Simulation intelligent dynamic dilution sniffing method and system

Abstract

The invention relates to a simulation intelligent dynamic dilution sniffing method and system, and the method comprises the steps: mixing an odor sample with clean air in proportion through a dynamic dilution module, and achieving instantaneous uniform dilution through a Venturi tube; the diluted gas is introduced into a plurality of groups of gas sensor arrays in parallel so as to simulate a plurality of olfactory discriminators; after being preprocessed, a sensor signal is input into a special artificial neural network model for analysis; the model deeply fuses a standard sniffing judgment rule, and can output simulated sniffing results which accord with specifications ('correct ', 'unknown' and'wrong '); the system automatically judges whether standard termination conditions are met or not according to multiple groups of results, adjusts the dilution ratio or terminates the experiment according to the standard termination conditions, and finally automatically calculates the odor concentration. Automatic equipment is used for completely replacing manual sniffing identification, subjective differences are thoroughly eliminated, and the detection efficiency, repeatability and standardization level are greatly improved.

Description

Technical Field

[0001] This invention relates to the technical field of environmental monitoring, and in particular to a simulated intelligent dynamic dilution olfaction method and system. Background Technology

[0002] Odor pollution is a significant environmental problem, and its monitoring primarily relies on sensory analysis, i.e., manual olfaction. Currently, the "dynamic dilution olfaction method" is widely used as a standard method both internationally and domestically (e.g., Chinese Ministry of Ecology and Environment standards HJ1262-2022 and HJ1416-2025). The basic principle of this method is as follows: an odor sample is mixed with odorless clean air at a specific ratio using a dynamic dilution apparatus. Trained olfaction experts then smell the diluted gas to determine if an odor can be detected. By progressively changing the dilution factor, the odor concentration is calculated based on the collective judgment of the olfaction team.

[0003] Although this method is considered a benchmark, its inherent limitations severely restrict monitoring efficiency and data quality:

[0004] 1. High subjectivity: The test results depend entirely on the olfactory sensitivity of the olfactory examiner and are significantly affected by individual physiological conditions (such as colds, fatigue), emotions, experience, etc., resulting in poor repeatability and low comparability of test results for the same sample from different people or at different times.

[0005] 2. Inefficient: It requires organizing multiple (usually 6) qualified olfaction testers to conduct experiments simultaneously. The process is time-consuming, taking several hours from sample preparation, dilution, olfaction to result calculation, which cannot meet the monitoring needs of large-scale and rapid response.

[0006] 3. Standardization and scaling difficulties: The training, screening and maintenance of olfactory identifiers are costly, and the differences between personnel in different laboratories are difficult to eliminate, which makes the data uncertain when comparing data across regions and laboratories.

[0007] 4. Labor costs and health risks: Long-term exposure to odor samples with unknown components may pose potential health risks to odor assessors, and professional odor assessors are in short supply.

[0008] To overcome the limitations of manual olfaction, electronic nose technology has been explored for odor identification. However, existing electronic nose technologies mostly perform qualitative or semi-quantitative analysis directly on raw or simply pre-treated gases. Their operating modes (such as static headspace sampling and fixed concentration response) are fundamentally different from the standard "dynamic dilution olfaction method" in principle, process, and judgment logic. Existing electronic nose systems are generally not directly compatible with the dynamic dilution process, complex dilution gradient logic (environmental samples and pollution source samples have opposite directions), and threshold judgment rules based on group statistics specified in current standards. Summary of the Invention

[0009] This invention aims to address the shortcomings of existing technologies by providing a simulated intelligent dynamic dilution odor identification method and system. It uses multiple gas sensor arrays to physically simulate the "olfactory organs" of multiple odor detectors, and uses an intelligent analysis algorithm embedded with standard judgment rules to simulate the "brain judgment" of odor detectors and the group decision-making process. It also drives a high-precision dynamic dilution module to automatically execute the standard dilution process, which can eliminate the subjectivity of manual odor identification, improve detection accuracy and repeatability, realize the full automation of odor concentration detection, and is compatible with the structure and standard methods of existing dynamic dilution odor identification methods.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] A simulated intelligent dynamic dilution olfaction method includes the following steps:

[0012] S1. Automated dynamic dilution and sample preparation:

[0013] Using a dynamic dilution module, the odor sample to be tested is mixed with odorless clean air in real time at a preset, precisely controllable dilution ratio to prepare a test gas with a specific dilution ratio.

