated, with benzene achieving remarkable sub-10-ppb resolution (R2= 0.9997, root-mean-square errors, RMSE = 6.1 ppb) at 100 ppb, establishing time-resolved analysis as a powerful tool for precise, real-world monitoring.A method of detecting and analyzing a fluid sample

[0246] In another aspect, a method of detecting and analyzing a fluid sample is described, the method comprises the following steps: providing a detection device according to any of the embodiments described herein, detecting the signal of the fluid sample at the plurality of time points within the time period to generate the plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; generating the temporal profile of the readouts by the first processor; and comparing the generated temporal profile of the fluid sample with the database by the second processor to determine the identity and / or concentration of the analyte in the fluid sample.

[0247] In some embodiments, the method includes providing a detection device according to any of the embodiments described herein, such as one of the detection devices disclosed in the sections of “a gas detection device” and “examples”.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0248] In some embodiments, the sensor detects a signal of the fluid sample at plurality of time points within a time period. In some embodiments, the sensor generates a readout at each of the time points within the time period. In some embodiments, the first processor plots the readouts from the sensor to generate a temporal profile of the analyte. In some embodiments, the temporal profile is a plot of signal showing the sensor response (Y-axis) to the analyte at each time points (X-axis) within the time period.

[0249] In some embodiments, the second processor compares the temporal profile of the analyte with the temporal profiles in the database. In some embodiments, the second processor determines the characteristics of the analyte (e.g., identity and concentration) based on the comparison if there is at least one match. In some embodiments, if the second processor fails to find a match in the database, the unmatched temporal profile is noted for the user for further decisions.

[0250] In some embodiments, the fluid sample comprises at least two analytes; and the method further comprises: deconvolute the generated temporal profile of the fluid sample to generate a temporal profile for each of the at least two analytes; and comparing the temporal profile of each of the at least two analytes with the database to determine the identity and / or concentration of each of the at least two analytes.

[0251] In some embodiments, one sensor can detect at least two analytes. The sensor detects signals of the at least two analytes to generate readouts for each of the analytes. In some embodiments, the first processor may then deconvolute the readouts to generate a temporal profile for each analyte. In some embodiments, the second processor subsequently compares each of the temporal profile with the database to determine the characteristics (e.g., identity and / or concentration) of the analytes. In some embodiments, the sensor may generate a plurality of readouts reflecting both analytes from the signal detected from the analytes. In some embodiments, the first processor may then generate a temporal profile from the plurality of readouts containing the responses of at least two analytes. In some embodiments, the second processor deconvolutes the temporal profile to generate one temporal profile for each analyte. In some embodiments, the temporal profile for each analyte is subsequently compared with the temporal profiles in the database to determine the characteristics (e.g., identity and / or concentration) of the analytes.

[0252] In another aspect, a method of detecting and analyzing an analyte in a fluid sample is described, the method comprising the following steps:Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 providing the detection device according to any of the embodiments described herein; detecting the analyte by the first sensor; and detecting the analyte by the second sensor.

[0253] In some embodiments, the method including providing a detection device described herein, such as one of the detection devices disclosed in the sections of “a gas detection device” and “examples”.

[0254] In some embodiments, at least two sensors are detecting the same analyte, where the two sensors have complimentary properties in detecting analytes. In some embodiments, the first sensor provides a low LOD compared to the LOD of the second sensor in detecting the at least one analyte. In some embodiments, the second sensor provides better sensitivity compared to the sensitivity of the first sensor. In some embodiments, the detection device then deconvolute the signals from the first and the second sensors to determine the characteristics of the analyte. In some embodiments, the results deconvoluted from multiple sensor signals have better accuracy compared to the results from a single sensor.

[0255] In some embodiments, the method uses a machine learning algorithms model to analyze the detected signals. In some embodiments, the machine learning model uses support vector classifiers (SVCs) and / or random forests (RFs) to identify and quantify the analytes. In some embodiments, RFs yield higher accuracy compared to the SVCs among all tested analytes.Examples

[0256] The various embodiments are further described with reference to the following non-limiting examples.Hardware Overview

[0257] In some embodiments, the detection device was designed around an Arduino Mega 2560 microcontroller board integrated with a series of modular, custom-fabricated printed circuit boards (PCBs). Each PCB incorporated a surface-mount sensor array comprising an array of sensors (Fig. 1 A), such as chemiresistive and environmental gas sensors. The sensor configurations were optimized to improve analyte sensitivity, minimize noise, reduce redundancy, and streamline power and routing constraints. The PCB integratedAttorney Docket Number: 0042697.00611WO1 Date of Electronic Filing: September 24, 2025 sensors were selected based on performance metrics-specifically response magnitude, noise floor (limitation of detection), and concentration dynamic range. Sensor data acquisition was executed via analog and digital input channels, with voltage outputs corresponding to analyte-induced changes in sensor resistance. The sensors used in the sensor array and the gases to which they are each expected to respond strongly are listed in Table 1. The sensors span a range of chemiresistive and NDIR mechanisms to collectively detect key indoor air pollutants, including VCs, VOCs, formaldehyde, and combustion byproducts. In some embodiments, the detective device has sensors from sensor No. 1-6 and sensor No. 7-1. In some embodiments, the detective device has sensors from sensor No. 1-6 and sensor No. 7-2.Table 1: Gas sensors integrated into the sensor array and their primary target analytes.

[0258] In some embodiments, the detective device comprises a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138-B sensor, a SEN0566-B sensor, a MiCS5524-B sensor, a MQ135-B sensor, and a MiCS 5914 sensor.

[0259] In some embodiments, the detective device comprises a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138-B sensor, a SEN0566-B, a MQ 138 sensor, a MiCS5524- B sensor, and a MiCS 5914 sensor.

[0260] In some embodiments, the detective device comprises a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138-B sensor, a SEN0566-B sensor, a MiCS5524-B sensor, and a MQ135-B sensor.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025Design of Experimental Platform

[0261] In some embodiments, a custom enclosure was developed to enable controlled delivery and purging of fluid samples (e.g., VCs or VOCs) for sensor characterization. As shown in Fig. IB, the system allows for precise modulation of analyte concentration and environmental humidity via mass flow controllers and a modular mixing setup. The integration of a third stream for humidification through dual bubblers facilitates systematic evaluation of humidity effects (Fig. 1C).VOC Introduction (dehumidified stream)

[0262] In some embodiments, the VOCs were introduced from two streams joined by a y- shaped (Fig. IB) connector leading to a sealed enclosure. These two streams each contained a mass flow controller, with one stream controlling the flow of VOC vapor, and the other controlling the flow of inert carrier gas (e.g., N2, air). In some embodiments, a Bronkhorst MASS-STREAM D64 flow controller (Hoskin Scientific, BC, Canada) was used to control the diluting stream. In some embodiments, the VOC-introducing stream utilized a Bronkhorst EL-FLOW Select flow controller (Hoskin Scientific, BC, Canada). The concentrations of inert carrier gas and of VOC were controlled by programming the respective mass flow controllers for the required flow rates and durations. When compressed gas cylinders were used for VOC introduction, these were upstream of the EL-FLOW.VOC Introduction (humidified stream)

[0263] In some embodiments, sensor characteristics in response to changes in ambient humidity were achieved by supplementing the pre-dehumidified gas introduction streams discussed above with the controlled introduction of water vapor (Fig. 1C). A pressure regulator was connected to the outlet of an N2 gas source and set to 50 psi. The flow rate was measured using a flow meter. Two gas bubblers were subsequently attached in series, with the first being suspended in a water bath at a set temperature. The water vapor flowing out of the second gas bubbler intersected with the stream carrying air as a carrying gas using a y- shaped connector. This stream further intersects with the stream of VOC vapors to be introduced into the sampling chamber. When using the bubbler system for VOC introduction, the bubbler was placed downstream of the EL-FLOW mass flow controller.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025Response Curve Generation

