Handheld AI electronic nose device, system and method

Abstract

The invention provides a handheld AI electronic nose device, system and method. The electronic nose device (901) comprises an environmental parameter monitoring unit (101), a multi-channel metal oxide gas sensor array unit (102), a gas path-gas chamber integrated assembly (103), an integrated pump valve module (104), a display and input module (105), an acquisition and execution control panel (106), a high-resolution analog-to-digital conversion module (107), an edge AI processing module (108) and a battery and management module (109). The detection efficiency can be improved, the long-term stability is enhanced, edge intelligence and controllable self-adaption are achieved, and portability and integrated optimization are achieved.

Description

Technical Field

[0001] This disclosure relates to the field of gas detection device technology, specifically to a handheld AI electronic nose device, system, and method. Background Technology

[0002] Metal oxide sensors (MOS) are widely used in gas detection due to their low cost and high sensitivity. However, existing MOS-based electronic nose systems still face a series of technical challenges in practical applications, especially in portable, on-site rapid detection scenarios.

[0003] First, the detection efficiency is low: the working cycle of a MOS sensor typically includes multiple stages such as preheating, adsorption detection, desorption cleaning, and recovery. In existing technologies, the preheating period often takes tens of seconds to several minutes to reach a stable operating temperature; the detection period requires waiting for the sensor response to reach a steady state before reliable judgment can be made, which is significantly time-consuming; the desorption cleaning stage is inefficient, relying on natural desorption or isothermal cleaning, resulting in difficulty in removing residual gases and slow baseline recovery. This makes a single complete detection cycle usually exceed 3 minutes, which is difficult to meet the time-sensitive application requirements such as emergency response and rapid screening in medical settings. Although there have been attempts to shorten some steps by using algorithms to make early judgments (early stop) or optimizing temperature control, the effect of a single strategy is limited. That is, relying solely on algorithms to stop early is prone to misjudgment, and optimizing temperature control alone cannot solve the problem of the lengthy detection period itself.

[0004] Secondly, environmental and device drift have a significant impact: MOS sensor performance is susceptible to environmental factors (such as fluctuations in temperature, humidity, air pressure, and gas flow rate) and the device's own condition (such as aging and contamination), leading to problems such as baseline drift, sensitivity changes, and response time alterations. This drift significantly reduces the accuracy and long-term stability of detection results, manifesting as an increased false positive or false negative rate. Traditional fixed threshold discrimination or offline calibration methods struggle to adapt to these dynamic changes in real time, especially in complex and variable field environments, making it difficult to guarantee the reliability of detection results. Existing technologies lack a real-time, online hierarchical compensation and self-calibration mechanism to systematically address drift from multiple sources.

[0005] Secondly, there is a lack of intelligence and adaptability: existing electronic nose systems mostly rely on the cloud for inference and updates of complex models. This results in devices being unable to work effectively or update models in environments with limited or no network access (such as disaster sites, the field, or secure isolation zones). Cloud-based model update cycles are long, making it difficult to quickly adapt to individual sensor differences, new environmental background gases, or sensor drift. Furthermore, uploading raw gas response data to the cloud for processing also poses privacy and security risks, such as the leakage of sensitive information (e.g., medical respiratory characteristics). Therefore, there is an urgent need for a mechanism that can achieve efficient and reliable inference and controllable adaptive learning at the device's local edge.

[0006] Finally, the system suffers from poor integration and portability: existing systems are mostly laboratory or benchtop structures, large in size and high in power consumption, making it difficult to meet the needs of portable and handheld applications. Gas path designs are often complex, resulting in large dead spaces and uneven gas flow across sensor channels, leading to asynchronous gas arrival times, affecting the accuracy of early feature extraction and multi-channel fusion, and increasing the risk of cross-contamination from residual gases. The dispersed layout of multiple modules is susceptible to electromagnetic interference (such as the impact of pump and valve actuation on weak signal acquisition), and lacks safety isolation design for high-temperature operating areas, which is detrimental to the safety and user experience of handheld operation. Insufficient power consumption optimization also limits the battery life of battery-powered devices.

[0007] In summary, the existing MOS electronic nose system has many shortcomings in terms of detection speed, anti-drift capability, environmental adaptability, intelligence level, portability and safety, which restricts its widespread application in rapid and reliable on-site detection scenarios. Summary of the Invention

[0008] (a) Technical problems to be solved

[0009] In view of the above problems, this disclosure provides a handheld AI electronic nose device, system and method to at least partially solve the problems of low efficiency, significant influence of environmental and device drift, insufficient intelligence and adaptability, poor system integration and portability of traditional gas detection.

[0010] (II) Technical Solution

[0011] This disclosure provides a handheld AI electronic nose device, comprising: an electronic nose device including an environmental parameter monitoring unit, a multi-channel metal oxide gas sensor array unit, an integrated gas path-gas chamber assembly, an integrated pump and valve module, a display and input module, a data acquisition and execution control board, a high-resolution analog-to-digital converter module, an edge AI processing module, and a battery and management module; the environmental parameter monitoring unit is located at the central mounting position of the multi-channel metal oxide gas sensor array unit, which is disposed within the integrated gas path-gas chamber assembly; the integrated pump and valve module is connected to the integrated gas path-gas chamber assembly; the data acquisition and execution control board is connected to the environmental parameter monitoring unit and the multi-channel metal oxide gas sensor array unit; the high-resolution analog-to-digital converter module is connected above the data acquisition and execution control board, converting the analog data acquired by the data acquisition and execution control board into digital data, and transmitting the converted digital data to the edge AI processing module connected below; the edge AI processing module is connected to and sends control commands back to the data acquisition and execution control board; the battery and management module provides energy and safety management for the electronic nose device; the display and input module displays the mode, temperature control curve, progress, results, and prompts of the electronic nose device.

[0012] The second aspect of this disclosure provides a handheld AI electronic nose system, including the aforementioned handheld AI electronic nose device. The system is functionally divided into a perception layer, a decision-making layer, and a control layer, which interact through a unified message / event bus.

[0013] According to embodiments of this disclosure, the sensing layer comprises an environmental parameter monitoring unit, a multi-channel metal oxide gas sensor array unit, an integrated gas path-gas chamber assembly, an integrated pump-valve module, and a high-resolution analog-to-digital conversion module, used for gas sampling and signal acquisition; gas introduction and distribution: external gas enters via the integrated pump-valve module, wherein the pump and valve are integrated in a compact structure, the pump is located on the left side of the integrated pump-valve module, and the valve is located on the right side of the integrated pump-valve module; the pump provides the driving force for gas flow, and the valve acts as an on / off switch and flow path switch; external gas enters the integrated gas path-gas chamber assembly through the valve under the action of the pump. The gas enters the sampling microcavities through the central channel and then through the inlets of each sampling microcavity. Within these microcavities, the gas interacts fully with the sensitive layer of the multi-channel metal oxide gas sensor array unit located at the sensor mounting interface. The gas exits from the sampling microcavities, flows into the central channel, and exits through the gas outlet. Signal and environmental quantity acquisition: A high-resolution analog-to-digital converter performs high-precision synchronous sampling of the multi-channel sensor signals from the multi-channel metal oxide gas sensor array unit. An environmental parameter monitoring unit outputs real-time environmental parameters, including temperature, humidity, flow rate, and air pressure, for subsequent online compensation and self-calibration. The sensing layer outputs the generated multi-channel sensor signals and environmental parameters to the decision layer.

[0014] According to embodiments of this disclosure, the decision layer comprises a dynamic temperature control and early stop discrimination unit, an anti-drift and self-calibration unit, and an edge AI and local incremental learning unit, operating with the data support of the temperature control / execution of the acquisition and execution control board and the high-resolution analog-to-digital conversion module. The dynamic temperature control and early stop discrimination unit, based on the temperature control curve issued by the acquisition and execution control board, sequentially executes pulse rapid preheating, low-temperature adsorption detection, two-stage early stop sequential probability ratio testing, multi-outlet temporal convolutional network, and high-temperature desorption cleaning processes. When a threshold is met, a discrimination conclusion and confidence level are generated, and a cut-in cleaning signal is sent to the control layer. The trigger signal; the anti-drift and self-calibration unit takes the environmental parameters of the environmental parameter monitoring unit and the historical baseline of the multi-channel metal oxide gas sensor array unit / high-resolution analog-to-digital conversion module as inputs, performs online environmental compensation, anchor baseline calibration, channel health scoring and gating / masking, confidence calibration, conformal rejection and outlier distribution identification, and outputs parameter update and resampling instructions; the edge AI and local incremental learning unit runs on the edge AI processing module, performs model inference, sample screening, constrained incremental update and version rollback on the preprocessed multi-channel data, and generates model / log versions.

