Medical vaporizer verification system

Abstract

The invention discloses a medical vaporizer verification system, and relates to the technical field of medical instrument detection. Comprising an environment perception acquisition module, an environment data preprocessing and structuring module, a condensation risk feature extraction and risk value quantification module, a deep learning driven condensation risk intelligent determination module and a particle dynamic identification and condensate droplet elimination module. Multi-dimensional physical parameters of environmental conditions are obtained in real time through a multi-type high-sensitivity sensor array arranged in a detection cabin. According to the invention, environmental changes are monitored in real time through a multi-sensor array, condensation risk features are extracted, and an intervention mechanism is triggered in advance by using deep learning intelligent pre-judgment; ultrasonic excitation and a laser interferometer are combined to collect particle vibration characteristics, and machine learning is matched to quickly distinguish drug particles and condensate drops, so that errors are effectively eliminated, the verification accuracy of an atomizer and the intelligent level of the system are improved, and the detection reliability and clinical safety are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of medical device detection, and particularly relates to a medical nebulizer verification system. Background Art

[0002] A medical nebulizer verification system refers to a technical system specifically used for performance testing, parameter calibration, and compliance assessment of medical nebulizer devices. This system usually integrates components such as high-precision sensors, flow control modules, particle size analyzers, data acquisition and processing units, etc., and can comprehensively detect key indicators such as the particle size distribution of aerosol particles output by the nebulizer, atomization rate, liquid medicine output volume, and continuous working stability under simulated actual use environments. By comparing and analyzing the test results with national or industry standards (such as YY 0672, ISO 27427, etc.), this system can be used to evaluate whether the working performance of the nebulizer meets the standards, and assist medical device manufacturers, testing institutions, or hospitals to ensure the safety, effectiveness, and consistency of nebulizer products.

[0003] The existing technology has the following deficiencies: In the existing technology, the performance verification of medical nebulizers usually relies on a particle size analysis module based on the principles of laser diffraction or light scattering to evaluate the particle size distribution of aerosol particles output by the nebulizer and the concentration of effective drug particles, and is supplemented by an ultrasonic generator to enhance the particle distribution uniformity or adjust the atomization state. However, in some environmental conditions with high humidity or rapid cooling with large temperature differences, during the atomization process, in addition to generating drug-containing aerosol particles, a large number of non-drug condensation microdroplets formed by the condensation of water vapor may also be induced. Since these condensation droplets are highly similar to drug particles in terms of particle size, morphological characteristics, and optical response characteristics, the existing particle size analysis module is difficult to effectively distinguish them, and identification errors are very likely to occur.

[0004] In the case where the detection system cannot accurately identify and eliminate these condensation droplets, the system misjudges them as effective drug particles, and then wrongly evaluates the output performance of the nebulizer. This situation of overestimating the output particle concentration and the quality of the particle size distribution may lead to the misjudgment of equipment with unqualified atomization performance as qualified equipment. When further applied clinically, the actual drug dose inhaled by patients is significantly lower than the standard dose requirement. Especially for special populations such as children and the elderly who suffer from chronic respiratory diseases such as asthma and chronic obstructive pulmonary disease (COPD), long-term use of such nebulizer devices with misjudged performance may lead to insufficient drug efficacy, delayed illness, and even high-risk medical events such as acute exacerbation of respiratory failure in severe cases, thus increasing medical risks, potential safety hazards in equipment use, and legal responsibilities of production enterprises and medical institutions.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and thus it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] An object of the present invention is to provide a medical nebulizer calibration system, which can monitor environmental changes in real time through a multi-sensor array, extract condensation risk characteristics and use deep learning to make intelligent predictions, and trigger an intervention mechanism in advance; combine ultrasonic excitation and a laser interferometer to collect particle vibration characteristics, cooperate with machine learning to quickly distinguish drug particles from condensate droplets, effectively eliminate errors, improve the calibration accuracy of the nebulizer and the intelligent level of the system, and ensure reliable detection and clinical safety, so as to solve the problems in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solution: A medical nebulizer calibration system, including an environmental perception and acquisition module, an environmental data preprocessing and structuring module, a condensation risk feature extraction and risk value quantification module, a deep learning-driven condensation risk intelligent determination module, and a particle dynamic recognition and condensate droplet exclusion module:

[0008] The environmental perception and acquisition module, during the calibration process of the medical nebulizer, through a multi-type high-sensitivity sensor array arranged inside the detection chamber, obtains multi-dimensional physical parameters of the environmental conditions in real time;