[0264] In some embodiments, sensor responses to target analytes were characterized using two experimental configurations: (1) a bubbler-based setup for generating humidified fluid samples (e.g., VOC) streams (Fig. 1C), and (2) a gas cylinder dilution system for concentration-controlled exposures (Fig. IB). In both setups, airflow was regulated using dual-channel mass flow controllers (MASS-STREAM and EL-FLOW, Bronkhorst). For humidified trials, water vapor was introduced via a dual-bubbler series circuit on a nitrogen gas line (Fig. 1C), while fluid samples (e.g., fluid samples containing VOCs such as benzene) were supplied from pressurized cylinders. Desired concentrations were achieved by adjusting flow ratios of dry air, humidified air, and analyte gas. Sensor firmware was developed in C++ using the Arduino IDE, with data acquisition handled via serial communication to a host PC running Python (PySerial). MFC operation, experimental timing, and logging were coordinated through Python scripts interfaced with FlowView software (Bronkhorst). Each trial followed a standardized exposure sequence to the gases: (1) 150 seconds of exposure to baseline (clean air); (2) 150 seconds of exposure to the fluid sample (comprising VOC), and (3) 600 seconds of exposure to clean air to remove the fluid sample attached to the sensors as sensor recovery (purging). Resistance values were recorded continuously at 0.2 Hz, with a minimum of three biological replicates per analyte and concentration.Limit of Detection Studies

[0265] In some embodiments, limit of detection (LOD) characterization was performed using certified gas cylinders (Airgas) containing benzene (20 and 100 ppm), toluene (100 ppm), or formaldehyde (9 ppm). Analyte streams were diluted in real-time using calibrated dual-channel mass flow controllers (MASS-STREAM and EL-FLOW, Bronkhorst) to achieve final concentrations ranging from 25 ppb to 10 ppm. These concentrations encompass typical occupational exposure limits (e.g., OSHA permissible exposure limits) as well as indoor air quality standards (e.g., LEED guidelines). Total flow was fixed at 10 LPM, with target concentrations verified by dynamic flow ratio calculations and validated experimentally via repeatability of sensor response. For each sensor, LOD was defined as the analyte concentration at which the response signal exceeded three times the baseline noise level. LOD of various sensor responding to four tested analytes (formaldehyde, benzene, toluene, and BTEX) are included in Table 2. Table 2 also includes additional performance metrics extracted from temporal profiles. The additional performance metrics includes theAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 linear regression coefficient (R2), signal-to-noise ratio (SNR), and adsorption / desorption time constants (rr, if).

[0266] Table 2 describes the sensor-specific performance metrics across four target analytes: formaldehyde, benzene, toluene, and BTEX. Each cell reports a tuple of six key values (listed from left to right in parentheses in Table 2): response time (rr), recovery time (if), signal-to-noise ratio (SNR), limit of detection (LOD), slope of the calibration curve, and coefficient of determination (R2). Metrics were extracted from temporal profiles under standardized exposure protocols.Table 2. Sensor-specific performance metrics across four target analytes: formaldehyde, benzene, toluene, and BTEX.Humidi fication Study

[0267] In some embodiments, the influence of relative humidity (RH) on sensor performance was evaluated using a three-stream gas mixing configuration comprising dry air, benzene-doped air (0.95 ppm), and humidified nitrogen passed through a dual-bubbler system. Humidity levels were adjusted in 10% increments from 0% to 90% RH by tuning the flow rate of the humidified stream via a precision needle valve, while maintaining a total flow rate of 5 LPM. RH was monitored in real time using a calibrated DHT11 sensor placed within the sensing chamber. At each RH level, sensors were exposed to benzene under steady-stateAttorney Docket Number: 0042697.00611WO1 Date of Electronic Filing: September 24, 2025 flow conditions, and the response was recorded. Deviation from the 10% RH baseline response was calculated to quantify the impact of humidity on sensing accuracy.Data Fusion and Processins

[0268] In some embodiments, sensor response data were collected using a Python-based acquisition script that queried the detection device via serial communication with the Arduino MEGA 2560. Resistance values from each sensor channel were recorded at 5-second intervals and logged as raw datasets in .csv format. For downstream analysis, all raw data were first normalized according to the expression:where Rb is the baseline resistance in clean air and Rgas(t) is the instantaneous resistance during fluid sample (e.g., VOC containing fluid sample) exposure. This normalized response, S(t), was used consistently across all machine learning workflows. For static analysis, a single scalar feature - Smax (the maximum response amplitude) - was extracted from each sensor per trial, yielding a 1-D feature vector across the sensor array. For time-series models, the full normalized response traces were retained for each sensor, forming a 2-D matrix (tsensorsxntimepoints) per sample. The detection device probed one measurement from all sensors every 5 seconds. These time-resolved matrices (readouts) served as input to temporal classification and regression algorithms to generate temporal profiles to identify (or classify) and quantify the analytes in the fluid samples.Machine Learning Algorithms

[0269] In some embodiments, for classification and quantification tasks, datasets were split into training and test sets (i.e., 80% training unless specified otherwise). Classification models included support vector classifiers (SVCs) and random forests (RFs), with RFs yielding highest accuracy across all gas classes. Regression models for analyte concentration estimation employ support vector regressors (SVRs) and RF regressors. All models were implemented using Python (scikit-leam vl.3) and evaluated with five-fold cross-validation. Performance metrics included accuracy, RMSE, and R2. Feature importance and ablation analyses were conducted to identify dominant sensor contributions for both classification and regression tasks.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025Sensor Array Calibration

[0270] In some embodiments, to characterize the analytical performance of the detection device, a hybrid chemiresi stive sensor array was exposed to four representative indoor air pollutants: formaldehyde, benzene, toluene, and a BTEX mixture. The BTEX mixture is comprised of 1 ppm benzene, 10 ppm toluene, 10 ppm ethylbenzene, 20 ppm m-xylene, and 20 ppm o-xylene. Each compound was delivered via controlled flow protocols in the range of 25 ppb to 5 ppm.

[0271] Representative temporal profiles for the SGP41, ENS160 R0, ENS160 R2, ENS160 R3, MiCS5524, MQ138 sensors are shown in Figs. 2A-1 to 2D-4. Fig. 2A-1 illustrates temporal profiles of a SGP41 sensor exposed to formaldehyde in a fluid sample at various concentrations. Fig. 2A-2 illustrates temporal profiles of an ENS160 R2 sensor exposed to formaldehyde in a fluid sample at various concentrations. Fig. 2A-3 illustrates temporal profiles of an MiCS5524 sensor exposed to formaldehyde in a fluid sample at various concentrations. Fig. 2A-4 illustrates temporal profiles of an ENS160 R0 sensor exposed to formaldehyde in a fluid sample at various concentrations. Fig. 2B-1 illustrates temporal profiles of a SGP41 sensor exposed to benzene in a fluid sample at various concentrations. Fig. 2B-2 illustrates temporal profiles of an ENS160 R2 sensor exposed to benzene in a fluid sample at various concentrations. Fig. 2B-3 illustrates temporal profiles of a MQ138 sensor exposed to benzene in a fluid sample at various concentrations. Fig. 2B-4 illustrates temporal profiles of a MiCS5524 sensor exposed to benzene in a fluid sample at various concentrations. Fig. 2B-5 illustrates temporal profiles of an ENS160 R0 sensor exposed to benzene in a fluid sample at various concentrations. The lines (e.g., dashed lines, dotted lines or solid lines) in each of the Figs. 2A-1 to 2D-4 represented mean of the multiple trails. The shaded regions in each of the Figs. 2A-1 to 2D-4 indicate standard deviation across multiple trials. In some embodiments, as shown in Figs. 2A-1 to 2D-4, the sensor response (y-axis) over time (x-axis) increases with the increasing of the analyte concentrations. In some embodiments, the sensor response is proportional to the analyte concentration.