[0015] According to embodiments of this disclosure, the control layer comprises an acquisition and execution control board, a high-resolution analog-to-digital conversion module, an edge AI processing module, a battery and management module, and a display and input module, responsible for execution, acquisition, energy consumption, and interaction. Execution and acquisition: The acquisition and execution control board controls the heating and temperature measurement of the integrated pump-valve module and the multi-channel metal oxide gas sensor array unit, and works in conjunction with the high-resolution analog-to-digital conversion module to achieve multi-channel synchronous sampling and preprocessing. Local computation: The edge AI processing module completes inference, confidence calibration, and incremental learning, and sends the control commands back to the acquisition and execution control board. Power supply and interaction: The battery and management module provides energy and safety management. The display and input module displays the mode, temperature control curve, progress, results, and prompts. Data and control path: The target link is the high-resolution analog-to-digital conversion module connected to the acquisition and execution control board, the acquisition and execution control board connected to the edge AI processing module, and the edge AI processing module connected to the display and input module. The control loop is the edge AI processing module connected to the acquisition and execution control board, used to trigger pump / valve / temperature control actions.

[0016] This disclosure provides a handheld AI electronic nose method, including the aforementioned handheld AI electronic nose system. The method includes: Step S1: Performing the initialization operation of the system and establishing a baseline profile; Step S2: Performing sampling start point detection and data preprocessing; Step S3: Performing a pulsed rapid preheating process in the first stage; Step S4: Performing low-temperature adsorption detection in the second stage and performing multi-level early stop discrimination; Step S5: Implementing online environmental compensation and data normalization processing; Step S6: Performing baseline tracking, anchor point calibration, and channel health status assessment; Step S7: Performing threshold adaptive adjustment, uncertainty control, and unknown class identification; Step S8: Performing constrained incremental learning and rollback operations at the edge AI end; Step S9: Implementing closed-loop collaborative control of dynamic temperature control and early stop strategy; Step S10: Performing system structure optimization and power consumption management; Step S11: Adjusting the parameter range and working mode according to the operating status.

[0017] According to an embodiment of this disclosure, the initialization operation of the system and the establishment of a baseline profile include: Step S1.1: Clean gas baseline acquisition: After the device is first deployed or powered on, the clean gas is automatically switched, and multi-channel signals for a preset time period are acquired to obtain the initial baseline for each channel. Noise flooring Dynamic range scale With rise time constant And simultaneously record the reference environment vector Step S1.2: Initial value fitting for environmental compensation: Using the reference environmental vector e0 as a reference, fit the environmental sensitivity coefficients of each channel; Step S1.3: Version management of calibration files: Generate calibration files with timestamps and store them in the database.

[0018] According to embodiments of this disclosure, the second stage of performing low-temperature adsorption detection and executing multi-level early cessation discrimination includes: Step S4.1: Low-temperature adsorption: acquiring response signals on a low-temperature platform and extracting features including early slope, amplitude, and cross-channel ratio from the response signals to improve information density; Step S4.2: Executing statistical early cessation discrimination based on sequential probability ratio test: calculating the log-likelihood ratio in real time. Defined as:

[0019]

[0020] in, This represents the alternative assumption that the target gas exists. This indicates the null hypothesis that the target gas does not exist. Represents the k-th observation; when Accept in advance when ≥logA ,when Accept when ≤logB ;when When in the hysteresis interval, stabilization is achieved by combining sliding window consistency and double threshold hysteresis; Step S4.3: Multi-exit deep early stop: Based on the temporal convolutional network, multiple exit points are set within the preset response time interval. When the discrimination confidence of any exit reaches the threshold, it is output in advance; The position of the exit is adjusted according to the scenario; Step S4.4: Perform fallback complete acquisition: If any of the above early stops is not triggered or the confidence does not meet the preset conditions, the response signal continues to be acquired until the predefined acquisition period.

[0021] According to embodiments of this disclosure, the execution of baseline tracking, anchor point calibration, and channel health status assessment includes: Step S6.1: Baseline tracking: Measuring and acquiring the latest baseline during the interval between the end of post-cleaning or between two detections. ,noise and time constant ; Calculate drift index ,in, , , The reference benchmark value is used as a reference. If any indicator exceeds the adaptive threshold band, the adaptive threshold band is determined based on the quantiles of the historical distribution. The target quantile interval is selected and marked as a "deviation" state. Step S6.2: Anchor point calibration: Triggered by a fixed period or the "deviation" condition generated in S6.1, the clean gas is switched to perform short sampling to refresh the reference benchmark value. With reference environment; and simultaneously fine-tuning S5 Coefficients; Step S6.3: Health Score and Gating: Define the health score of channel i at time t. for:

[0022]

[0023] in, This is a normalization function that maps the deviation to the interval [0,1]. Assess the cross-correlation consistency score between this channel and the other channels; w k Configurable weights; applied before feature fusion / model input. Channels are weighted; channels whose health scores are continuously below a preset threshold are placed into candidate blocking status and the event is reported.

[0024] (III) Beneficial Effects

[0025] 1. Improved detection efficiency: By using pulse rapid preheating, multi-stage early stop discrimination during the low-temperature adsorption period (SPRT and multi-outlet TCN), and high-temperature desorption and cleaning after early stop triggering, this system can shorten the time of the three stages of preheating, adsorption and cleaning, and shorten the overall detection cycle while ensuring the reliability of discrimination.

[0026] 2. Enhanced long-term stability: Through layered anti-drift and self-calibration mechanisms such as online environmental compensation, anchor baseline calibration, and channel health gating / shielding, the effects of temperature, humidity, flow rate, and device aging can be reduced, thereby maintaining the long-term consistency and reliability of test results.

[0027] 3. Edge intelligence and controllable adaptation: Inference and constrained incremental learning are performed locally on the device, and combined with versioning and rollback mechanisms, it can adapt to new scenarios and individual differences even in environments without or with weak networks, while avoiding the uploading of private data and improving the independence and security of the system.

[0028] 4. Portability and Integrated Optimization: It adopts a CFD-optimized one-to-many small dead-cavity air chamber, low-adsorption inner wall and pump valve integrated package, combined with analog / digital / power three-domain isolation circuit and power consumption management, to improve multi-channel response synchronization and signal-to-noise ratio, reduce cross-contamination, and meet the needs of handheld applications in terms of size, power consumption and thermal insulation safety. Attached Figure Description

[0029] To gain a more complete understanding of this disclosure and its advantages, reference will now be made to the following description taken in conjunction with the accompanying drawings, wherein:

[0030] Figure 1 A schematic perspective view of the electronic nose device structure provided in an embodiment of this disclosure is shown.

[0031] Figure 2 This schematic diagram illustrates the structure of the integrated gas path-gas chamber assembly provided in an embodiment of the present disclosure;

[0032] Figure 3 A flowchart illustrating the rapid detection process provided in an embodiment of this disclosure is shown schematically.

[0033] Figure 4 The flowchart illustrating the anti-drift and self-calibration process provided in the embodiments of this disclosure is shown in the schematic diagram.

[0034] Figure 5 This schematically illustrates a flowchart of edge AI and local incremental learning provided in an embodiment of the present disclosure;

[0035] Figure 6 The schematic diagram illustrates the early stopping detection principle provided in the embodiments of this disclosure;

[0036] Figure 7 The illustration shows experimental data diagrams from Embodiment 1 provided in this disclosure.

[0037] Figure 7 A schematic illustration shows the classification confusion matrix of the pest Bactroceradorsalis at a normal decision point in Embodiment 1 of this disclosure, showing an overall classification accuracy of 95.5%.