[0014] S2. Multi-channel sensor array response acquisition:

[0015] The test gas prepared in step S1 is introduced into several sets of gas sensor arrays in parallel, and each set of gas sensor arrays contains multiple heterogeneous gas sensors.

[0016] S3. Sensor signal preprocessing and feature extraction:

[0017] The baseline signal and response signal to the test gas of each sensor in each gas sensor array are collected, and calculations and preprocessing are performed to obtain the feature vector characterizing the intensity of gas stimulation.

[0018] S4. Intelligent olfaction and judgment based on artificial intelligence:

[0019] The multiple feature vectors obtained in step S3 are input into a pre-trained artificial neural network model for processing. The artificial neural network model simulates the judgment logic of multiple olfactory judges based on the input signal features and outputs the judgment result corresponding to each sensor array that conforms to the standard olfactory judgment procedure. The type of the judgment result is selected from one or more of "correct", "unknown" and "incorrect".

[0020] S5. Automated process control and concentration calculation:

[0021] Based on the multiple judgment results output in step S4, a judgment is made according to the preset experimental termination conditions that are consistent with the standard method for artificial dynamic dilution olfaction.

[0022] If the termination condition is not met, the dilution factor will be automatically adjusted, and the process will return to step S1 to start the next round of testing.

[0023] If the termination condition is met, the experiment will be terminated, and the odor concentration value will be automatically calculated and output based on the dilution factor of all detection rounds and the corresponding judgment results.

[0024] In step S1, real-time mixing is achieved through a Venturi mixer: odorless clean air is used as the power source to flow through the throat of the Venturi tube to generate a high-speed jet and negative pressure, which entrains the odor sample to be tested, and the two achieve instantaneous turbulent mixing in the Venturi tube.

[0025] In step S3, the calculation and preprocessing specifically include: using the sensor data processing module to remove noise, interference, drift, and interference information generated by random factors; combining hardware filtering and software filtering methods; adding an operational amplifier; and using a state equation-based method to estimate the true value of the signal in real time to process random noise and system errors. For each sensor, the normalized response change rate Δ% to the test gas is calculated: Δ% = (test response value - baseline value) / baseline value × 100%. The calculated multiple Δ% values ​​are normalized to form a feature vector.

[0026] In step S4, the artificial neural network model is a fully connected feedforward neural network, and the output layer is configured differently according to the sample type:

[0027] When detecting ambient air or fugitive emission samples, the output layer is set with three neurons, corresponding to three categories of judgment results: "correct", "unknown", and "incorrect".

[0028] When testing exhaust gas samples from stationary pollution sources, the output layer is equipped with two neurons, corresponding to the "correct" and "incorrect" judgment results respectively.

[0029] In step S4, the decision-making process of the artificial neural network model incorporates the judgment rules of the standard olfactory discrimination method:

[0030] For ambient air or fugitive emission samples, if the model determines it to be "correct", the corresponding sensor signal characteristics must meet the preset first rule: Δ% of any one sensor > first threshold, or Δ% of at least two sensors > second threshold; the first threshold is greater than the second threshold.

[0031] For stationary pollution source exhaust gas samples, if the model determines it to be "correct", the corresponding sensor signal characteristics must meet the preset second rule: Δ% of any sensor > the third threshold.

[0032] The first threshold is 50%, the second threshold is 20%, and the third threshold is 30%.

[0033] In step S5, the experimental termination conditions are consistent with the standard artificial dynamic dilution olfactory identification method:

[0034] For ambient air or fugitive emission samples, the termination condition is: the M value calculated based on the judgment results of all groups at the current dilution factor is less than or equal to 0.58;

[0035] For stationary pollution source exhaust gas samples, the termination condition is: two consecutive sets of judgment results are "error".

[0036] A simulated intelligent dynamic dilution olfaction system is provided to implement the aforementioned simulated intelligent dynamic dilution olfaction method, comprising:

[0037] The dynamic dilution and gas mixing unit is used to mix the odor sample to be tested with odorless clean air in a precise ratio and output the test gas with a specified dilution factor.