[0009] The environmental data preprocessing and structuring module preprocesses the raw environmental data collected in real time and constructs a standardized environmental condition data set according to the sampling period;

[0010] The condensation risk feature extraction and risk value quantification module extracts key feature indicators reflecting that the environmental conditions reach the condensation risk from the preprocessed data set through feature engineering techniques, and comprehensively analyzes the extracted key feature indicators to quantify the environmental condition risk value;

[0011] The deep learning-driven condensation risk intelligent determination module inputs the risk key indicators after comprehensive analysis into a deep learning model pre-trained based on historical data, and through the deep learning model, makes an intelligent evaluation of the environmental conditions to determine whether the current environment reaches the trigger condensation risk critical point;

[0012] The particle dynamic recognition and condensate droplet exclusion module, when the environmental conditions trigger the condensation risk critical point, dynamically adjusts the frequency and amplitude of the ultrasonic generator to make the drug particles and condensate droplets generate micro-amplitude vibrations, captures the particle dynamic vibration spectrum information through a high-sensitivity laser interferometer, analyzes the spectrum differences of the particle vibrations in real time, and uses machine learning algorithms to quickly identify and exclude non-drug condensate droplets.

[0013] Preferably, during the verification of a medical nebulizer, multi-type high-sensitivity sensor arrays deployed inside the detection chamber are used to obtain multi-dimensional physical parameters in real time. The specific steps are as follows:

[0014] First, according to the structure of the detection chamber and the characteristics of the air flow distribution, various sensor nodes are arranged to ensure that all key areas are covered;

[0015] Second, a unified sampling period and timestamp mechanism are set to synchronously collect and control various sensors to ensure the consistency of data timing;

[0016] Third, the data stream collected by each sensor node is read and cached in real time to form high-frequency continuous raw physical parameter data;

[0017] Finally, the acquired data is preliminarily integrated and identified through the central control processing module or the data acquisition card, providing a structured and multi-dimensional data input basis for subsequent data preprocessing, feature extraction, and condensation risk judgment.

[0018] Preferably, key feature indicators reflecting that the environmental conditions reach the condensation risk are extracted from the preprocessed data set through feature engineering techniques. The extracted key feature indicators include the relative humidity mutation rate per unit time and the local area air flow temperature drop amplitude inside the chamber. The relative humidity mutation rate per unit time and the local area air flow temperature drop amplitude inside the chamber are comprehensively analyzed under the detection window to generate a relative humidity rate change reference value and an air flow critical temperature drop reference value respectively, and the environmental condition risk value is quantified through the relative humidity rate change reference value and the air flow critical temperature drop reference value.

[0019] Preferably, the specific steps for comprehensively analyzing the relative humidity mutation rate per unit time under the detection window to generate a relative humidity rate change reference value are as follows:

[0020] The relative humidity numerical data at adjacent sampling periods is collected in real time. Based on the absolute difference between adjacent sampling points, the humidity mutation amplitude feature is extracted. The extraction formula is as follows:

[0021] ΔRH i =|RH i+1 -RH i |

[0022] , where RH i is the relative humidity value collected at the i-th sampling point, RH i+1 is the relative humidity value collected at the (i + 1)-th sampling point, and ΔRH i is the absolute value amplitude of the relative humidity change between the i-th and (i + 1)-th sampling points;

[0023] On the basis of completing the extraction of the humidity mutation amplitude characteristics, a non-linear amplification and accumulation mechanism is introduced to generate a reference value for the relative humidity rate of change. The generation formula is as follows:

[0024]

[0025] , where γ RH is the reference value for the relative humidity rate of change, N is the total number of sampling points, γ is the mutation amplification coefficient, and tanh(·) is the hyperbolic tangent function.

[0026] Preferably, the specific steps for comprehensively analyzing the temperature drop amplitude of the local area air flow in the cabin under the detection window to generate a reference value for the critical air flow cooling are as follows:

[0027] The air flow temperature data is collected in real time through temperature sensors arranged in the local area of the detection cabin. Based on the collected temperature change information, the air flow temperature change rate factor is calculated. The calculation expression is as follows:

[0028]

[0029] , where ΔT is the air flow temperature change amplitude, Δt is the sampling time interval, β is the temperature change response adjustment factor, T0 is the initial temperature reference value of the cabin, e is the natural base, and Λ is the air flow temperature change rate factor;

[0030] After obtaining the temperature change rate factor Λ, combined with the humidity change characteristics in the detection environment, a reference value for the critical air flow cooling is generated. The generation formula is as follows:

[0031]

[0032] , where Ψ is the reference value for the critical air flow cooling, H is the current detected environment humidity value, H max is the set maximum humidity reference value, is the thermal-humidity coupling adjustment factor, and T env is the real-time average temperature of the detection environment.