[0038] Figure 7 B schematically illustrates the classification confusion matrix of the same pest at the earliest decision point provided in the embodiments of this disclosure, showing an overall classification accuracy of 90.9%;

[0039] Figure 7 C schematically illustrates a bar chart showing the average time required for each infection group of Bactrocera dorsalis to reach a classification decision, as provided in the embodiments of this disclosure.

[0040] Figure 7 D schematically illustrates the classification confusion matrix for the pest Bactrocera correcta at a normal decision point provided in the embodiments of this disclosure, showing an overall classification accuracy of 93.3%;

[0041] Figure 7 E schematically illustrates the classification confusion matrix of the same pest at the earliest decision point provided in the embodiments of this disclosure, showing an overall classification accuracy of 88.9%;

[0042] Figure 7 F schematically illustrates a bar chart showing the average time required for each infection group of Bactrocera correcta to reach a classification decision, as provided in the embodiments of this disclosure.

[0043] Figure 8This illustration schematically shows an early stop decision-making diagram provided in an embodiment of the present disclosure;

[0044] Figure 9 The illustration shows a schematic diagram of the collaborative inspection of a customs conveyor belt and a robotic arm provided in an embodiment of this disclosure.

[0045] Explanation of reference numerals in the attached figures:

[0046] 101-Environmental parameter monitoring unit; 102-Multi-channel metal oxide gas sensor array unit; 103-Integrated gas path-gas chamber assembly; 104-Integrated pump and valve module; 105-Display and input module; 106-Acquisition and execution control board; 107-High-resolution analog-to-digital conversion module; 108-Edge AI processing module; 109-Battery and management module; 201-Air inlet; 202-Sampling microcavity; 203-Sampling microcavity inlet; 204-Sampling microcavity outlet; 205-Sensor mounting interface; 206-Air outlet; 901-Electronic nose device; 902-Robotic arm; 903-Conveyor belt; 904-Sample; 905-Sample box. Detailed Implementation

[0047] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0048] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0049] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0050] 1. Overall System Composition

[0051] like Figure 1As shown, the handheld AI electronic nose device 901 disclosed herein includes an environmental parameter monitoring unit 101, a multi-channel metal oxide gas sensor array unit 102, an integrated gas path-gas chamber assembly 103, an integrated pump and valve module 104, a display and input module 105, an acquisition and execution control board 106, a high-resolution analog-to-digital conversion module 107, an edge AI processing module 108, and a battery and management module 109.

[0052] The environmental parameter monitoring unit 101 is located in the central mounting position of the multi-channel metal oxide gas sensor array unit 102. The multi-channel metal oxide gas sensor array unit 102 is set inside the gas path-gas chamber integrated component 103. The integrated pump valve module 104 is connected to the gas path-gas chamber integrated component 103. The acquisition and execution control board 106 is connected to the environmental parameter monitoring unit 101 and the multi-channel metal oxide gas sensor array unit 102. The high-resolution analog-to-digital conversion module 107 is connected to the acquisition and execution control board 106 above and converts the analog data it acquires into digital data. Then, the converted digital data is transmitted to the edge AI processing module 108 connected below. The edge AI processing module 108 is connected and sends control commands back to the acquisition and execution control board 106. The battery and management module 109 provides energy and safety management for the electronic nose device 901. The display and input module 105 displays the mode, temperature control curve, progress, results and prompts of the electronic nose device 901.

[0053] The environmental parameter monitoring unit 101 is located in the central mounting position of the multi-channel metal oxide gas sensor array unit 102. It collects environmental parameters such as temperature, humidity, flow rate and air pressure as inputs for online compensation and self-calibration.

[0054] The multi-channel metal oxide gas sensor array unit 102 can preferably have 8 channels, each channel having independent heating and independent temperature measurement, for performing differentiated dynamic temperature control.

[0055] like Figure 2 As shown, the integrated gas path-gas chamber assembly 103 includes an air inlet 201, a central channel, at least one sampling microcavity 202, and an air outlet 206. The sampling microcavity 202 is provided with a sampling microcavity inlet 203 and a sampling microcavity outlet 204. A sensor mounting interface 205 is provided on the lower inner side of the sampling microcavity 202. A sensitive layer of a multi-channel metal oxide gas sensor array unit 102 is provided at the sensor mounting interface 205 to ensure that the gas acts on the sensitive layer of the multi-channel metal oxide gas sensor array unit 102 within the sampling microcavity 202.

[0056] The integrated pump and valve module 104 is connected to the integrated gas path-gas chamber component 103, supporting constant flow sampling, pulse sampling and cleaning modes.

[0057] The acquisition and execution control board 106 is responsible for the initial acquisition of sensor signals, the execution of array heating curves and pump valve drive control, and is synchronized with the high-resolution analog-to-digital conversion module 107.

[0058] The high-resolution analog-to-digital converter module 107 is used to perform high-precision (multi-channel synchronous) sampling of the output of each channel of the multi-channel metal oxide gas sensor array unit 102 to ensure the quality of weak signals in the early stage.

[0059] The edge AI processing module 108 is used to perform model inference, confidence calibration and constrained incremental learning locally, and to version and roll back.

[0060] The battery and management module 109 is used to provide power, charge / discharge and safety management, and supports low-power sleep mode.

[0061] The display and input module 105 is used for mode selection and result display.

[0062] 2. System Functional Architecture

[0063] Based on the aforementioned device, the electronic nose system is functionally divided into a perception layer, a decision-making layer, and a control layer, which interact through a unified message / event bus.

[0064] The sensing layer consists of an environmental parameter monitoring unit 101, a multi-channel metal oxide gas sensor array unit 102, an integrated gas path-gas chamber assembly 103, an integrated pump and valve module 104, and a high-resolution analog-to-digital conversion module 107, which are used to complete gas sampling and signal acquisition.

[0065] Specifically, firstly, the sensing layer can be used for gas introduction and distribution: external gas first enters through the integrated pump-valve module 104, where the pump and valve are integrated into a compact structure. The pump is located on the left side of the integrated pump-valve module 104, and the valve is located on the right side. The pump provides the driving force for gas flow, while the valve acts as an on / off switch and flow path switch. Under the action of the pump, the external gas enters the inlet 201 of the integrated gas path-gas chamber assembly 103 through the valve, and then enters the sampling microcavities 202 through the central channel and the inlets 203 of each sampling microcavity. In the sampling microcavities 202, the gas fully interacts with the sensitive layer of the multi-channel metal oxide gas sensor array unit 102 located at the sensor mounting interface 205, and then flows into the central channel from the sampling microcavity outlet 204 and is discharged through the outlet 206. Secondly, it can also be used for signal and environmental quantity acquisition: the high-resolution analog-to-digital conversion module 107 performs high-precision synchronous sampling of the multi-channel electrical signals (i.e., sensor signals) of the multi-channel metal oxide gas sensor array unit 102; the environmental parameter monitoring unit 101 outputs environmental parameters such as temperature, humidity, flow rate, and air pressure in real time for subsequent online compensation and self-calibration. Finally, the sensing layer can output to the decision layer: forming an input of "multi-channel sensor signals + environmental parameters" (corresponding to the full-link S2 described later).

[0066] The decision-making layer consists of three functional units: a dynamic temperature control and early stop discrimination unit, an anti-drift and self-calibration unit, and an edge AI and local incremental learning unit. It operates with the data support of the temperature control / execution and high-resolution analog-to-digital conversion module 107 of the acquisition and execution control board 106.

[0067] Among them, such as Figure 3 As shown, the dynamic temperature control and early stop discrimination unit executes the following process based on the temperature control curve issued by the acquisition and execution control board 106: pulse rapid preheating → low-temperature adsorption detection → two-stage early stop (sequential probability ratio test (SPRT) and multi-outlet temporal convolutional network (TCN)) → high-temperature desorption cleaning. When the threshold is met, a discrimination conclusion and confidence level are generated, and a trigger signal for starting cleaning (corresponding to S3–S4) is sent to the control layer.

[0068] like Figure 4 As shown, the anti-drift and self-calibration unit takes the environmental parameters of the environmental parameter monitoring unit 101 and the historical baseline of the multi-channel metal oxide gas sensor array unit 102 / high-resolution analog-to-digital conversion module 107 as inputs, performs online environmental compensation, anchor baseline calibration, channel health scoring and gating / masking, confidence calibration, conformal rejection and outlier distribution (OOD) identification, and outputs parameter update and resampling instructions (corresponding to S5).