[0038] The multi-channel sensing and detection unit includes several parallel arrays of gas sensors for simultaneously acquiring response signals to the test gas.

[0039] The signal acquisition and preprocessing unit is connected to the multi-channel sensing and detection unit and is used to acquire, process, and extract feature vectors from sensor signals.

[0040] The intelligent analysis and decision-making unit has a built-in artificial neural network model, which is used to receive feature vectors, analyze them, and output judgment results that conform to the standard olfactory identification procedure.

[0041] The central control and computing unit connects the intelligent analysis and decision-making unit and the dynamic dilution and gas distribution unit. It is used to judge the experimental progress based on the judgment results, control the adjustment of the dilution ratio, and calculate the final odor concentration when the experiment ends.

[0042] The human-computer interaction unit is used to set experimental parameters, start / stop experiments, display the detection process and intermediate results in real time, and view and output the final detection report.

[0043] The dynamic dilution and gas distribution unit includes a venturi mixer and multiple mass flow controllers, which are used to precisely control the flow rates of the odor sample and the odorless clean air, respectively.

[0044] Each gas sensor array in the multi-channel sensing unit is equipped with an independent flow control component at its front end to ensure that the gas flow through each sensor array is constant and consistent.

[0045] The beneficial effects of this invention are: this invention can replace the traditional manual dynamic dilution odor identification process, and integrates high-precision gas sensing, dynamic dilution, intelligent signal processing and artificial intelligence decision-making technologies to achieve objective, rapid and automated measurement of odor concentration. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the steps of the simulated intelligent dynamic dilution olfaction method of the present invention;

[0047] Figure 2 This is a structural diagram of the simulated intelligent dynamic dilution olfaction system of the present invention;

[0048] Figure 3 This is a schematic diagram of the dynamic dilution and gas distribution unit in Specific Embodiment 1;

[0049] Figure 4 This is a schematic diagram of the multi-channel sensing and detection unit in specific embodiment 1;

[0050] The following will describe in detail, with reference to the accompanying drawings, embodiments of the present invention. Detailed Implementation

[0051] The principles and features of the present invention are described below with reference to the accompanying drawings. The embodiments given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the invention will become clearer from the following description. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the invention.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0054] A simulated intelligent dynamic dilution olfaction method, such as Figure 1 As shown, it includes the following steps:

[0055] S1. Automated dynamic dilution and sample preparation:

[0056] Using a dynamic dilution module, the odor sample to be tested is mixed with odorless clean air in real time, quickly and uniformly at a preset, precisely controllable dilution ratio to prepare a test gas with a specific dilution ratio.

[0057] The dynamic dilution module comprises multiple gas paths precisely regulated by a mass flow controller. One path introduces odorless clean air, while the others are connected in parallel and introduce the odor sample to be tested. Crucially, a Venturi mixer is used as the core mixing unit: clean air, acting as the driving force, flows through the throat of the Venturi tube, forming a high-speed jet that generates negative pressure to entrain the odor sample. Intense momentum and mass exchange occurs between the two within the throat and diffuser section, achieving instantaneous and thorough turbulent mixing. This method boasts extremely high mixing efficiency, no dead volume, and rapid response, effectively avoiding the adsorption, residue, and uneven mixing problems that may arise with traditional mixing chambers. The dilution factor is achieved by precisely controlling the flow ratio of the odor sample to the odorless clean air, covering a wide range from low to high dilution factors (e.g., 10 to 100,000 times) to meet the testing needs of different sample types (ambient air / fugitive emissions / stationary pollution source exhaust gas).

[0058] S2. Multi-channel sensor array response acquisition:

[0059] The test gas prepared in step S1, after being diluted to a specific ratio, is introduced in parallel into multiple gas sensor arrays. Each sensor array consists of multiple (preferably five) gas sensors that have specific or cross-sensitivity to different types of odorous substances. These sensors are specifically screened and combined to cover the typical odorous substance spectrum of the target industry (such as wastewater treatment, chemical industry, and waste treatment). To ensure the consistency of detection conditions, each sensor array is equipped with an independent miniature mass flow meter or flow controller at its front end to ensure that the gas flow rate through each sensor array is constant and equal.