[0272] The SGP41 exhibited comparably rapid adsorption-desorption kinetics with high reproducibility across replicates, particularly evident in its benzene and toluene responses (Figs. 2B-1, 2C-1), where clear signal plateaus - representing equilibrium - and recovery phases emerged even at sub-ppm concentrations. Conversely, the ENS160 - particularly the R3 feature output-demonstrated higher amplitude responses (y-axis), but with reduced signal- to-noise ratio (SNR). This divergence in behavior is attributed to inherent materialAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 differences and readout stability, emphasizing that signal amplitude alone is insufficient when evaluating sensor utility for real-world trace gas detection.

[0273] In some embodiments, the temporal profiles showed variations in the time period prior to 100 s, such as shown in Figs. 2A-1, 2B-1, 2C-1, 2D-1. In a non-limiting embodiment, the variation in signals for times less than 100 s was due to the warm-up time of the SGP41 sensor. A baseline was observed as the signal after the warmup period but before the introduction of the analyte into the system (i.e. from -100 - -170 s in Fig. 2A-1), as well as after the analyte had been successfully purged from the testing enclosure. The baseline changed only minimally for concentrations differing by more than an order of magnitude. This presented a small upper limit on rates of gradual sensor drift, even when exposed to high analyte concentrations. This degree of drift could feasibly be accounted for through a combination of signal processing and regular recalibration. Drift in this way should be defined as the change over time in signal response to similar concentrations of the same mixtures of gases.

[0274] To facilitate quantitative analysis, signal intensities were plotted against concentrations on a log-linear scale (Figs. 3 A-D). Data extracted (e.g., slope of the calibration curve, and coefficient of determination R2) from Figures 3 A-D is also listed in Table 2.In some embodiments, for benzene detection (Table 2), the SGP41 exhibited the most favorable balance between signal stability and analytical performance, achieving the highest signal -to-noise ratio (SNR) of 116.1 among all sensors evaluated. Its sensitivity was moderate (slope = 0.4636; AR / Ro per log-fold concentration), but the combination of low baseline noise and high linearity (R2= 0.9905) enabled a 3o-computed limit of detection (LOD) of 0.1 ppb, with clear experimental response observed down to 56 ppb. In contrast, the ENS160 R3 channel (Table 2) demonstrated the highest sensitivity (slope = 0.7635) and a strong linear fit (R2= 0.9918), but its elevated signal variability resulted in a lower SNR of 14.98 and a higher LOD of 33.9 ppb. Similar trends were observed for the ENS160 R2 and R0 outputs (Table 2), with slopes of 0.5281 and 0.4277, respectively, and LODs ranging from 20.5 to 37.4 ppb. While these channels provided strong signal magnitudes, their utility was limited by increased noise. The MiCS5524 offered solid performance (Table 2) with an LOD of 11 ppb and high linearity (R2= 0.9944), whereas the MQ138 and SEN0566 exhibited lower sensitivities (slopes < 0.4) and elevated LODs (> 100 ppb), indicating more constrained applicability in trace-level benzene detection.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0275] In some embodiments, for formaldehyde (Table 2), both the SGP41 and ENS 160 sensors demonstrated strong analytical performance, though through differing trade-offs between noise and sensitivity. The SGP41 offered a high SNR of 93.47, rapid response time (rr = 5 s), and excellent linearity (R2= 0.9986), enabling a So-computed LOD of 0.5 ppb and experimentally observed response down to 25 ppb. Its sensitivity was 0.6264, sufficient for sub-ppm quantification with minimal baseline drift. Among the ENS 160 readouts, the R3 channel delivered the highest sensitivity (slope = 0.7507) and nearly perfect linearity (R2= 0.9989), although its SNR (27.5) was significantly lower than that of the SGP41. The corresponding LOD was 9.5 ppb, computed at the 3c level, with an experimental minimum signal distinguishable at about 50 ppb. Other ENS160 outputs (R2 and R0) performed comparably, with sensitivities of 0.6396 and 0.6681, and LODs of 8.7 and 135 ppb, respectively, the latter elevated due to baseline instability in the R0 trace. The MiCS5524 and SEN0566 yielded LODs of 10.9 and 1.8 ppb, respectively, but lower linearity (R2= 0.6325 and 0.4659) and reduced SNRs (5.78 and 2.39) indicate limited robustness for low- concentration measurements. MQ138, while exhibiting a fast recovery time and modest sensitivity (0.1194), had poor regression characteristics (R2= 0.2636), confirming its unsuitability for reliable quantification. Taken together, these results affirm the SGP41 as the most stable and broadly applicable sensor for formaldehyde at low concentrations, while the ENS160 R3 channel offers heightened sensitivity when paired with post-processing tools capable of compensating for its elevated noise.

[0276] In some embodiments, to sense toluene (Table 2), the SGP41 again achieved the highest signal-to-noise ratio (SNR = 141.7), enabling clear and stable detection across the tested concentration range. Despite its modest sensitivity (slope = 0.3516), its combination of rapid response (rr = 35 s), high linearity (R2= 0.9585), and excellent noise suppression allowed it to resolve concentrations below 100 ppb, with an estimated experimental LOD approaching 0.1 ppm. The ENS160 R3 channel, by comparison, offered the highest sensitivity (slope = 0.7467) and strongest linear regression (R2= 0.9985), though with an SNR of 14.7 and a 3o-computed LOD of 28.3 ppb. ENS160 R2 and R0 followed similar trends, with slopes of 0.6358 and 0.6149, and LODs of 21.2 and 158.9 ppb, respectively. The MiCS5524 displayed a strong SNR of 85.8 and an extremely low computed LOD of 1.5 ppb, but with diminished linearity (R2= 0.9461) compared to the ENS 160 channels, suggesting a performance ceiling due to response nonuniformity at higher concentrations. The MQ138, while offering moderate sensitivity (0.4242) and good linearity (R2= 0.9632), had a higherAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025LOD of 122.3 ppb, reflecting greater baseline variation. The SEN0566 and ENS160 RO outputs showed limited applicability, with lower SNRs (5.11 and 2.92, respectively) and elevated LODs. Altogether, the SGP41 provided the most robust and stable performance for low-concentration toluene detection, while the ENS 160 R3 remained the most sensitive sensor — ideal for applications requiring high slope readouts when supported by advanced signal processing. The MiCS5524 may serve as a valuable complementary feature due to its strong raw SNR, further justifying its inclusion in hybrid sensor arrays.