[0069] like Figure 5As shown, the edge AI and local incremental learning unit run on the edge AI processing module 108 to perform model inference, sample selection, constrained incremental updates and version rollback on the preprocessed multi-channel data, and generate model / log versions (corresponding to S6).

[0070] The control layer consists of an acquisition and execution control board 106, a high-resolution analog-to-digital conversion module 107, an edge AI processing module 108, a battery and management module 109, and a display and input module 105, and is responsible for execution, acquisition, energy consumption, and interaction.

[0071] Execution and Acquisition: The acquisition and execution control board 106 controls the integrated pump valve module 104 and the multi-channel metal oxide gas sensor array unit 102 for heating / temperature measurement, and works in conjunction with the high-resolution analog-to-digital conversion module 107 to achieve multi-channel synchronous sampling and preprocessing.

[0072] Local computing: The edge AI processing module 108 completes inference, confidence calibration and incremental learning, and sends control commands (such as cut-in cleaning / re-sampling) back to the acquisition and execution control board 106.

[0073] Power supply and interaction: Battery and management module 109 provides energy and safety management; display and input module 105 displays modes, temperature control curves, progress, results and prompts.

[0074] Data and control path: The typical target link is high-resolution analog-to-digital conversion module 107 → acquisition and execution control board 106 → edge AI processing module 108 → display and input module 105; the control loop is edge AI processing module 108 → acquisition and execution control board 106 (triggering pump valve / temperature control action).

[0075] 3. End-to-end workflow

[0076] Step S1: Initialization and baseline profile establishment (deployment / power-on).

[0077] Step S1.1: Clean Gas Baseline Acquisition: After the device is initially deployed or powered on, the clean gas is automatically switched, and multi-channel signals are acquired for a short period of time (i.e., a preset period) to obtain the initial baseline for each channel. Noise flooring Dynamic range scale With rise time constant Simultaneously record the reference environment vector .

[0078] Step S1.2: Initial value fitting for environmental compensation: Using e0 as a reference, fit the environmental sensitivity coefficient of each channel (linear or second-order polynomial; initial values ​​of Kalman / Bayes filter parameters can be selected).

[0079] Step S1.3: Version management of calibration files: Generate "calibration file v0" (including channel parameters, environmental regression coefficients, initial threshold values, and rejection strategy parameters), store it in the database, and add a timestamp.

[0080] Step S2: Sampling start point detection and data preprocessing (started with each detection).

[0081] Starting point detection: The intake starting point is determined by cross-correlation between pump flow rate, valve switch, and channel signal.

[0082] Preprocessing: Denoising (preferably low-order Savitzky-Golay or Kalman filtering), zero-mean / dimensional normalization, and channel filtering are performed on the original sequence to form a preprocessed data stream; the acquisition link may include a high-resolution analog-to-digital conversion module 107 (such as ADS1256) to ensure the quality of weak early signals.

[0083] Step S3: In the first stage, i.e., stage I, pulsed rapid preheating (preheating acceleration) is performed.

[0084] Short-duration high-power heating enables the sensor to quickly enter the low-temperature detection platform (located in the medium-low temperature range of the safe operating range, preferably reaching 200–300℃ within 1–5 seconds); closed-loop temperature measurement is used to prevent overshoot; the preheating sequence and amplitude can be adaptively adjusted according to the channel heat capacity and array consistency.

[0085] Step S4: In the second stage, namely stage II, low-temperature adsorption detection + multi-stage early stop judgment (mid-stage acceleration) is performed.

[0086] Step S4.1: Low-temperature adsorption: Collect response signals on a low-temperature platform and extract features such as early slope, amplitude and cross-channel ratio from the response signals to improve information density.

[0087] Step S4.2: Perform statistical early stopping discrimination based on the sequential probability ratio test, i.e., statistical early stopping (Level-1, SPRT): calculate the log-likelihood ratio in real time using the sequential probability ratio test. , Defined as:

[0088] in, This represents the alternative assumption that the target gas exists. This indicates the null hypothesis that the target gas does not exist. This represents the k-th observation.

[0089] when Accept in advance when ≥logA (Entering cleaning), when Accept when ≤logB ;when When in the hysteresis interval, stabilization is achieved by combining sliding window consistency and double threshold hysteresis.

[0090] Step S4.3: Multi-exit deep early stop (Level-2, TCN): Based on the temporal convolutional network (TCN), multiple exit points are set within a preset response interval (e.g., 1–6s) (preferably 2–5s). When the discrimination confidence of any exit reaches the threshold, it is output in advance; the exit position can be adaptive according to the scenario, rather than a fixed specific number of seconds.

[0091] Step S4.4: Fallback, i.e., perform fallback complete acquisition: If any of the above early stops is not triggered or the confidence level is insufficient, i.e. the preset conditions are not met, then continue to acquire response signals until the complete cycle (predefined acquisition cycle).

[0092] Step S5: Online environmental compensation and data normalization processing (throughout the entire process).

[0093] like Figure 4 As shown, dynamic compensation is performed on each channel, and the compensated signal... Calculated using the following formula:

[0094]

[0095] in, β represents the original signal of the i-th sensing channel at time t; i,T β i,H β i,P β i,Q This represents the first-order compensation coefficient of the i-th channel for different environmental parameters; ΔT represents the higher-order compensation coefficient of the i-th channel for the interaction of environmental parameters; ΔT, ΔH, ΔP, ΔQ... represent the difference or product of the current environmental parameter and the reference environmental parameter; This represents the difference between the current environmental parameter value and the reference environmental parameter value; the coefficient φ is weighted by a sliding window or exponentially and adaptively updated at a slowly varying rate to avoid following instantaneous noise; for Perform normalization and denoising to output a multi-channel sequence. Provided for early shutdown / TCN and subsequent modules to share.

[0096] like Figure 4 As shown, step S6: baseline tracking, anchor point calibration, and channel health status assessment (periodic / conditional triggering).

[0097] Step S6.1: Baseline Tracking: Measure and acquire the latest baseline during the interval between the end of post-washing or between two detections. ,noise and time constant ; Calculate drift index ,in, , , The reference benchmark value is used; if any indicator exceeds the adaptive threshold band (determined based on the quantiles of the historical distribution, preferably in the 90–99% (target quantile) range), it is marked as "deviation".

[0098] Step S6.2: Anchor point calibration: Triggered by a fixed period or the "deviation" condition generated in S6.1, switch to clean gas for short sampling to refresh. In conjunction with the reference environment, and simultaneously fine-tuning the S5 coefficient.

[0099] Step S6.3: Health Score and Gating: Define the health score of channel i at time t. for:

[0100]

[0101] in, This is a normalization function that maps the deviation to the interval [0,1]. Score the cross-correlation consistency between this channel and the other channels; Configurable weights; applied before feature fusion / model input. Channels are weighted; channels whose health scores are continuously below a preset threshold are placed into candidate blocking status and the event is reported.

[0102] like Figure 4 As shown, step S7: perform threshold adaptation, uncertainty control and unknown class (OOD) identification.

[0103] Step S7.1: Confidence calibration: Perform temperature calibration or minimize expected calibration error (ECE / MMCE) calibration on the model output (SPRT / TCN) to make the confidence interpretable.

[0104] Step S7.2: Conformal Dynamic Threshold: Calculate the non-conformance score within the sliding window W. This score can be defined in various ways, including but not limited to: in classification tasks, it can be defined as... The maximum class probability is a measure of the uncertainty of the prediction result; in regression or continuous value prediction tasks, it can be defined as the prediction residual, such as the difference between the predicted value and the reference value.

[0105] The system selects an appropriate definition of non-compliance score based on the detection task and calculates its distribution quantile within window W. When the non-compliance score of the current sample is greater than or equal to When the threshold is set, the system determines the data as unreliable or requires rejection, thus achieving dynamic threshold discrimination with controllable coverage.

[0106] Step S7.3: OOD Recognition: Maintaining various centers in the intermediate feature space With covariance Calculate Mahalanobis distance Or energy score, where z is the feature representation of the sample to be identified in the intermediate layer of the model; k is the class index. If the value exceeds the historical quantile threshold, it is classified as OOD. OOD samples directly follow the "full cycle → resampling" path and do not enter incremental learning.