[0060] S3. Sensor signal preprocessing and feature extraction:

[0061] The baseline signal (raw value) of each sensor in each sensor array is acquired when clean air is introduced, as well as the response signal (test value) when diluted sample gas is introduced. The raw signals are subjected to hardware filtering (such as using a high-precision operational amplifier circuit) and software filtering (such as state estimation algorithms based on Kalman filtering or moving average) to suppress noise, temperature drift, and random interference.

[0062] After temperature and humidity compensation preprocessing, the normalized response change rate Δ% for each sensor is calculated: Δ% = (test value - original value) / original value × 100%. For negative responses, Δ% is treated as 0. All Δ% values ​​are normalized to the interval [-1, 1] to form a feature vector characterizing the stimulation intensity of the gas composition on the sensor array at the current dilution factor.

[0063] S4. Intelligent olfaction and judgment based on artificial intelligence:

[0064] The multiple feature vectors obtained in step S3 are input into a pre-trained intelligent analysis system, which is essentially an artificial neural network model designed specifically for simulating olfactory discrimination tasks.

[0065] Network Architecture: A three-layer fully connected feedforward neural network is preferred. The number of neurons in the input layer equals the total number of sensors (number of groups × number of sensors per group). The feature abstraction layer (hidden layer) adopts an "overcomplete" design, with the number of neurons configured as (2m+1) × number of groups (m is the number of sensors per group, for example, 5, a single group corresponds to 11 computing units, in 4 / 6 group scenarios, the total number of computing units is 44 / 66 respectively), forming an overcomplete feature representation space, which can simultaneously capture the signal differences of 5 sensors within a single group and the cooperative response patterns of sensors between groups.

[0066] The hyperbolic tangent function (Tanh function) is used as the activation function to enhance the extraction of weak nonlinear features. This function maps the input signal to the [-1,1] interval, which can enhance the sensitivity to weak signal changes (such as the 20%-50% signal change rate in environmental samples) and prevent low-concentration odor signals from being masked by noise.

[0067] The feature learning mechanism achieves cross-channel signal coupling through a weight matrix, focusing on strengthening the weight ratio of "signal features that meet the judgment rules" (such as "single Δ%>50%" and "two Δ%>20%" in environmental samples, and "single Δ%>30%" in pollution source samples), thereby improving the recognition priority of target signals.

[0068] The output layer is dynamically configured according to the sample type: For ambient air / unorganized samples, three output neurons are set up, corresponding to three judgment results: "correct (√, odor can be detected)", "unknown (?, uncertain)" and "incorrect (×, odor cannot be detected)" (unique thermal encoding: [1,0,0] is correct, [0,1,0] is unknown, and [0,0,1] is incorrect); For stationary pollution source exhaust gas samples, two output neurons are set up, corresponding to "correct (√)" and "incorrect (×)" (unique thermal encoding: [1,0] is correct and [0,1] is incorrect).

[0069] Rule Embedding and Decision Logic: During training and inference, the network deeply integrates manually determined rules from standard methods. The network output is a probability distribution for each category (obtained through a Softmax normalized exponential function). The final decision is the category with the highest probability, but this result must satisfy the corresponding physical rule constraints:

[0070] Environmental sample rule constraints: If judged as "correct", the corresponding sensor signal must meet the following conditions: "at least one of the five sensors has Δ% > 50%" or "at least two have Δ% > 20%"; if judged as "unknown", only one sensor has Δ% between 20% and 50%; if judged as "incorrect", all sensors have Δ% ≤ 20%.

[0071] Pollution source sample rule constraints: If judged as "correct", it is required that "at least one sensor's Δ% > 30%" is met; if judged as "incorrect", it is required that "all sensors' Δ% ≤ 30%" is met.

[0072] This design ensures that the "black box" decision-making of artificial intelligence is closely integrated with explainable industry-standard logic.

[0073] S5. Automated process control and concentration calculation:

[0074] The multiple sets of "smell recognition results" output by the intelligent analysis system are sent to the automated control platform, which makes a judgment based on the experimental termination conditions specified by standard methods.

[0075] For environmental samples: Calculate the judgment results of all groups at the current dilution factor, calculate the individual olfactory threshold and the average olfactory threshold of the group according to the standard formula, and further calculate the M value. If the M value > 0.58, it is determined that the termination condition has not been met.