[0277] In some embodiments, in BTEX mixture sensing (Table 2), the ENS 160 R3 feature demonstrated the highest sensitivity (slope = 0.8205) and excellent linearity (R2= 0.9881), offering precise signal scaling across concentrations. However, its SNR (29.2) was surpassed by the SGP41, which delivered the most stable performance with an SNR of 95.1, a low LOD of 0.1 ppb, and strong linear regression (R2= 0.9273). While the SGP41’s slope (0.5096) was lower, its superior noise characteristics enabled robust detection of complex vapor mixtures without advanced filtering. Among the ENS 160 variants, both R2 and R0 channels exhibited similarly strong linearity (R2> 0.995) with sensitivities of 0.6529 and 0.6727, and LODs of 6.6 ppb and 22.2 ppb, respectively. These channels contributed valuable distinct features with reliable calibration behavior, particularly across mid-ppm exposures. The MiCS5524, notable for its high SNR (60.3) and low LOD (1.4 ppb), showed slightly diminished linearity (R2= 0.8616), possibly suggesting nonlinear behavior at higher concentrations. The MQ138, while showing moderate sensitivity and regression quality (slope = 0.4183, R2= 0.8348), was limited by its higher LOD of 30.9 ppb and lower reproducibility. The SEN0566 yielded the least favorable performance (SNR = 4.85, LOD = 40.6 ppb), indicating marginal utility for multicomponent detection tasks. Collectively, these results support the use of both SGP41 and ENS 160 R3 as cornerstone sensors in multianalyte VOC classification. The signal stability of SGP41 and discriminability of BTEX at sub-ppm levels make it an ideal baseline sensor, while ENS160 channels contribute high-resolution feature gradients crucial for mixture deconvolution with machine learning models disclosed herein.

[0278] In some embodiments, effective sensor selection for the detection device requires careful trade-offs among sensitivity (quantified by slope), reliability (SNR), and linearity (R2). No single sensor exhibited optimal performance across all evaluated target VOCs - formaldehyde, benzene, toluene, ethylbenzene, and xylenes — necessitating a multiplexed approach. In some embodiments, a multiplexer is used. In some embodiments, theAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 multiplexer not only enhanced flexibility in sensor array configuration but also mitigated the cumulative effect of pull-up resistors required by certain devices. In some embodiments, the multiplexer allowed a single master device or controller to interface with multiple I2C or 13 C slave / peripheral devices, which enhances flexibility and simplifies routing complexity. This allowed for easier control with one controller, lowered the chance of voltage mismatching, and reduced the board space occupied by the system. In some embodiments, the multiplexer also allowed for the use of multiple devices with the same address (e.g., the SGP40 and SGP41), since it could identify each of them individually. Additionally, in some embodiments, when multiple devices or sensors requiring pull-up resistors were placed in parallel (as they would be without a multiplexer), it could lead to issues with signal integrity and power consumption, due to parasitic capacitance or other factors. In some embodiments, having a multiplexer helped mitigate this issue, since the multiplexer could isolate each of the devices. In some embodiments, the multiplexer distributed the bus capacitance across the system. In some embodiments, the multiplexer reduced the negative cumulative effect of the pull-up resistors.

[0279] The SGP41 consistently delivered high SNRs and low computed and experimental LODs, making it the most reliable baseline detector for trace-level quantification, especially under LEED and OSHA-relevant thresholds. In contrast, the ENS160 - particularly its R3 output - offered the highest sensitivity and regression fidelity (R2> 0.99 across all gases), though its elevated signal variability demands computational denoising strategies (e.g., PCA, random forests) to fully realize its value. Importantly, accurate classification and quantification of chemically similar VOCs in mixtures such as BTEX depend not only on individual sensor performance but also on collective signal diversity. High-dimensional machine learning models for mixture deconvolution require at least as many orthogonal sensors features as there are analytes to resolve. For the five-target system examined here, this necessitates a minimum of five decor-related sensor readouts, each contributing a unique signal response profile across the compound set. Sensors with highly overlapping or correlated responses - regardless of sensitivity - are insufficient for selective identification. An optimal sensor suite would include: SGP41 (for stable low-noise detection across all analytes), ENS160 R3 (for high sensitivity and broad dynamic range), MiCS5524 (for complementary high-SNR outputs, especially for toluene and BTEX), ENS 160 R2 (as a second decorrelated ENS output), and MQ138 (to introduce orthogonal response kinetics despite its higher noise). This combination offers complementary performance characteristicsAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 and non-redundant signal behaviors, forming a foundation for robust machine-learning-based analyte (e.g., VOC) classification and quantification in real-world indoor environments. The key sensing metrics belying this sensor selection are further illustrated in Figs. 3 A-3D, showing suitable sensing linearity and experimental LODs at concentrations relevant for monitoring target VOCs near OSHA-recommended concentration limits.Effect of Humidity on Regression and Inference

[0280] In some embodiments, indoor air quality monitoring solutions that determine the identities and concentrations of analytes in fluid samples must be able to adjust according to reasonable changes in ambient environmental conditions. The effects of changes in relative humidity on the behavior of detection devices were studied using regression tools (e.g., simple and machine learning-enabled) to complement the multiplexed sensor array to account for this interfering effect. The general effect of increased relative humidity was to attenuate the sensor response to analytes (e.g., VOCs) in the fluid samples. The attenuation is evidenced by the weakened peak response of the SGP41 sensor (Fig. 4A) exposed to benzene at concentration of ~1 ppm. However, two of the seven evaluated sensors showed average increases in sensor response at elevated humidities - namely the MQ138 and SEN0556. This is expressed by their median percent differences to the response at 10% RH to benzene, which were +58.2 % and +76.3 %, respectively (Fig. 4H). The ENS160-R3 attribute (Fig. 4D), which previously showed promise due to its high sensitivity and large linear response range, showed close to zero deviation in response in higher humidity conditions (4.7% interquartile range). The ENS160 R2 (Fig. 4C) showed similar humidity response to ENS160 R3. The ENS160 R0 (Fig. 4B), MQ138 (Fig. 4E), and SEN0556 (Fig. 4G) had the largest variances in their difference to the 10% RH signal, having interquartile ranges of 35.9%, 43.7%, and 48.0% respectively. MICS5524 (Fig. 4F) showed reduced sensor response with regard to increasing humidity. These reflect strong influences of ambient humidity on sensor response characteristics, though not necessarily that these are influences that cannot be accounted for in downstream processing. A larger variation in response at higher humidities is not necessarily strictly detrimental to sensor value. Since ambient humidity is measured by the detection device and may consequently be included in the calculations of gas identity and concentration, as long as there is a consistent effect of temperature and humidity on a sensor, this relationship can be accounted for by signal processing or machine learning methods.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0281] In some embodiments, to assess whether the humidity-induced drift observed across the sensor array could be algorithmically corrected, two modeling approaches were applied to map sensor responses across relative humidity (RH) levels: a standard linear regression (Fig. 41) and a machine learning (ML) inference model using random forests (Fig. 4J). Both models used 10-fold randomized cross-validation and were trained to map raw sensor outputs to a reference signal at 10% RH, enabling correction prior to classification or quantification. These models leverage the fact that the detection device concurrently measures ambient humidity and sensor resistance, allowing for software-side compensation without additional hardware complexity.