[0033] Preferably, the reference value for the relative humidity rate of change and the reference value for the critical air flow cooling after comprehensive analysis are input into a deep learning model pre-trained based on historical data. The deep learning model generates a condensation risk coefficient, and the environmental conditions are intelligently evaluated through the condensation risk coefficient to determine whether the current environment reaches the trigger condensation risk critical point.

[0034] Preferably, the condensation risk coefficient generated when the environmental conditions are intelligently evaluated through a deep learning model pre-trained based on historical data is compared and analyzed with a preset condensation risk coefficient reference threshold to determine whether the current environment reaches the trigger condensation risk critical point. The judgment logic is as follows:

[0035] If the condensation risk coefficient is greater than the preset condensation risk coefficient reference threshold, it is determined that the current environment reaches the trigger condensation risk critical point; if the condensation risk coefficient is less than or equal to the preset condensation risk coefficient reference threshold, it is determined that the current environment does not reach the trigger condensation risk critical point.

[0036] Preferably, when the environmental conditions trigger the condensation risk critical point, the frequency and amplitude of the ultrasonic generator are dynamically adjusted to make the drug particles and the condensate droplets generate micro-amplitude vibrations. The specific steps of capturing the dynamic vibration spectrum information of the particles by a highly sensitive laser interferometer, analyzing the spectrum difference of the particle vibrations in real time, and using machine learning algorithms to quickly identify and exclude non-drug condensate droplets are as follows:

[0037] When the environmental conditions trigger the condensation risk critical point, the ultrasonic excitation device is started, and the ultrasonic excitation parameters are adjusted in real time according to the condensation risk coefficient CRC deviation value. The adjustment formulas for frequency and amplitude are as follows:

[0038]

[0039] , where f us is the dynamically adjusted ultrasonic excitation frequency, f0 is the reference ultrasonic frequency, α is the frequency adjustment coefficient, CRC is the condensation risk coefficient, CRC th is the condensation risk coefficient reference threshold, θ is the frequency adjustment index factor, A us is the dynamically adjusted ultrasonic excitation amplitude, A0 is the reference ultrasonic amplitude, μ is the amplitude adjustment coefficient, and η is the amplitude adjustment index factor;

[0040] Under the action of ultrasonic excitation, the particles will generate micro-amplitude vibration responses. A highly sensitive laser interferometer is used to collect the particle vibration spectrum signals in real time, and a spectrum difference characteristic index is constructed. The formula is as follows:

[0041]

[0042] , where VSI is the vibration spectrum difference index, which is used to quantify the dynamic vibration behavior difference between the current particle to be identified and the drug particles under the acoustic excitation condition, Ω P (j) is the vibration amplitude of the standard drug particle at the j-th frequency point, Ω D (j) is the vibration amplitude of the current particle to be identified at the j-th frequency point, M is the total number of frequency points, and κ is the nonlinear amplification factor.

[0043] Based on the intelligent identification of particle categories and the exclusion of condensate droplets by the machine learning model, the constructed vibration spectrum difference index VSI is input into the trained machine learning model. By comparing the discrimination score output by the model with the decision threshold, the category of the particle is judged. The judgment logic is as follows:

[0044]

[0045] , where Class particle is the determination result of particle category, and σ(VSI, Θ ML ) is the output classification score of the machine learning model, Θ ML is the set of machine learning model parameters, σ(·) is a function symbol indicating "substituting VSI into the model and combining the model parameters to obtain an output score", and H is the particle category determination threshold;

[0046] If the recognition result is a condensate droplet, the condensate particles are removed from the detection data and do not participate in the evaluation calculation of the atomizer performance to ensure the authenticity and effectiveness of the final drug output particle concentration evaluation.

[0047] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0048] The present invention monitors the changes in environmental conditions in real time through a highly sensitive multi-sensor array, extracts the key features of the condensation risk by combining feature engineering, and uses a deep learning model to realize the intelligent prediction of the environmental state, enabling the system to actively identify and trigger an intervention mechanism before the condensation risk appears; further, by dynamically adjusting the ultrasonic excitation and using a laser interferometer to capture the particle vibration characteristics in real time, and combining machine learning algorithms to quickly distinguish drug particles from condensate droplets, accurately removing non-drug components, and ensuring the accuracy and reliability of the verification data. Overall, this solution significantly improves the adaptability and intelligent level of the medical atomizer verification system under complex environmental conditions, greatly reduces the risk of detection misjudgment caused by environmental condensation interference, and thus provides a solid guarantee for equipment quality control and clinical treatment safety. Description of the Drawings

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments recorded in the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings.