[0076] For pollution source samples: determine whether there are two consecutive sets (simulating two odor judges) and judge as "error".

[0077] If the termination condition is not met, the automated control platform will instruct the dynamic dilution module to automatically adjust to the next dilution level according to the preset logic (increase the dilution factor for environmental samples and decrease the dilution factor for pollution source samples), and jump to step S1 to start a new cycle.

[0078] If the termination conditions are met, the experiment is terminated. Based on the dilution factor at the time of termination and the judgment results of each group, the automated control platform automatically calculates the odor concentration (Y value) of the sample in strict accordance with the calculation formula specified in standard HJ1416-2025, and generates a complete and standardized test report.

[0079] A simulated intelligent dynamic dilution olfaction system, such as Figure 2 As shown, it includes:

[0080] Dynamic dilution and gas mixing unit: Used to mix the odorous gas sample to be tested with odorless clean air in a precise ratio, outputting test gas at a specified dilution factor. It includes multiple high-precision mass flow controllers and a Venturi mixer. The mass flow controllers control the flow rates of clean air and sample gas separately, while the Venturi mixer achieves instantaneous and uniform mixing of the gases, outputting test gas at the specified dilution factor.

[0081] Multi-channel sensing and detection unit: includes several groups (preferably four or six groups) of identical gas sensor arrays arranged in parallel, each group of gas sensor arrays containing multiple heterogeneous gas sensors; each group of arrays is equipped with an independent flow control component to ensure that the detection conditions of each channel are consistent.

[0082] Signal acquisition and preprocessing unit: includes a multi-channel high-precision analog-to-digital converter, signal conditioning circuit, and a microprocessor that runs filtering algorithms, used for real-time acquisition, filtering, and calculation of the normalized response change rate Δ of each sensor.

[0083] Intelligent analysis and decision-making unit: includes memory and processor. The memory stores the trained neural network model and the embedded standard judgment rules. The processor is used to execute the model, analyze the preprocessed sensor signals, and output the "sniffing" judgment result (correct / unclear / incorrect) for each group of sensor arrays.

[0084] Central control and computing unit: including programmable logic controller or industrial computer, used to coordinate and control the operation of the entire system: based on the output of intelligent analysis and decision unit, it determines whether the experiment should be terminated; if not terminated, it sends an instruction to dynamic dilution unit to adjust the dilution ratio; if terminated, it automatically calculates the final odor concentration based on the judgment results of all rounds and generates a report.

[0085] Human-computer interaction unit: including touch screen or connected host computer software, used to set experimental parameters (sample type, initial dilution factor, etc.), start / stop experiment, display detection progress and intermediate results in real time, and view and output final test report.

[0086] This invention has the following significant advantages:

[0087] Complete objectivity and high repeatability: The influence of human factors is completely eliminated, and the test results depend only on the performance of the sensor and algorithm, with extremely high repeatability and inter-laboratory comparability.

[0088] High efficiency and automation: It enables fully automated operation of the entire process, reducing the time for a single test from several hours to tens of minutes, and enables continuous batch testing, greatly improving monitoring efficiency.

[0089] Standard compatibility: It strictly follows and simulates every technical step and judgment logic of the current national standard method, ensuring the equivalence of the test results with the results of the manual method in principle.

[0090] Intelligent and robust: The neural network model used can learn the response patterns of complex odor mixtures. Combined with embedded physical rules, it has good resistance to cross-interference and background gas noise, improving the accuracy of odor detection for low concentrations or complex components.

[0091] Reduced operating costs and risks: There is no need to train and maintain a team of olfactory detectors, avoiding personnel health risks and significantly reducing long-term operating costs. Specific Implementation Example 1:

[0093] The system mainly includes the following modules:

[0094] 1. Dynamic dilution and gas distribution unit, such as Figure 3 As shown: including a first gas path, a second gas path, a third gas path, and a fourth gas path connected to a Venturi mixer.

[0095] A mass flow controller MFC-D01 is installed in the first air line. MFC-D01 controls the total clean air flow rate (e.g., 10 L / min).