[0282] In some embodiments, linear regression (Fig. 41) applied to the maximum normalized sensor response (Smax) revealed high humidity -induced RMSE for ENS 160 RO (18.1%), SEN0566 (40.7%), and MQ138 (13.0%), while ENS160 R3 and SGP41 had minimal errors (2.6% and 3.1%, respectively), consistent with their humidity stability observed in Fig. 4B, 4G, 4E, and 4D. Notably, the SGP41 exhibited a largely linear attenuation trend with humidity, enabling relatively accurate correction via linear fits. Conversely, sensors such as SEN0566 and MQ138 exhibited more non-linear or unstable humidity dependencies that were not effectively captured by simple regression. When employing a random forest model to capture potential non-linearities (Fig. 4J), performance improved across nearly all sensors. RMSE dropped significantly for SEN0566 (from 40.7% to 24.5%) and ENS160 R0 (from 18.1% to 14.3%). The ENS160 R3 feature again emerged as the most resilient (RMSE < 2%), while SGP41 maintained low error (3.8%), reinforcing its linear and predictable behavior. These results highlight the importance of both sensormaterial stability and downstream analytical flexibility. In particular, pairing sensor-specific knowledge (e.g., the linearity of SGP41 or the sensitivity of ENS160 R3) with appropriate correction algorithms ensures more robust gas quantification. While some sensors exhibit strong humidity dependence, as long as the response follows a consistent and learnable trend, it can be effectively mitigated through calibration or inference, enabling reliable use in real- world indoor air quality monitoring.Machine Learning Models for Analyte Characterization in a Fluid Sample

[0283] In some embodiments, to assess the full sensing capabilities of the detection device, a targeted study was performed using benzene, toluene, and formaldehyde - three prevalent indoor VOCs with distinct health implications. Each gas was tested individuallyAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 across five concentrations (ranging from 0.25 to 5 ppm) using a dynamic exposure protocol in a sealed chamber with a controlled flow system. Time-resolved resistance data were collected from an array of chemiresi stive sensors mounted on the detection device. Each exposure cycle included baseline air flow, VOC introduction at fixed concentration, and a post-exposure purge, resulting in temporally structured response profiles for each compoundsensor pair. The goal of this experiment was not only to examine raw signal characteristics, but also to determine whether feature-rich sensor data could support robust classification and quantification of VOCs using data-driven approaches.

[0284] In some embodiments, dimensionality reduction through principal component analysis (PCA) revealed distinct chemical separability across the sensor feature space. The temporal profile of formaldehyde in response to a SGP41 sensor at 25 ppb (the lowest concentration tested uniformly across all three VOCs) is shown in Fig. 5A. The timedependent sensor signals clustered cleanly by compound in PCA space (Fig. 5B). The first two principal components captured 69.2% of the total variance, and notably, benzene and toluene — despite their structural similarity — occupied orthogonal trajectories relative to these components. This separation underscores the discriminative value of multivariate temporal patterns (temporal profiles) in the chemiresi stive signal. To further examine the influence of concentration, each compound’s vectorial evolution in PCA space was tracked across the full concentration range (Fig. 5C). The result was a set of compound-specific trajectories, each exhibiting unique directional growth in response to increasing analyte level. Importantly, these trajectories were not strictly radial or linear, highlighting that different VOCs modulate the sensor array in fundamentally different ways, even when concentration is the only changing variable.

[0285] In some embodiments, beyond principal component space, the raw sensor traces themselves revealed nuanced temporal behaviors that contributed to compound differentiation. Figs. 5D-5I visualizes the complete matrix of sensor features across the three VOCs. While many sensors responded to all three compounds (benzene, formaldehyde, and toluene) to varying degrees, the temporal kinetics (e.g., rise time, decay slope, peak width, and recovery profile) reflected in the temporal profile varied significantly between analytes. For instance, formaldehyde (Figs. 5F-5G) produced sharper adsorption and recovery transitions (less time to reach peak signal and return to base signal) compared to the broader, more gradual features (more time to reach peak signal and return to base signal) observed for toluene (Figs. 5H-5I). These kinetic fingerprints are not easily captured by static or peak-onlyAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 analyses, yet they form the basis of the advantage of temporal processing of signal data. In particular, subtle differences in response curvature and desorption lag contributed disproportionately to compound classification in downstream machine learning models. This is further supported by the sensor correlation maps in Fig. 5J, which illustrate both redundancy and divergence in feature relationships. Some sensor pairs showed high correlation (X-X), though they may have different correlations of response kinetics with different compounds (X-Y), thus providing additional identifying information from seemingly redundant sensors. This enables an array of broadly specific sensors to behave as a coherent and highly specific chemical classifier.

[0286] In some embodiments, leveraging this time-dependent dataset, support vector machine (SVM) models were trained for both classification and regression tasks. For analyte identity classification, the support vector classifier (SVC) achieved perfect accuracy (100%) across all test samples (n = 13), as visualized in the confusion matrix (Fig. 5K). This is particularly notable given the close chemical similarity of benzene and toluene, and the low absolute concentrations involved. Regression performance using support vector regression (SVR) was similarly strong (Fig. 5L). Across the same dataset, concentration predictions for benzene, formaldehyde, and toluene yielded R2values of 0.9997, 0.9451, and 0.9949, respectively, with root-mean-square errors (RMSE) of 6.09 ppb, 20.7 ppb, and 142 ppb. These levels of accuracy are well within regulatory thresholds for indoor air safety and, critically, approach or exceed the resolution offered by more expensive and complex detection systems such as GC-MS. The benzene model, in particular, performed with sub- 10 ppb resolution, which is significantly below the 1 ppm limit by OSHA and within striking range of 1 ppb aspiration level by LEED. Together, these results demonstrate that the combination of a low-cost array of sensors (e.g., chemiresi stive sensors) with temporal signal processing and machine learning yields a highly effective platform for real-time analyte (e.g., VOC) monitoring in indoor environments.Time-dependent Analyte Classification

[0287] In some embodiments, the chemiresi stive response of gas sensors is inherently time-resolved: signal evolution over time reflects the interplay of analyte adsorption to the sensors, surface reaction kinetics at the surface of the sensors, and desorption dynamics over the surface of the sensors. Static features, such as peak magnitude, discard this rich temporal structure. To investigate the added value of time-dependent analysis, a set of classificationAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 and regression tasks was designed comparing static approaches to models trained on full time-series data. Each analyte (benzene, toluene, and formaldehyde) was delivered at five concentrations (0.25 - 1.0 ppm) using the same controlled delivery protocol as in previous experiments. Fig. 6A displays representative SGP41 response curves (temporal profile) at 0.25 ppm, showing how even at the same concentration, the three analytes generate markedly different signal shapes. Benzene exhibits a faster rise and higher saturation (higher peak) than toluene, with formaldehyde returning more quickly (shorter time period) to baseline. These features are not well captured by static metrics such as maximum signal alone, limiting the range of compounds that may be effectively distinguished by a deployed sensor, and motivating a shift to time-aware modeling.

[0288] In some embodiments, to quantitatively assess the benefit of time-dependent analysis, random forest classifiers were trained on both static and time-resolved (temporal) data (Figs. 6B and 6C). In the static case (Fig. 6B), each sensor datapoint (sampled every 5 seconds) was treated as an independent input (n = 2952), using features such as max signal and mean slope. For the time-dependent model (Fig. 6C), each full exposure curve was treated as a single input sample (n = 42), allowing the model to learn from shape and dynamics. Classification accuracy improved modestly from 94.5% to 100%, suggesting that while static data are informative, the full curve provides greater discriminatory power and robustness. The true advantage of time-series analysis emerged in regression performance. Figs. 6D and 6E compare the results of support vector regression models trained to predict benzene concentration. The static model (Fig. 6D) exhibits significant scatter and multivalued predictions at several concentration levels, as evidenced by the background contour spread. This ambiguity reflects the limitations of using maximum signal alone, especially when signal saturation or sensor non-linearity occurs. In contrast, the time-dependent model (Fig. 6E) achieved an R2of 0.9997 and RMSE of just 6.09 ppb. The over an order of magnitude improvement demonstrates precise and unambiguous quantification enabled by temporal features.