[0050] Figure 1 It is a module schematic diagram of a medical atomizer verification system of the present invention. Detailed Embodiments

[0051] Now, the exemplary embodiments will be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided so that the present disclosure will be more complete and comprehensive, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0052] The present invention provides a medical nebulizer calibration system as Figure 1 shown, which includes an environmental perception and acquisition module, an environmental data preprocessing and structuring module, a condensation risk feature extraction and risk value quantification module, a deep learning-driven intelligent condensation risk determination module, and a particle dynamic recognition and condensate droplet exclusion module:

[0053] The environmental perception and acquisition module, during the medical nebulizer calibration process, uses a multi-type high-sensitivity sensor array deployed inside the detection chamber to obtain multi-dimensional physical parameters of the environmental conditions in real time;

[0054] During the medical nebulizer calibration process, in order to accurately perceive and dynamically monitor the environmental changes inside the detection chamber, the system usually reasonably deploys a multi-type high-sensitivity sensor array inside the chamber. These sensors include temperature sensors, humidity sensors, air flow velocity sensors, air pressure sensors, and dew point detection modules, etc. Through this array, synchronous and real-time acquisition of multiple key physical parameters in the environmental conditions can be achieved, so as to comprehensively master the thermal and humidity state, air flow characteristics, and micro-environment change trend inside the detection chamber. These multi-dimensional environmental parameters collected at high frequency not only provide a data basis for subsequent condensation risk prediction and dynamic response, but also can be used to assist in analyzing the stability of atomized particle generation, diffusion characteristics, and the environmental adaptability of the nebulizer working performance, which is of great significance for improving the accuracy and intelligent level of the overall calibration system.

[0055] During the medical nebulizer calibration process, multi-dimensional physical parameters are obtained in real time through a multi-type high-sensitivity sensor array deployed inside the detection chamber. The specific steps are as follows: First, according to the structure of the detection chamber and the characteristics of air flow distribution, arrange various sensor nodes such as temperature, humidity, air pressure, wind speed, and dew point to ensure that all key areas are covered; Second, set a unified sampling period and timestamp mechanism to synchronously collect and control various sensors to ensure the consistency of data time sequence; Third, read and cache the data stream collected by each sensor node in real time to form high-frequency continuous raw physical parameter data; Finally, preliminarily integrate and identify the obtained data through the central control processing module or data acquisition card to provide a structured and multi-dimensional data input basis for subsequent data preprocessing, feature extraction, and condensation risk judgment. The key to this process is to achieve high-precision perception and spatio-temporal synchronous perception of environmental parameters, providing a stable and reliable perception support for the calibration system.

[0056] The environmental data preprocessing and structuring module preprocesses the real-time collected raw environmental data and constructs a standardized environmental condition data set according to the sampling period;

[0057] Preprocessing the raw environmental data collected in real time refers to performing a series of data cleaning operations such as denoising, filling missing values, removing outliers, and smoothing on the raw multi-dimensional physical data such as temperature, humidity, and air pressure collected by sensors in the detection cabin, with the aim of improving data quality and stability. Subsequently, the cleaned data is time-synchronized and formatted according to a set sampling period (for example, every second or every minute) to construct an environmental condition dataset with consistent structure and standardized dimensions. This standardized dataset can serve as a standard input source for subsequent feature extraction, condensation risk identification, and model analysis. Its main function is to ensure the continuity and comparability of the data, eliminate the interference of sensor noise and sampling differences on the system's judgment results, and thus provide an accurate and reliable data foundation for intelligent evaluation.

[0058] The condensation risk feature extraction and risk value quantification module uses feature engineering technology to extract key characteristic indicators reflecting the environmental conditions that reach the condensation risk from the preprocessed data set, and conducts a comprehensive analysis of the extracted key characteristic indicators to quantify the environmental condition risk value;

[0059] Through feature engineering technology, key characteristic indicators reflecting the risk of condensation under environmental conditions are extracted from the preprocessed data set. The extracted key characteristic indicators include the relative humidity mutation rate per unit time and the temperature drop of the airflow in the local area of the cabin. The relative humidity mutation rate per unit time and the temperature drop of the airflow in the local area of the cabin are comprehensively analyzed within the detection window to generate relative humidity rate change reference value and airflow critical temperature drop reference value, respectively. The relative humidity rate change reference value and airflow critical temperature drop reference value are used to quantify the environmental condition risk value.