[0096] The second, third, and fourth gas paths are connected in parallel, through which the odor sample to be tested is introduced. A mass flow controller MFC-D02 is installed on the second gas path, MFC-D03 on the third, and MFC-D04 on the fourth. MFC-D02, MFC-D03, and MFC-D04 are used to finely control the dilution ratio at different ranges (e.g., 2 mL / min, 10 mL / min, and 1 L / min).

[0097] Clean air and the odor sample to be tested are instantly mixed in the negative pressure zone of the Venturi mixer.

[0098] 2. Multi-channel sensing and detection unit, such as Figure 4 As shown: Four sensor arrays are set up, each containing five different types of metal oxide semiconductor (MOS) gas sensors (such as models sensitive to hydrogen sulfide, ammonia, methanethiol, volatile organic compounds (VOCs), and benzene compounds). Each array is fronted by a miniature rotor flow meter (A01, A02, A03, A04) to ensure consistent flow rates (e.g., all 200 mL / min).

[0099] 3. Signal Acquisition and Preprocessing Unit: A 24-bit high-precision ADC is used to acquire sensor voltage signals. A microprocessor (such as an ARM Cortex-M series) runs the program in real time, acquiring a set of data every 10 seconds, performing moving average filtering, and calculating the Δ% of each sensor's value relative to its most recent clean air baseline value.

[0100] 4. Intelligent Analysis and Decision-Making Unit: Employs an embedded AI acceleration chip (such as a Neural Processing Unit, NPU) or a high-performance microprocessor. Deploys a pre-trained fully connected neural network model. The input layer has 20 neurons (4 groups × 5 sensors). The feature abstraction layer has 44 neurons ((2*5+1)*4). For environmental samples, the output layer has 3 neurons. This model has been trained using a large amount of historical manual olfaction experimental data and its corresponding sensor response data, and can accurately map sensor response patterns to "correct / unclear / incorrect" judgments.

[0101] 5. Central Control and Computing Unit: An industrial computer (IPC) runs the control software. The software controls the entire experimental process: setting the initial dilution factor (e.g., 10x) and initiating detection. After each round, it receives four sets of judgment results and calculates the M value. If the M value > 0.58, the dilution factor is automatically increased by one level (e.g., from 10x to 30x), and the next round begins. The experiment terminates when the calculated M value in a certain round is ≤ 0.58. Based on the results of the terminated round and all previous rounds, the IPC uses a built-in standard calculation formula to calculate the odor concentration Y.

[0102] 6. Human-computer interaction unit: The touch screen connected to the IPC displays the operation interface and real-time data curves.

[0103] Workflow:

[0104] Users connect the ambient air sample bag to the sample gas inlet, select the "Ambient Air" detection mode on the touchscreen, set the initial parameters, and then click Start.

[0105] The system flushes all air paths and sensors with clean air and collects baseline values.

[0106] The dynamic dilution unit distributes gas according to the initial dilution factor (e.g., 10 times), and then mixes the clean air and the diluted sample gas according to the dilution factor and introduces them into the corresponding channels of the four sensor arrays respectively (the system internally assigns the task of "smelling" the mixed gas to each array).

[0107] The signal acquisition unit acquires the responses of four sensor arrays to three gases, calculates Δ%, and sends the preprocessed data to the intelligent analysis and decision-making unit.

[0108] The intelligent analysis and decision-making unit runs a neural network model, combines it with the embedded "environmental sample judgment rules", and outputs four sets of results, such as: [√,?,×,√].

[0109] The central control unit calculates the M value for this round. Assuming M = 1.2 > 0.58, the process does not terminate. A control command is sent to the dynamic dilution unit to adjust the dilution ratio to 30 times.

[0110] Repeat the above steps. Assuming that when the dilution factor reaches 100 times, the four results are: [?, ×, ×, ?]. The calculated value is M = 0.45 ≤ 0.58, which satisfies the termination condition.

[0111] Based on the results of the three rounds of judgment (10x, 30x, and 100x), the central control unit automatically calculates the odor concentration of the ambient air sample (e.g., Y=85), displays the results on the touch screen, and generates a PDF test report.

[0112] The results of the olfaction identification using this system are compared with those of existing methods, and the comparison is shown in Table 1:

[0113] Table 1. Comparison of olfaction results between our system and existing methods.