[0289] In some embodiments, to visualize how sensor responses evolve over time for each compound, principal component trajectories were constructed using the full-time sequence (temporal profile) of the sensor array data (Figs. 7A-2 to 71-5). Each trial begins at a consistent baseline in PCA space, from which the response vector advances during analyte adsorption, reaches a peak corresponding to equilibration, and then returns along a desorption path. These closed-loop contours — here referred to as “leaflets” — exhibit unique shapes forAttorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 each VOC, determined by the interplay of molecular properties, surface interactions, and transport dynamics. The drawing in Fig. 7A showed an illustration of the leaflets and the adsorption, desorption, and equilibrium states of the analyte over the surface of the sensor. The trajectory’s direction reflects the relative timing and contribution of different sensor features, while the curvature encodes the kinetics of adsorption and recovery. The path of a particular analyte (e.g., a compound, a VOC) is therefore dependent on its unique physicochemical properties, which govern its interactions with sensor surfaces and transport behavior. As a result, both the shape and orientation of the leaflet provide a compoundspecific fingerprint of the sensor-analyte interaction. Furthermore, the area enclosed by the trajectory (leaflet) increases with concentration, reflecting the overall magnitude and duration of signal excursion in feature space. In some embodiments, the direction of the leaflet (theta) indicates the chemistry of the analyte. Together, these trajectories capture the multidimensional dynamics of the sensing process and offer a structured, interpretable basis for downstream machine learning and real-time chemical discrimination.

[0290] In some embodiments, the direction of the leaflet (theta, or 9) is obtained by computing the principal orientation of the trajectory formed in feature space during an analyte exposure cycle. In some embodiments, the trajectory is projected into a reduced- dimensionality space (e.g., first two principal components of a PCA, or other embedding). In some embodiments, a line is then fitted between the starting position (baseline prior to exposure) and the equilibrium adsorption position (steady-state response). In some embodiments, the direction 9 is defined as the angle of this fitted line relative to a reference axis (e.g., the first principal component axis or time axis). In some embodiments, 9 can be calculated as the arctangent of the slope determined by the change in signal features (Ay / Ax) between baseline and equilibrium. In some embodiments, this orientation encodes the dominant adsorption kinetics and reflects analyte-sensor chemical interactions, enabling chemical identity to be inferred from leaflet direction. For example, in some embodiments as shown in Figs. 7E-1 to 7E-5, toluene showed theta values range from 278.7 to 392.8 tested at a concentration range of 9.28 ppm to 9.5 ppm. The theta values were plotted in Fig. 71, where it showed that the mean theta value is 296.3°. Similarly, Figs. 7F-1 to 7F-5 showed that the theta values for formaldehyde are 396.9sto 336.1°. The mean theta value for formaldehyde is 316.6sas shown in Fig. 71. Similarly, Figs. 7G-1 to 7G-5 showed that the theta value for toluene was 289.8 to 301.6s. The mean theta value for toluene was 287.2 / as shown in Fig. 71.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0291] In some embodiments, quantification (e.g., concentration) could be decoded in area / volume (in higher dimensions) enclosed by the leaflet. In some embodiments, the enclosed area (or volume) of the leaflet reflected the concentration of the analyte. The enclosed areas of the leaflets in Figs. 7B-1 to 7G-5 were calculated and plotted in Fig. 7H. In some embodiments, the contour integral (area) of the leaflet increased as the concentration increases for toluene, formaldehyde, and benzene.

[0292] In some embodiments, device level limit of detection is reached when the leaflet shape was unrecognizable or / and the leaflet collapses. In some embodiments, such as shown in Figs. 7B-1, 7D-1, and 7F-1, the leaflet shape was close to being unrecognizable, which indicated the detection limit for the corresponding detective device was reached when the concentration level in the sample was tested. In some embodiments, such as shown in Figs. 7C-1 and 7E-1, the leaflet shape was close to being collapsed, which indicated the detection limit for the corresponding detective device was reached when the concentration level in the sample was tested.

[0293] In some embodiments, frequency of probing (i.e., how often the sensor has a readout within the time period) was based on leaflet geometrical resolution. In some embodiments, the leaflet has a smooth boarder line indicating sufficient frequency of probing. In some embodiments, the frequency of probing corresponds to the numbers of readouts or time points during the time period of testing. In some embodiments, the frequency of probing was proportional to the numbers of readouts or time points during the time period of testing. In some embodiments, the frequency of probing needed to be increased to achieve smoother outline of the leaflet for better precision.

[0294] In some embodiments, adaptively changing sensor frequency could distinguish certain analytes that cannot be distinguished otherwise. In some embodiments, especially when the target analytes had close chemical structures, their leaflet shapes may be close to each other and may not be distinguishable. In some embodiments, the leaflet shapes may be more defined with higher frequency of sensor detection, enabling the detective device to distinguish analytes that cannot be distinguished at lower frequencies.

[0295] In some embodiments, the sensor array and target classes can be varied, and the dimensionality of the analysis need not be limited to a specific technique such as principal component analysis (PCA) or t-distributed stochastic neighbor embedding (t-SNE). In some embodiments, more generally, the multi-dimensional sensor response space, denoted as IRn, where n represents the number of features or sensors, may be mapped into a lower-Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 dimensional representation H1, where m < n. In some embodiments, such dimensionality reduction can be accomplished using any linear or nonlinear transformation, including, but not limited to, linear projections (e.g., PC A, linear discriminant analysis), manifold learning approaches (e.g., t-SNE, UMAP, Isomap), or autoencoder-based mappings. In some embodiments, the mapping may extend beyond real-valued spaces into complex vector spaces (e.g., (Cm) or other structured mathematical spaces, thereby allowing the representation of phase, orientation, or frequency-domain features not easily captured in standard real- valued projections. In some embodiments, these mappings preserve, emphasize, or disentangle the geometric and statistical structure of sensor trajectories (“leaflets”), enabling analyte-specific signatures to be more effectively classified or regressed upon by downstream machine learning models.

[0296] In some embodiments, these findings underscore an important insight that timedependent analysis enables low-cost sensor (e.g., chemiresi stive sensor) arrays to operate with higher specificity and resolution by encoding dynamic response patterns, much like the mammalian olfactory system, which interprets odors not as static snapshots, but as evolving temporal signals tracking the entire signal arc rather than a single point. By mimicking this process, the detection device transforms basic sensors into a far more intelligent, adaptable detection system.Conclusion

[0297] In some embodiments, the detection device demonstrated that low-cost, rapidly prototyped sensor arrays can achieve high-performance indoor VOC monitoring when coupled with targeted data analysis strategies. Iterative PCB design and sensor selection yielded ppb-level limits of detection for benzene, toluene, formaldehyde, and BTEX mixtures, which are below the thresholds required by OSHA and LEED standards. Controlled studies under varying relative humidity revealed a consistent and quantifiable bias on sensor responses, where conventional linear regression models were outperformed by machinelearning-based inference for bias correction. This capability addresses one of the most persistent challenges in MOS gas sensing - environmental interference - without requiring additional hardware filtration. Critically, the approach capitalized on the intrinsic adsorptiondesorption kinetics of each sensor / compound pair, extracting richer temporal signatures than steady-state values could provide.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025

[0298] In some embodiments, dimensionality reduction via PCA confirmed distinct separation of analytes in feature space and served as a valuable interpretive tool, revealing a “leaflet” structure in the time-series data-patterns that make the underlying separation not only accessible to ML models, but also interpretable by human analysts. While support vector classifiers and regressors achieved competent performance, Random Forest models consistently outperformed them in both classification and concentration regression tasks, particularly in the challenging low-concentration regime. This performance advantage reflects RF’s ability to handle nonlinear decision boundaries and noisy sensor outputs, producing 100% classification accuracy and sub-100-ppb regression errors for target analytes. Together, the integration of kinetic-aware feature extraction, humidity -robust modeling, and transparent visualization frameworks establishes the detection device as both a deployable IAQ monitoring solution and a generalizable methodology for rapid, applicationspecific e-nose development.