[0107] like Figure 4 As shown, step S7.4: Resampling control and fallback.

[0108] Triggering conditions: A series of events such as rejection / OOD / abnormal health score occur, or key indicators cross the adaptive threshold (historical percentile setting).

[0109] Resampling strategy: After entering the resampling phase, the sampling time can be appropriately extended or the pump flow rate adjusted within the safety / compliance range; a reassessment is performed after the complete cycle is completed.

[0110] Collaborative control: During the resampling phase, issue an instruction to the early stop unit to "prohibit early stop / increase the threshold" to prioritize ensuring the reliability of the conclusion; if necessary, it is recommended to extend the post-cleaning or increase the intensity of high-temperature cleaning.

[0111] like Figure 5 As shown, step S8: Perform constrained incremental learning and rollback (offline / quasi-online) on the edge AI.

[0112] Step S8.1: Sample selection: Only samples with "high confidence + passing S1–S8 quality control" are included in the candidate set; replay caching is used to maintain class balance and old knowledge.

[0113] Step S8.2: Lightweight Update: Preferably, only fine-tune the last-level classifier head / normalized statistics; if full network fine-tuning is required, introduce Elastic Weight Fixation (EWC) constraints, and adjust the loss function. for

[0114]

[0115] Among them, L CE That is, cross-entropy loss; λ is a hyperparameter; θ j This represents the value of the j-th parameter in the current model; This represents the original value of the j-th parameter after training on the old task is completed.

[0116] Step S8.3: Local Validation and Rollback: After incremental updates, monitor metrics such as accuracy, ECE, and average early stop time on the local validation set; when ΔAcc is below the threshold (e.g. If the rejection rate rises above a threshold (e.g., >10%), rollback will be initiated. And clear the recently updated sample pool.

[0117] Step S8.4: Versioning and Auditing: Version the model and calibration configuration (v1, v2, etc.), record timestamps, trigger reasons and metric changes and store them in encrypted form.

[0118] Data path (example): ADS1256 → Feature extraction → AI model inference → Output prediction and confidence; Early stop is triggered when the confidence is ≥ the threshold (e.g., 0.90), otherwise sampling continues.

[0119] like Figure 3 As shown, step S9: Implement closed-loop coordinated control of dynamic temperature control and early stop strategy.

[0120] Synergy in the time dimension: Dynamic temperature control is used to shorten the preheating and cleaning period, while early shutdown is used to shorten the adsorption and detection period; the two complement each other to achieve full-cycle acceleration.

[0121] Triggering coordination: When the early stop output is ahead of schedule, the dynamic temperature control immediately switches to the high-temperature desorption stage; if the early stop does not meet the standard, the complete temperature control curve is maintained.

[0122] Safety Collaboration: When the health score is low or rejections / OODs occur frequently, the early stop threshold is automatically raised or early stop is temporarily disabled; when the environment and health return to stability, the default threshold is restored.

[0123] Step S10: Perform system structure optimization and power consumption management.

[0124] Flow splitting and uniformity: The intake end is divided into eight streams. CFD optimization of the splitting angle and pipe resistance ensures that the flow difference is preferably no more than 5%, thereby improving the synchronicity of multi-channel response and enhancing the availability of early characteristics.

[0125] Low adsorption and small dead space: the valve is close to the gas chamber inlet to shorten the pipeline; the inner wall of the gas path is coated with a low adsorption coating; the dead space volume is preferably no more than 0.5 mL.

[0126] Modular and miniaturized: Pumps, valves, gas chambers, arrays and batteries are integrated into one package; modular arrays facilitate replacement or upgrades.

[0127] PCB three-domain isolation: physical partitioning of analog acquisition, digital AI and power drive to improve anti-interference and signal-to-noise ratio.

[0128] Power consumption and safety: Battery management monitors power consumption and supports low-power mode; heat insulation is implemented in high-temperature areas; the overall size and weight of the device are suitable for handheld use; landscape UI supports one-handed operation.

[0129] Step S11: Adjust the parameter range and working mode according to the operating status.

[0130] Temperature control platform: The low temperature detection platform and the high temperature desorption platform adopt the "range + optimization + safety upper limit" setting (for example, the low temperature is 200-300℃, and the high temperature is 350-500℃ without damaging the device).

[0131] Early stop exit: Multiple candidate points are set in the 1–6s range (preferably 2–5s), and the settings are adaptive according to the scenario.

[0132] Confidence threshold (optional mode): The threshold can be set to 0.75 in the emergency rapid mode and 0.90 in the medical screening mode.

[0133] Conformal window and confidence level: window W is on the order of tens to hundreds of frames, and ε is on the order of percentage points; typical settings are given in the examples.

[0134] Health score weighting: ∈[0,1] and has been verified.

[0135] Anchor point cycle: A dual strategy of "fixed interval + event triggering" is adopted.

[0136] Dynamic cleaning strategy: dynamically match temperature (300–500℃) and duration (10–20s) according to the contamination index R.

[0137] Resampling: Within safe / compliant limits, moderately extend the sampling time or adjust the pump flow rate to enhance the characteristics.

[0138] Example 1: Rapid Screening System for Fruit Fly Infection in Imported Fruits at Ports of Entry

[0139] 1. Application Scenarios

[0140] This embodiment is applied at customs entry ports for rapid on-site screening of fruit fly pests (Bactrocera dorsalis, Bactrocera correcta) in tropical fruits (mango, wax apple, guava).

[0141] like Figure 9 As shown, in a batch quarantine scenario, the electronic nose device 901 can work in conjunction with the existing conveyor belt 903 and robotic arm 902: fruit sample 904 enters the sampling position of the electronic nose device 901 along with the conveyor belt 903. After the system completes rapid identification, it transmits the results to the robotic arm 902, which sorts the sample 904 into different sample boxes 905, realizing a closed-loop process of "automatic conveying - rapid detection - instant sorting".

[0142] 2. System Configuration

[0143] Sensor array: 8-channel MOS gas sensor (preferably TGS2602 or TGS2620);

[0144] Dynamic temperature control unit: The detection range is preferably 250±10℃, and 450℃ pulse heating is used during the cleaning stage, with a duty cycle of about 1:3;

[0145] Edge AI Model: Based on a multi-exit TCN structure, with exit points set at three key feature points:

[0146] Feature point 1: The point of maximum rate of change, Max(ds / dt), i.e., the inflection point where the signal rises rapidly, such as... Figure 6 As shown;

[0147] Feature point two: Response peak point, i.e., the point where the signal amplitude is the largest;

[0148] Feature point 3: The minimum rate of change point Min(ds / dt), that is, the point of gradual change in the descent phase;

[0149] Drift compensation module: Based on dynamic coefficient updates using Kalman filtering, it corrects the effects of temperature and humidity fluctuations in real time.

[0150] Automated collaboration: The electronic nose device 901 outputs the detection results through the edge AI processing module 108, communicates with the robotic arm 902 and the conveyor belt 903, and drives the robotic arm 902 to sort the samples 904 into different sample boxes 905.

[0151] 3. Testing Process

[0152] Step S1: Initialization;

[0153] Rapid preheating: After the device is powered on, the sensor is quickly heated to the detection temperature zone (approximately 250±10℃) using a pulse heating method (preferably 1–3 seconds), thus shortening the preheating time;

[0154] Baseline calibration: After the sensor reaches the detection temperature range, air is introduced as a cleaning gas, the initial baseline signal is collected, and a baseline profile (including noise level and dynamic response time constant) is established.

[0155] Step S2: Sample detection and early cessation detection;

[0156] Fruit sample 904 enters the sampling position of electronic nose device 901 via conveyor belt 903; the air pump of electronic nose device 901 operates in pulse mode (preferably 2000 mL / min), and the signal is input to AI model in real time; the system outputs the discrimination result before feature point one (about 1–2 s), feature point two (about 3–5 s), or feature point three / complete cycle; the discrimination result is transmitted to robotic arm 902, and robotic arm 902 performs sorting according to the result.

[0157] Step S3: Cleaning and compensation;

[0158] After the judgment is completed, immediately switch to high-temperature pulse cleaning (preferably 450℃), and simultaneously introduce air to flush the sensor cavity, remove residual gas and restore the baseline. The drift compensation module updates the parameters synchronously.