[0114]

[0115] Standard methods rely on manual olfactory identification, and the results are affected by subjective factors such as individual olfactory sensitivity and physical condition, resulting in poor repeatability. As shown in the comparison data in the table above, the relative standard deviation (RSD) of the standard method in the laboratory can reach 15.4%–22.0% at an odor concentration of 20. In contrast, the RSD of the method of this invention is only 7.8% at the same concentration, with a significantly narrowed data fluctuation range. This indicates that the present invention, through automated detection and judgment, significantly reduces the variation introduced by human factors, improving the objectivity and consistency of the test results.

[0116] The standard method requires 8 people, with an average testing time of 20 minutes per sample, which is inefficient. The intelligent testing method of this invention requires only 1 person, and the testing time is shortened to 10 minutes. While significantly reducing manpower requirements, the testing time for a single sample is reduced by half, providing a technical foundation for achieving higher throughput odor detection and effectively solving the problems of low efficiency and inability to adapt to rapid or large-scale testing scenarios of traditional methods.

[0117] Repeatability is defined as the maximum difference between two independent test results performed on the same sample under normal and correct operating conditions, by the same operator, in the same laboratory, using the same instrument, and within a short period of time, at a 95% probability level. Comparative data shows that this method is more concentrated and stable overall than the standard method at different concentrations.

[0118] The integration of existing electronic nose technology into dynamic dilution olfaction has not yet been achieved, particularly in the threshold determination and dilution gradient adaptation for complex mixed odors, where mature solutions are lacking. This invention addresses this gap. It is not simply a replacement for an electronic nose, but rather integrates dynamic dilution control and intelligent sensing and judgment algorithms. It can automatically adapt to dilution gradients and accurately and stably determine the threshold values ​​for complex mixed odors. This is demonstrated by the intelligent testing method's ability to output a series of stable test values, showcasing its breakthrough in automation and intelligence while simulating the manual olfaction process.

[0119] This invention successfully addresses the three major challenges of traditional olfactory testing methods—high subjectivity, low efficiency, and difficulty in standardization—through automation and intelligence. It not only demonstrates clear advantages in specific indicators, but more importantly, it proposes a mature solution integrating an intelligent judgment system within a dynamic dilution framework, filling a technological gap in this field and promoting the development of olfactory detection towards objectivity, efficiency, and standardization.

[0120] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any improvements made using the inventive concept and technical solution of the present invention, or direct application to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. A simulated intelligent dynamic dilution olfaction method, characterized in that, Includes the following steps: S1. Automated dynamic dilution and sample preparation: Using a dynamic dilution module, the odor sample to be tested is mixed with odorless clean air in real time at a preset, precisely controllable dilution ratio to prepare a test gas with a specific dilution ratio. S2, Multi-channel sensor array response acquisition: The test gas prepared in step S1 is introduced into several sets of gas sensor arrays in parallel, and each set of gas sensor arrays contains multiple heterogeneous gas sensors. S3. Sensor signal preprocessing and feature extraction: The baseline signal and response signal to the test gas of each sensor in each gas sensor array are collected, and calculations and preprocessing are performed to obtain the feature vector characterizing the intensity of gas stimulation. S4. Intelligent olfaction and judgment based on artificial intelligence: The multiple feature vectors obtained in step S3 are input into the pre-trained artificial neural network model for processing; The artificial neural network model simulates the judgment logic of multiple olfactory judges based on the characteristics of the input signal, and outputs the judgment result corresponding to each sensor array that conforms to the standard olfactory judgment procedure; the type of judgment result is selected from "correct", "unclear" and "incorrect"; S5. Automated process control and concentration calculation: Based on the multiple judgment results output in step S4, a judgment is made according to the preset experimental termination conditions that are consistent with the standard method for artificial dynamic dilution olfaction. If the termination condition is not met, the dilution factor will be automatically adjusted, and the process will return to step S1 to start the next round of testing. If the termination condition is met, the experiment will be terminated, and the odor concentration value will be automatically calculated and output based on the dilution factor of all detection rounds and the corresponding judgment results.