[0299] Various modifications and additions can be made to the exemplary embodiments discussed without departing from the scope of the present disclosure. For example, while the embodiments described above refer to particular features, the scope of this disclosure also includes embodiments having different combinations of features and embodiments that do not include all of the described features. Accordingly, the scope of the present disclosure is intended to embrace all such alternatives, modifications, and variations as falling within the scope of the claims, together with all equivalents thereof.

Claims

Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025CLAIMSWhat is claimed is:

1. A detection device comprising: at least one sensor configured to detect a signal of a fluid sample comprising at least one analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated from a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; and a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors can be the same or different processors.

2. The detection device of claim 1, wherein the at least one sensor comprises an array of sensors.

3. The detection device of claim 2, wherein the sensors in the array of sensors are parallelly or sequentially connected.

4. The detection device of any of the preceding claims, wherein the detection device further comprises a redundant sensor configured to replace the at least one sensor in case of failure.

5. The detection device of claim 4, wherein the detection device is configured to connect or disconnect the redundant sensor.

6. The detection device of any of the preceding claims, wherein the detection device is a hand-hold gadget, a tabletop station, or an in-line apparatus of an industrial process.

7. The detection device of any of the preceding claims, wherein the at least one sensor is selected from the group consisting of a chemiresi stive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, and a hybrid multi-modal sensor.

8. The detection device of claim 7, wherein the gravimetric sensor is a quartz crystal microbalance sensor.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 20259. The detection device of any of the claims 1-7, wherein the at least one sensor is a chemiresi stive sensor.

10. The detection device of any of the claims 1-7, wherein the at least one sensor is a chemiresi stive metal-oxide sensor, a conductive polymer sensor, a carbon nanotube sensor, or a graphene sensor.

11. The detection device of any of the claims 1-7, wherein the at least one sensor is an optical sensor selected from the group consisting of a fluorescence sensor, a reflection sensor, a surface plasmon resonance sensor, and a Raman sensor.

12. The detection device of any of the claims 1-7, wherein the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, a MQ135 sensor, and a MiCS 5914 sensor.

13. The detection device of any of the claims 1-7, wherein the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, and a MiCS 5914 sensor.

14. The detection device of any of the claims 1-7, wherein the at least one sensor is selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a SCD40 sensor, a MQ138 sensor, a SEN0566 sensor, a MiCS5524 sensor, and a MQ135 sensor.

15. The detection device of any of the preceding claims, wherein the fluid sample comprises a liquid, a vapor, and / or a solid.

16. The detection device of any of the preceding claims, wherein the fluid sample is a vapor of a volatile compound mixed in a gaseous sample.

17. The detection device of any of the preceding claims, wherein the fluid sample is a vapor of a volatile compound, or a water vapor mixed in a gaseous sample.

18. The detection device of any of the preceding claims, wherein the analyte is a volatile compound.

19. The detection device of any of the preceding claims, wherein the analyte is selected from the group consisting of benzene, formaldehyde, toluene, xylenes, BTEX, methane, propane, a sulfur gas, ammonia, nitrogen dioxide, hydrogen, an alcohol, smoke, carbon monoxide, carbon dioxide, or a combination thereof.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 202520. The detection device of any of the preceding claims, wherein the detection device further comprises a humidity sensor to detect humidity of the fluid sample, and wherein the humidity sensor is configured to detect the humidity level from about 0% to about 90%.

21. The detection device of claim 20, wherein the detective device is configured to detect at a temperature ranges from -450F to 3000F.

22. The detection device of any of the preceding claims, wherein the first or second processor is configured to execute a machine learning model to reduce a humidity- induced detection error of the at least one sensor.

23. The detection device of any of the preceding claims, wherein the first processor is configured to process the readouts.

24. The detection device of any of the preceding claims , wherein the second processor is configured to execute a machine learning model.

25. The detection device of claim 24, wherein the machine learning model is a linear regression model, a support vector machine model, a support vector classifier model, a random forest model, or a deep neural network model.

26. The detection device of any of the preceding claims, wherein the memory is configured to be capable of storing additional temporal profile of new known analyte to the database.

27. The detection device of any of the preceding claims, wherein the memory is configured to be trained by a machine learning model to add additional temporal profile of new known analyte to the database.

28. The detection device of any of the preceding claims, wherein the fluid sample comprises at least two analytes; and wherein the second processor is configured to: deconvolute the plurality of readouts of the fluid sample to generate a temporal profile for each of the at least two analytes; and compare each of the temporal profile with the database to determine the identity and / or concentration of each of the at least two analytes.

29. The detection device of any of the preceding claims, wherein the detection device further comprises an additional sensor configured to detect an additional signal of the fluid sample at the plurality of time points within the time period to generate an additional plurality of readouts of the additional signal each measured at one of the plurality of time points within the time period;Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 the first processor is configured to generate an additional temporal profile of the additional plurality of readouts; and the second processor is configured to compare the additional temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample.

30. The detection device of any of the preceding claims, wherein the second processor is configured to determine the identity of the analyte by a linear regression model or a support vector regression model.

31. The detection device of any of the preceding claims, wherein the second processor is configured to determine the concentration of the analyte by a linear regression model or a support vector regression model.

32. The detection device of any of the preceding claims, wherein the temporal profile is converted to a leaflet comprising a closed-loop trajectory formed by projecting the temporal signal into a principal component space.

33. The detection device of claim 32, wherein the at least one sensor is configured to allow the adsorption and desorption of the analyte over a surface of the at least one sensor within the time period, and the leaflet is configured to reflect the adsorption and desorption process.

34. The detection device of claim 32, wherein the leaflet’s orientation is configured to reflect a pathway of the adsorption and desorption of the analyte over the surface of the at least one sensor.

35. The detection device of claim 32, wherein the direction of the leaflet reflects an identity of the analyte.

36. The detection device of claim 32, wherein a curvature of the leaflet is configured to reflect a hysteresis and recovery kinetics of the adsorption and desorption of the analyte over the surface of the at least one sensor.

37. The detection device of claim 32, wherein the geometrical resolution of the leaflet is configured to be higher with increased numbers of readouts at the time points within the time period.

38. The detection device of claim 32, wherein the enclosed area of the leaflet is configured to be proportional with a concentration of the analyte.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 202539. The detection device of any of the preceding claims, wherein the detection device further comprises an interface for inputting user instructions to control the at least one sensor and / or the first and / or second processors.

40. The detection device of any of the preceding claims, further comprises a communication module configured to communicate the generated temporal profile and the determined identity and / or concentration of the analyte to a user.

41. The detection device of any of the preceding claims, wherein the fluid sample is an air sample from a building, a ship, a hotel, a hospital, an industrial site, an industrial process, a farm, a wildfire detection site, a breath sample, or an atmosphere.

42. The detection device of any of the preceding claims, wherein the time period is about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or 15 seconds or about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, or 15 minutes.

43. The detection device of any of the preceding claims, wherein the at least one sensor frequency ranges from 1 Hz to 1G Hz.