[0159] Step S4: Sorting collaboration, such as... Figure 9 As shown,

[0160] Normal result: Sample 904 continues to flow with conveyor belt 903, or is placed into qualified sample box 905 by robotic arm 902; Suspected result: Robotic arm 902 removes sample 904 to isolated sample box 905; Uncertain result: The system triggers a fallback, and sample 904 enters sample box 905 to be reviewed.

[0161] (1) Experimental verification data

[0162] like Figure 7 As shown, Bactrocera dorsalis achieved the following accuracy rates: 95.5% for regular decision points, 90.9% for earliest decision points, and a 3 / 4 reduction in average advance decision time.

[0163] like Figure 7 As shown, Bactrocera Correcta's accuracy at regular decision points is 93.3%; accuracy at earliest decision points is 88.9%; and the average advance decision time is reduced by approximately 1 / 2–3 / 4 of the cycle.

[0164] The bar chart results show that the average discrimination time in the control group was 4.5s, while in the infection group it was extended to 7–9.6s in the early, middle and late stages, respectively. Our system can trigger discrimination in 1–2 seconds.

[0165] (2) Schematic diagram of rapid detection and collaboration

[0166] like Figure 8 As shown, the period shortening ratios corresponding to different discrimination points are: feature point one is about 1 / 4 of the original, feature point two is about 1 / 2, and feature point three is about 3 / 4.

[0167] like Figure 9 As shown, the electronic nose device 901 works in conjunction with the conveyor belt 903, the robotic arm 902, and the sample box 905 to achieve rhythmic detection and automatic sorting of batch fruit samples 904.

[0168] 4. Technical Effects

[0169] Accelerated full-cycle testing: The average detection time has been reduced from the traditional 182 seconds to less than 38 seconds;

[0170] Automated processing: Through the collaboration of conveyor belt 903 and robotic arm 902, the automatic conveying, detection and sorting of samples 904 are realized;

[0171] Reliability assurance: The accuracy rate of the earliest decision point remains above 88%, combined with a complete cycle fallback mechanism and the 905 sample box to be verified, to ensure safety;

[0172] Application flexibility: Sample 904 is not limited to apples and can be a variety of fruits; the number of sample boxes 905 can also be configured according to on-site needs and is not limited by the number of schematic diagrams.

[0173] Example 2: Medical Breath Analysis System

[0174] 1. Application Scenarios

[0175] This embodiment is applied in a medical setting to detect key biomarkers in human exhaled breath, including acetone (aided diagnosis of diabetes), hydrogen sulfide (oral / liver function monitoring), and methane (gut microbiota status). Through individualized incremental learning, multimodal environment compensation, and three-level safety protection, this system achieves high reliability, speed, and clinical adaptability.

[0176] 2. System Configuration

[0177] (1) Sensor array

[0178] The preferred filters are TGS822 (acetone), TGS825 (hydrogen sulfide), and TGS2600 (methane), with an optional version, TGS2611-E00, featuring an anti-alcohol interference filter to reduce ethanol interference with breathalyzer testing.

[0179] (2) Dynamic temperature control unit

[0180] Detection range: 200–230℃, to avoid high-temperature decomposition of biogas;

[0181] Rapid preheating: The temperature rises to the target temperature zone within 2 seconds, ensuring that the patient can be tested after a short exhalation;

[0182] Cleaning temperature: 360–400℃. The cleaning time should be adjusted adaptively according to the frequency of use (e.g., 360℃ / 12 seconds for ≤5 times / day, 400℃ / 18 seconds for >10 times / day) to extend the sensor life.

[0183] (3) AI discrimination and learning module

[0184] Model structure: Multi-exit TCN, with exit points set at 2–4 seconds;

[0185] Incremental learning mechanism: Healthy individuals: only fine-tuning of the classifier head parameters (<1% of parameters); Chronic disease patients: fine-tuning of the classifier head and BN layer (<5% of parameters); Abnormal samples: triggering golden sample annotation, with batch updates concentrated during idle periods. Patient profile binding mechanism: generating a patient ID upon initial testing, associating it with age, BMI, and underlying disease information, and linking it with the incremental learning module to form an individualized adaptation model.

[0186] (4) Multimodal environment compensation module

[0187] Temperature compensation: Corrects the sensor signal based on the patient's sublingual temperature. ;

[0188] Breathing pattern classifier: Steady breathing: Uses conventional sampling; Shallow breathing or wheezing: Initiates segmented integration and extends sampling to 8 seconds; Cough interruption: Automatically discards the cycle and prompts for resampling.

[0189] (5) Three-level security protection system

[0190] Primary protection: The confidence threshold is dynamically adjusted according to the usage pattern (0.75 for emergency mode and 0.90 for screening mode).

[0191] Secondary arbitration: When the confidence level is between 0.80 and 0.89, a 3-second pulse retest is automatically triggered, and the result is output based on the principle of 2 / 3 consistency among the three tests;

[0192] Ultimate protection: Establish a pathological anomaly code library. When the result matches the abnormal feature spectrum, the system forces a manual review process.

[0193] 3. Testing Process

[0194] Step S1: Initialization;

[0195] The system quickly preheats to 200–230°C, introduces air to refresh the baseline, and uses patient profile parameters (such as body temperature and underlying disease information) to perform individualized baseline correction.

[0196] Step S2: Sampling and detection;

[0197] Subjects exhaled through a disposable mouthpiece, with the pump flow rate controlled at 100–150 mL / min. Sensors collected signals and input them into the AI ​​model in real time.

[0198] Step S3: Early cessation and rejection;

[0199] If the model's confidence level in the exit point at 2–4 seconds is greater than or equal to the threshold, the result will be output early; if the confidence level is in the middle range, a retest will be triggered; if the confidence level is insufficient, the full detection cycle (10–12 seconds) will begin.

[0200] Step S4: Cleaning and restoration;

[0201] After the test is completed, select the 360–400℃ pulse cleaning mode according to the frequency of use, introduce air to remove residual gas and restore the baseline.

[0202] 4. Technical Effects

[0203] This embodiment achieves the following through the above technical solution:

[0204] Personalized adaptation: Based on patient profiles and hierarchical incremental learning strategies, the detection model can be individually adjusted within 24 hours, significantly improving the adaptability of breath analysis for patients with chronic diseases;

[0205] Multimodal compensation: Combining body temperature and respiratory pattern correction, it avoids interference with results caused by shallow breathing, wheezing, or coughing, thereby improving the success rate of detection;

[0206] Level 3 security protection: Dynamic threshold, retest arbitration and abnormal feature database triple mechanism to reduce the risk of misjudgment and ensure clinical safety in medical scenarios;

[0207] Hardware lifespan optimization: Through anti-alcohol interference filter and graded thermal management strategy, the sensor lifespan is increased from 1,800 cycles to more than 3,500 cycles, reducing equipment maintenance costs.

[0208] Example 3: Environmental Emergency Detection System

[0209] 1. Application Scenarios

[0210] This embodiment is used for public safety and emergency monitoring, including enclosed or semi-enclosed environments such as subway stations, factories, warehouses, tunnels, and chemical industrial parks, for monitoring carbon monoxide (CO), benzene compounds (such as benzene, toluene, and xylene), and ammonia (…). It can quickly screen and classify dangerous gases such as ) and provide early warning, and is suitable for scenarios such as sudden leaks, dangerous operation inspections and fire rescue.

[0211] 2. System Configuration

[0212] (1) Sensor array

[0213] CO / Channels: TGS2610 and TGS2611 are preferred, with the CO channel having an added molecular sieve membrane to improve selectivity for ethanol / formaldehyde;

[0214] Benzene series channel: PID (photoionization) sensor is preferred to achieve ppb-level detection capability;

[0215] Or other toxic gas channels: can be equipped with MOS or electrochemical channels;

[0216] Channel-specific maintenance: MOS channels support high-temperature cleaning; PID channels perform light source calibration and zero-point calibration, and do not participate in high-temperature cleaning.