2. The simulated intelligent dynamic dilution olfaction method according to claim 1, characterized in that, In step S1, real-time mixing is achieved through a Venturi mixer: odorless clean air is used as the power source to flow through the throat of the Venturi tube to generate a high-speed jet and negative pressure, which entrains the odor sample to be tested, and the two achieve instantaneous turbulent mixing in the Venturi tube.

3. The simulated intelligent dynamic dilution olfaction method according to claim 1, characterized in that, In step S3, the calculation and preprocessing specifically include: using the sensor data processing module to remove noise, interference, drift, and interference information generated by random factors; combining hardware filtering and software filtering methods; adding an operational amplifier; and using a state equation-based method to estimate the true value of the signal in real time to process random noise and system errors. For each sensor, the normalized response change rate Δ% to the test gas is calculated: Δ% = (test response value - baseline value) / baseline value × 100%. The calculated multiple Δ% values ​​are normalized to form a feature vector.

4. The simulated intelligent dynamic dilution olfaction method according to claim 3, characterized in that, In step S4, the artificial neural network model is a fully connected feedforward neural network, and the output layer is configured differently according to the sample type: When detecting ambient air or fugitive emission samples, the output layer is set with three neurons, corresponding to three categories of judgment results: "correct", "unclear" and "incorrect". When testing exhaust gas samples from stationary pollution sources, the output layer is equipped with two neurons, corresponding to the "correct" and "incorrect" judgment results respectively.

5. The simulated intelligent dynamic dilution olfaction method according to claim 4, characterized in that, In step S4, the decision-making process of the artificial neural network model incorporates the judgment rules of the standard olfactory discrimination method: For ambient air or fugitive emission samples, if the model determines it to be "correct", the corresponding sensor signal characteristics must meet the preset first rule: Δ% of any one sensor > the first threshold, or Δ% of at least two sensors > the second threshold; the first threshold is greater than the second threshold. For stationary pollution source exhaust gas samples, if the model determines it to be "correct", the corresponding sensor signal characteristics must meet the preset second rule: Δ% of any sensor > the third threshold.

6. The simulated intelligent dynamic dilution olfaction method according to claim 5, characterized in that, The first threshold is 50%, the second threshold is 20%, and the third threshold is 30%.

7. The simulated intelligent dynamic dilution olfaction method according to claim 1, characterized in that, In step S5, the experimental termination conditions are consistent with the standard artificial dynamic dilution olfactory identification method: For ambient air or fugitive emission samples, the termination condition is: the M value calculated based on the judgment results of all groups at the current dilution factor is less than or equal to 0.58; For stationary pollution source exhaust gas samples, the termination condition is: two consecutive sets of judgment results are "error".

8. A simulated intelligent dynamic dilution olfaction system, used to implement the simulated intelligent dynamic dilution olfaction method as described in any one of claims 1-7, characterized in that, include: The dynamic dilution and gas mixing unit is used to mix the odor sample to be tested with odorless clean air in a precise ratio and output the test gas with a specified dilution factor. The multi-channel sensing and detection unit includes several parallel arrays of gas sensors for simultaneously acquiring response signals to the test gas. The signal acquisition and preprocessing unit is connected to the multi-channel sensing and detection unit and is used to acquire, process, and extract feature vectors from sensor signals. The intelligent analysis and decision-making unit has a built-in artificial neural network model, which is used to receive feature vectors, analyze them, and output judgment results that conform to the standard olfactory identification procedure. The central control and computing unit connects the intelligent analysis and decision-making unit and the dynamic dilution and gas distribution unit. It is used to judge the experimental progress based on the judgment results, control the adjustment of the dilution ratio, and calculate the final odor concentration when the experiment ends. The human-computer interaction unit is used to set experimental parameters, start / stop experiments, display the detection process and intermediate results in real time, and view and output the final detection report.

9. The simulated intelligent dynamic dilution olfaction system according to claim 8, characterized in that, The dynamic dilution and gas distribution unit includes a venturi mixer and multiple mass flow controllers, which are used to precisely control the flow rates of the odor sample and the odorless clean air, respectively.

10. A simulated intelligent dynamic dilution olfaction system according to claim 9, characterized in that, Each gas sensor array in the multi-channel sensing unit is equipped with an independent flow control component at its front end to ensure that the gas flow through each sensor array is constant and consistent.

Images