44. The detection device of any of the preceding claims, wherein the time period comprises 5, 10, 20, 30, 40, 50, 60, 70, 80, 90 or 100 time points.

45. The detection device of claim any of the preceding claims, wherein the time period comprises 100, 500, 1000, 1500, 2000, 2500, 3000, 3500, 4000, 4500, 5000, 8000, 10000, 50000, 100000, 200000, 500000, 1 million, 1.5 million, 5 million, 10 million, 15 million, or 20 million time points.

46. The detection device of any of the preceding claims, wherein the detection device further comprises a gas-flow device configured to deliver the fluid sample to the at least one sensor.

47. The detection device of claim 46, wherein the detection device further comprises a gas-removal device to remove the fluid sample from the surface of the at least one sensor.

48. A detection device comprising an array of sensors configured to detect an analyte in a fluid sample, comprising: a first sensor selected based on a first property in its detection of the analyte; and a second sensor selected based on a second property different from the first property in its detection of the analyte;Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 wherein the first and second properties are each selected from the group consisting of limit of detection, sensitivity, reliability, reproducibility, and linearity in detection of the analyte.

49. The detection device of claim 48, wherein the first sensor’s first property is superior to the second sensor’s first property in detecting the analyte.

50. The detection device of any of the claims 48-49, wherein the second sensor’s second property is superior to the first sensor’s second property in detecting the analyte.

51. The detection device of any of the claims 48-50, wherein the detection device further comprises a third sensor selected based on a third property in its detection of the analyte, wherein the third property is selected from the group consisting of limit of detection, sensitivity, reliability, and linearity in detection of the analyte.

52. The detection device of any of the claims 48-51, wherein the limit of detection is from about 0.1 ppb to about 150 ppm.

53. The detection device of any of the claims 48-52, wherein the limit of detection is from about 0.01 ppb to about 50 ppb.

54. The detection device of any of the claims 48-53, wherein the limit of detection is from about 0.01 ppb to about 0.5 ppb.

55. The detection device of claim 52 or 54, wherein the analyte is benzene, formaldehyde, toluene, or BTEX.

56. The detection device of any of the claims 48-55, wherein the sensitivity is measured by a slope of the calibration curve.

57. The detection device of claim 56, wherein the slope of the calibration curve is from about 0.05 to 0.99.

58. The detection device of any of the claims 48-57, wherein the reliability is measured by signal-to-noise ratio.

59. The detection device of claim 58, wherein the signal-to-noise ratio is from about 0.0001 to about 0.18.

60. The detection device of any of the claims 48-59, wherein the linearity is measured by a coefficient of determination.

61. The detection device of claim 60, wherein the coefficient of determination is from about 0.26 to 0.999.

62. The detection device of any of the claims 48-61, wherein the first and / or second sensors are configured to detect the identity or concentration of the analyte.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 202563. The detection device of any of the claims 48-62, wherein the analyte is one or more volatile organic compounds.

64. The detection device of any of the claims 48-63, wherein the analyte is selected from the group consisting of benzene, formaldehyde, toluene, xylenes, BTEX, methane, propane, a sulfur gas, ammonia, nitrogen dioxide, hydrogen, an alcohol, smoke, carbon monoxide, carbon dioxide, or a combination thereof.

65. The detection device of any of the claims 48-64, wherein the analyte is benzene, toluene, formaldehyde, BTEX, or a combination thereof.

66. The detection device of any of the claims 48-65, wherein the at least one sensor is selected from the group of a chemiresi stive sensor, an optical sensor, a gravimetric sensor, an electrochemical sensor, and a hybrid multi-modal sensor.

67. The detection device of any of the claims 48-66, wherein the first and second sensors are chemiresi stive sensors.

68. The detection device of any of the claims 48-67, wherein the first and second sensors are chemiresistive metal-oxide sensors.

69. The detection device of any of the claims 48-68, wherein the first and second sensors are each selected from the group consisting of a SGP41 sensor, a ENS160 sensor, a MQ138 sensor, a MiCS5524 sensor, and a SEN0566 sensor, and a SCD40 sensor.

70. The detection device of any of the claims 48-69, wherein the first sensor is a SGP41 sensor selected for its reproducibility and low limitation of detection of the analyte.

71. The detection device of any of the claims 48-69, wherein the first sensor is a ENS160 sensor selected for its heightened sensitivity in detecting the fluid sample.

72. The detection device of any of the claims 48-69, wherein the first sensor is a MQ138 sensor selected for its high linearity in detecting the fluid sample to provide complimentary properties different from the properties from the second sensor.

73. The detection device of any of the claims 48-69, wherein the first sensor is a MiCS5524 sensor selected for its low limit of detection and high linearity in detecting the fluid sample.

74. The detection device of any of the claims 48-69, wherein the first sensor is a SEN0566 sensor selected for its high linearity in detecting the fluid sample.

75. The detection device of claim any of the claims 48-74, wherein the detection device comprises a SGP41 sensor and a ENS160 sensor.

76. The detection device of claim 75, further comprises a SCD 40 sensor.Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 202577. The detection device of claim 76, further comprises a MQ138-B sensor, a SEM0566- B sensor, and / or a MiCS5524-B sensor.

78. The detection device of claim 77, further comprises a MiCS5524-B sensor and a MiCS5914 sensor.

79. The detection device of any of the claims 48-78, wherein the array of sensors comprises a SGP41 sensor, a MiCS5524 sensor, a ENS160 R2 sensor, a ENS160 R3 sensor, and a MQ138 sensor.

80. The detection device of claim any of the claims 48-79, wherein at least one of the first sensor and the second sensor is configured to detect a signal of the analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; and wherein the detection device further comprises a first processor configured to generate a temporal profile of the readouts; a memory storing a database comprising a plurality of temporal profiles, each generated by a plurality of readouts of the signal of a known analyte measured by the sensor at the plurality of time points within the time period; and a second processor configured to compare the generated temporal profile of the fluid sample with the database to determine the identity and / or concentration of the analyte in the fluid sample; wherein the first and second processors are the same or different processors.

81. A method of analyzing a fluid sample, comprising: providing a detection device of any of the claims 1 to 47, detecting the signal of the fluid sample at the plurality of time points within the time period to generate the plurality of readouts of the signal each measured at one of the plurality of the time points within the time period; generating the temporal profile of the readouts by the first processor; and comparing the generated temporal profile of the fluid sample with the database by the second processor to determine the identity and / or concentration of the analyte in the fluid sample.

82. The method of claim 81, wherein the fluid sample comprises at least two analytes; and the method further comprises:Attorney Docket Number: 0042697.00611WO1Date of Electronic Filing: September 24, 2025 deconvolute the generated temporal profile of the fluid sample to generate a temporal profile for each of the at least two analytes; and comparing the temporal profile of each of the at least two analytes with the database to determine the identity and / or concentration of each of the at least two analytes.

83. A method of detecting an analyte in a fluid sample, comprising: providing the detection device of claims 48 to 80; detecting the analyte by the first sensor; and detecting the analyte by the second sensor.

84. The method of claim 83, further comprising determining the limit of detection, sensitivity, reliability, reproducibility, or linearity in detection of the analyte by the first or second sensor.

85. The method of claim 84, wherein the method further comprises: detecting a signal of the analyte at a plurality of time points within a time period to generate a plurality of readouts of the signal measured at one of the plurality of the time point within the time period by the first or second sensor; generating a temporal profile of the readouts by a first processor; and comparing the generated temporal profile of the fluid sample with the database by a second processor to determine the identify and / or concentration of the analyte in the fluid sample.

Images