[0217] (2) Dynamic temperature control and cleaning life management

[0218] MOS detection range: 280–320℃; optimal preheating time is 1–2 seconds;

[0219] Adaptive pulse thermal cleaning protocol:

[0220] Regular cleaning: 480℃, approximately 15 seconds, continuous mode;

[0221] Heavy pollution: 500℃, approximately 8 seconds, pulse mode (preferably 5Hz);

[0222] Routine maintenance: 450℃, approximately 20 seconds, gradual temperature increase;

[0223] Health status monitoring: Calculates the quality factor Q in real time, and triggers a maintenance alarm when Q continuously decreases by more than 30%.

[0224] (3) Extreme Environment Adaptation Module

[0225] Hardware protection:

[0226] The air chamber is equipped with a PTC heating element (maintaining 60±5℃) to prevent condensation;

[0227] The pipeline is coated with a nano-hydrophobic coating (contact angle preferably >150°) to reduce moisture adsorption;

[0228] Environmental sensing: Integrated temperature and humidity sensor, with a sampling rate preferably ≥10Hz;

[0229] Algorithm compensation:

[0230] High humidity conditions (RH>80%): eigenvector × [1–α·(RH–80)], where α is the gradual change compensation coefficient;

[0231] Low temperature conditions (T<10℃): Start the preheating compensation mode, heat to 50±5℃ and keep constant temperature before proceeding with routine sampling.

[0232] (4) Anti-crossing and interference module

[0233] Hardware suppression: CO channel molecular sieve filter membrane; benzene series channel uses PID instead of MOS;

[0234] Algorithm subtraction: Construct a feature library of interfering substances (ethanol, formaldehyde, acetone, etc.), and subtract them according to the cross-sensitivity coefficient k. i Real-time correction

[0235]

[0236] Among which, C i is the estimated interference concentration, and k i is the experimentally calibrated value.

[0237] (5)Collaboration between AI model and hierarchical early warning

[0238] Discrimination model: SPRT + multi-output TCN, and the early-stop output is preferably selected within 1 - 3 seconds;

[0239] Hierarchical early warning mechanism:

[0240] Level III (warning): TLV < concentration < STEL → the device flashes a prompt;

[0241] Level II (medium risk): STEL ≤ concentration < IDLH → local audible and visual alarm + cloud push;

[0242] Level I (high risk): concentration ≥ IDLH → strong audible and visual alarm + linkage for exhaust or power-off;

[0243] Networking collaboration: Automatically generate a pollution diffusion model based on the positioning information and send it to adjacent devices to enter the "high-sensitivity mode".

[0244] 3. Detection process

[0245] Step S1: Initialization and health check;

[0246] The MOS channel is quickly preheated to 280 - 320 °C; the PID channel performs light source stabilization and zero calibration; air is introduced to refresh the baseline, and the temperature and humidity are collected and initial compensation is performed.

[0247] Step S2: High-flow sampling;

[0248] Start the high-flow pump (preferably 300 mL / min), the gas enters the gas chamber, the PTC maintains 60 ± 5 °C, and the hydrophobic coating reduces the condensation residue.

[0249] Step S3: Anti-interference deduction and early-stop discrimination;

[0250] The system performs real-time correction according to the interference library and sends the net response to SPRT / TCN:

[0251] If the confidence threshold is reached at the output within 1 - 3 seconds, a hierarchical early warning is issued in advance;

[0252] Otherwise, enter the complete cycle (10 - 15 seconds) to complete the discrimination.

[0253] Step S4: Hierarchical early warning and collaboration n;

[0254] The system outputs warning levels based on TLV / STEL / IDLH rules and can be linked to ventilation / power outages; at the same time, it uploads data to the cloud to drive nearby devices to enter high-sensitivity monitoring.

[0255] Step S5: Cleaning and Lifespan Management;

[0256] The MOS channel performs adaptive pulse thermal cleaning; the PID channel performs zero-point calibration; the Q value is updated and written to the file, and a maintenance alarm is triggered if the Q value drops beyond the threshold.

[0257] 4. Technical Effects

[0258] Anti-interference capability: The hardware filter membrane, PID control, and interference subtraction algorithm work together to significantly reduce false alarms in the presence of ethanol and formaldehyde.

[0259] Extreme environment adaptability: PTC heating, hydrophobic coating and high humidity / low temperature compensation algorithm ensure that the system can still work stably in high humidity and low temperature environments.

[0260] Rapid early warning capability: Early stop exits can trigger an early warning within 1–3 seconds, meeting the real-time requirements of public safety.

[0261] Tiered response and coordination: Based on a three-tiered early warning mechanism of TLV / STEL / IDLH, and supporting multi-device collaboration, it enables spread tracking and emergency response.

[0262] Extended lifespan and convenient maintenance: Through pulsed thermal cleaning and Q-value monitoring, the lifespan of the MOS channel is increased to approximately 5,000 cycles, and maintenance prompts are provided.

[0263] Those skilled in the art will understand that the features described in the various embodiments of this disclosure can be combined and / or combined in various ways, even if such combinations and / or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments of this disclosure can be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0264] Although this disclosure has been shown and described with reference to specific exemplary embodiments thereof, those skilled in the art will understand that various changes in form and detail may be made to this disclosure without departing from the spirit and scope of the disclosure as defined by the appended claims and their equivalents. Therefore, the scope of this disclosure should not be limited to the above embodiments, but should be defined not only by the appended claims, but also by their equivalents.

Claims

1. A handheld AI electronic nose device, characterized in that, include: The electronic nose device (901) includes an environmental parameter monitoring unit (101), a multi-channel metal oxide gas sensor array unit (102), an integrated gas path-gas chamber assembly (103), an integrated pump and valve module (104), a display and input module (105), an acquisition and execution control board (106), a high-resolution analog-to-digital conversion module (107), an edge AI processing module (108), and a battery and management module (109). The environmental parameter monitoring unit (101) is located in the central mounting position of the multi-channel metal oxide gas sensor array unit (102). The multi-channel metal oxide gas sensor array unit (102) is installed inside the gas path-gas chamber integrated assembly (103). The integrated pump valve module (104) is connected to the gas path-gas chamber integrated assembly (103). The acquisition and execution control board (106) is connected to the environmental parameter monitoring unit (101) and the multi-channel metal oxide gas sensor array unit (102). The high-resolution analog-to-digital conversion module (107) is located above the acquisition... Connected to the execution control board (106), the analog data collected by the acquisition and execution control board (106) is converted into digital data, and the converted digital data is transmitted to the edge AI processing module (108) connected below. The edge AI processing module (108) connects to and sends control commands back to the acquisition and execution control board (106). The battery and management module (109) provides energy and safety management for the electronic nose device (901). The display and input module (105) displays the mode, temperature control curve, progress, results and prompts of the electronic nose device (901).

2. The handheld AI electronic nose device according to claim 1, characterized in that, The environmental parameter monitoring unit (101) is used to collect environmental parameters as inputs for online compensation and self-calibration. The environmental parameters include temperature, humidity, flow rate and air pressure. Each channel of the multi-channel metal oxide gas sensor array unit (102) has independent heating and independent temperature measurement, which is used to perform differentiated dynamic temperature control; The integrated gas path-gas chamber assembly (103) includes an air inlet (201), a central channel, at least one sampling microcavity (202), and an air outlet (206). The sampling microcavity (202) is provided with a sampling microcavity inlet (203) and a sampling microcavity outlet (204). A sensor mounting interface (205) is provided on the lower inner side of the sampling microcavity (202). A sensitive layer of a multi-channel metal oxide gas sensor array unit (102) is provided at the sensor mounting interface (205) to ensure that the gas acts on the sensitive layer in the sampling microcavity (202). The integrated pump and valve module (104) is connected to the integrated gas path-gas chamber assembly (103) and supports constant flow sampling, pulse sampling and cleaning modes; The acquisition and execution control board (106) is responsible for the initial acquisition of sensor signals, the execution of array heating curves and the control of pump valve drive, and is synchronized with the high-resolution analog-to-digital conversion module (107); The high-resolution analog-to-digital converter module (107) is used to perform high-precision multi-channel synchronous sampling of the output of each channel of the multi-channel metal oxide gas sensor array unit (102) to ensure the quality of weak signals in the early stage. The edge AI processing module (108) is used to perform model inference, confidence calibration and constrained incremental learning locally, and to perform versioning and rollback. The battery and management module (109) is used to provide power, charge / discharge and safety management, and supports low-power sleep mode; The display and input module (105) is used for mode selection and result display.

3. A handheld AI electronic nose system, characterized in that, The handheld AI electronic nose device, as described in any one of claims 1 to 2, is divided into a perception layer, a decision-making layer, and a control layer, which interact through a unified message / event bus.

4. The handheld AI electronic nose system according to claim 3, characterized in that, The sensing layer consists of the environmental parameter monitoring unit (101), the multi-channel metal oxide gas sensor array unit (102), the gas path-gas chamber integrated component (103), the integrated pump valve module (104), and the high-resolution analog-to-digital conversion module (107), and is used to complete gas sampling and signal acquisition. Gas introduction and distribution: External gas enters through the integrated pump and valve module (104), wherein the gas pump and the gas valve are integrated into a compact structure. The gas pump is located on the left side of the integrated pump and valve module (104), and the gas valve is located on the right side of the integrated pump and valve module (104). The gas pump provides the driving force for gas flow, and the gas valve plays the role of switching and flow path switching. Under the action of the gas pump, the external gas enters the inlet (201) of the gas path-gas chamber integrated component (103) through the gas valve, and then enters the sampling microcavity (202) through the central channel and through the inlet (203) of each sampling microcavity. In the sampling microcavity (202), it fully interacts with the sensitive layer of the multi-channel metal oxide gas sensor array unit (102) located at the sensor mounting interface (205), and flows into the central channel from the sampling microcavity outlet (204) and is discharged through the outlet (206). Signal and environmental quantity acquisition: The high-resolution analog-to-digital conversion module (107) performs high-precision synchronous sampling of the multi-channel sensor signals of the multi-channel metal oxide gas sensor array unit (102); The environmental parameter monitoring unit (101) outputs environmental parameters including temperature, humidity, flow rate and air pressure in real time for subsequent online compensation and self-calibration; The perception layer outputs the generated multi-channel sensor signals and environmental parameters to the decision layer.

5. The handheld AI electronic nose system according to claim 3, characterized in that, The decision layer consists of a dynamic temperature control and early stop discrimination unit, an anti-drift and self-calibration unit, and an edge AI and local incremental learning unit, and operates with the support of temperature control / execution of the acquisition and execution control board (106) and data from the high-resolution analog-to-digital conversion module (107). The dynamic temperature control and early stop discrimination unit executes pulse rapid preheating, low temperature adsorption detection, two-stage early stop sequential probability ratio test and multi-outlet time convolution network and high temperature desorption cleaning process in sequence according to the temperature control curve issued by the acquisition and execution control board (106). When the threshold is met, a judgment conclusion and confidence level are generated, and a trigger signal for initiating cleaning is sent to the control layer; The anti-drift and self-calibration unit takes the environmental parameters of the environmental parameter monitoring unit (101) and the historical baseline of the multi-channel metal oxide gas sensor array unit (102) / high-resolution analog-to-digital conversion module (107) as inputs, performs online environmental compensation, anchor baseline calibration, channel health scoring and gating / masking, confidence calibration, conformal rejection and outlier distribution identification, and outputs parameter update and resampling instructions. The edge AI and local incremental learning unit run on the edge AI processing module (108) to perform model inference, sample screening, constrained incremental updates and version rollback on the preprocessed multi-channel data, and generate model / log versions.

6. The handheld AI electronic nose system according to claim 3, characterized in that, The control layer consists of an acquisition and execution control board (106), a high-resolution analog-to-digital conversion module (107), an edge AI processing module (108), a battery and management module (109), and a display and input module (105), and is responsible for execution, acquisition, energy consumption, and interaction. Execution and Acquisition: The acquisition and execution control board (106) controls the heating and temperature measurement of the integrated pump valve module (104) and the multi-channel metal oxide gas sensor array unit (102), and works in conjunction with the high-resolution analog-to-digital conversion module (107) to achieve multi-channel synchronous sampling and preprocessing; Local computation: The edge AI processing module (108) completes inference, confidence calibration and incremental learning, and sends the control commands back to the acquisition and execution control board (106). Power supply and interaction: The battery and management module (109) provides energy and safety management; the display and input module (105) displays the mode, temperature control curve, progress, results and prompts; Data and control path: The target link is a high-resolution analog-to-digital conversion module (107) connected to the acquisition and execution control board (106), the acquisition and execution control board (106) connected to the edge AI processing module (108), and the edge AI processing module (108) connected to the display and input module (105); the control loop is an edge AI processing module (108) connected to the acquisition and execution control board (106), used to trigger pump valve / temperature control actions.

7. A handheld AI electronic nose method, characterized in that, The method comprising the handheld AI electronic nose system according to any one of claims 3 to 6, wherein the method includes: Step S1: Perform the initialization operation of the system and establish a baseline profile; Step S2: Perform sampling start point detection and data preprocessing; Step S3: Perform a pulsed rapid preheating process in the first stage; Step S4: In the second stage, perform low-temperature adsorption detection and execute multi-level early stop judgment; Step S5: Implement online environmental compensation and data normalization processing; Step S6: Perform baseline tracking, anchor point calibration, and channel health status assessment; Step S7: Perform threshold adaptive adjustment, uncertainty control, and unknown class identification; Step S8: Perform constrained incremental learning and rollback operations at the edge AI. Step S9: Achieve closed-loop coordinated control of dynamic temperature control and early stop strategy; Step S10: Perform system structure optimization and power consumption management; Step S11: Adjust the parameter range and working mode according to the operating status.

8. The method according to claim 7, characterized in that, The initialization operation of the system and the establishment of a baseline profile include: Step S1.1: Clean Gas Baseline Acquisition: After the device is initially deployed or powered on, the clean gas is automatically switched, and multi-channel signals are acquired for a preset time period to obtain the initial baseline for each channel. Noise flooring Dynamic range scale With rise time constant And simultaneously record the reference environment vector ; Step S1.2: Initial value fitting for environmental compensation: Using the reference environmental vector e0 as a reference, fit the environmental sensitivity coefficients of each channel; Step S1.3: Version management of calibration files: Generate calibration files with timestamps and store them in the database.

9. The method according to claim 7, characterized in that, The second stage involves low-temperature adsorption detection and multi-level early cessation discrimination, including: Step S4.1: Low-temperature adsorption: The response signal is collected on a low-temperature platform, and features including early slope, amplitude and cross-channel ratio are extracted from the response signal to improve information density; Step S4.2: Perform statistical early stopping detection based on sequential probability ratio test: calculate log-likelihood ratio in real time. Defined as: in, This represents the alternative assumption that the target gas exists. This indicates the null hypothesis that the target gas does not exist. This represents the k-th observation. when Accept in advance when ≥logA ,when Accept when ≤logB ;when When in the hysteresis interval, stabilization is achieved by combining sliding window consistency and double threshold hysteresis. Step S4.3: Multi-exit deep early stopping: Based on a temporal convolutional network, multiple exit points are set within a preset response time interval. When the discrimination confidence of any exit reaches a threshold, the exit is output in advance; the position of the exit is adjusted according to the scenario. Step S4.4: Perform fallback complete acquisition: If any of the above early stop is not triggered or the confidence level does not meet the preset conditions, continue to acquire the response signal until the predefined acquisition period.

10. The method according to claim 7, characterized in that, The execution of baseline tracking, anchor point calibration, and channel health status assessment includes: Step S6.1: Baseline Tracking: Measure and acquire the latest baseline during the interval between the end of post-washing or between two detections. ,noise and time constant ; Calculate drift index ,in, , , The reference benchmark value is used; if any indicator exceeds the adaptive threshold band, the adaptive threshold band is determined based on the quantiles of the historical distribution, the target quantile interval is selected, and it is marked as "deviation" state; Step S6.2: Anchor point calibration: Triggered by a fixed period or the "deviation" condition generated in S6.1, switch to clean gas for short sampling to refresh the reference value. With reference environment; and simultaneously fine-tuning S5 coefficient; Step S6.3: Health Score and Gating: Define the health score of channel i at time t. for: in, This is a normalization function that maps the deviation to the interval [0,1]. Score the cross-correlation consistency between this channel and the other channels; Configurable weights; applied before feature fusion / model input. Channels are weighted; channels whose health scores are continuously below a preset threshold are placed into candidate blocking status and the event is reported.

Images