[0060] The environmental condition risk is quantified by the relative humidity rate change reference value and the airflow critical temperature drop reference value, aiming to make dynamic, accurate and quantifiable judgments on the potential condensation risk in the test cabin environment. The relative humidity rate change reference value reflects the severity of the humidity mutation per unit time and can capture the trend of water vapor rapidly approaching saturation; while the airflow critical temperature drop reference value is used to measure the abnormal drop in airflow temperature in a local area of the cabin within a short period of time, reflecting the thermal change behavior that may cause local supercooling or condensation. Through the coordinated quantitative analysis of these two reference values, not only can the trend of environmental conditions evolving from stability to the condensation boundary be identified in advance at the microscopic scale, but it can also provide a quantitative judgment basis for whether the system triggers subsequent intervention mechanisms (such as ultrasonic excitation recognition), thereby realizing intelligent perception and precise control of environmental risks and effectively avoiding detection misjudgments caused by drastic environmental changes.

[0061] A significantly accelerated rate of relative humidity change per unit time typically indicates that the test chamber environment is rapidly approaching a critical condensation state and is a key early warning indicator for determining whether the environment has reached a condensation risk. Relative humidity itself indicates the level of water vapor saturation in the air. A rapid increase in relative humidity over a very short period of time indicates that the air within the chamber is rapidly accumulating water vapor or that a sudden drop in temperature is approaching saturation. Especially in confined spaces or those with limited airflow, sudden humidity changes can easily cause the air to approach or even exceed the dew point, prompting a phase change in water vapor, condensing into microdroplets that adhere to the chamber walls or float in the air. These condensation droplets can easily be misidentified as drug aerosol particles during nebulizer testing, leading to detection errors. By monitoring the dynamic changes in relative humidity, particularly its rate of change per unit time, it is possible to accurately identify the potential for condensation in an environment before condensation actually occurs. Therefore, this indicator is highly sensitive and has predictive value in condensation risk assessment.

[0062] The specific steps for comprehensively analyzing the relative humidity mutation rate per unit time within the detection window to generate a reference value for the relative humidity rate change are as follows:

[0063] The relative humidity data of adjacent sampling periods are collected in real time. Based on the absolute difference between two adjacent sampling points, the humidity mutation amplitude characteristics are extracted. The extraction formula is as follows:

[0064] ΔRH i =|RH i+1 -RH i |

[0065] , where RH i is the relative humidity value collected at the i-th sampling point, RH i+1 is the relative humidity value collected at the i+1th sampling point, ΔRH i is the absolute value of the relative humidity change between the i-th and i+1-th sampling points;

[0066] Through the above extraction process, the slowly changing data components in the overall change trend can be stripped away, retaining only the sudden humidity change characteristics, thereby capturing the subtle changes in the potential condensation risk in the detection cabin environment at an early stage and forming the data basis for subsequent index construction.

[0067] On the basis of extracting the humidity mutation amplitude feature, a nonlinear amplification and accumulation mechanism is introduced to generate a reference value of the relative humidity rate change. The generation formula is as follows:

[0068]

[0069] , where γ RHis the reference value of the relative humidity rate change, N is the total number of sampling points, γ is the mutation amplification coefficient, which is the mutation amplitude amplification coefficient introduced when performing non-linear amplification processing on the humidity mutation amplitude each time. Its usual value range is between 1 and 5. tanh(·) is the hyperbolic tangent function, which is used as a non-linear amplification operator to enhance the contribution weight of large-amplitude humidity mutations to the overall index and suppress the influence of small perturbations. Its output value range is (-1, 1).

[0070] Through the above non-linear processing and accumulation, the perception ability of the local environmental mutation trend in the detection chamber can be effectively strengthened, making the reference value of the relative humidity rate change more sensitive to the dynamic response of the condensation risk and having anti-interference characteristics, thereby providing a highly reliable quantitative basis for the subsequent condensation risk determination of the system.

[0071] The larger the reference value of the relative humidity rate change generated by comprehensively analyzing the relative humidity mutation rate per unit time under the detection window, the faster the relative humidity in the environmental conditions is approaching the saturation state, that is, the water vapor content in the air is increasing rapidly or the temperature is dropping rapidly, thus making the air more likely to reach or exceed the dew point temperature and trigger the condensation phenomenon. Therefore, the larger the reference value of the relative humidity rate change, the higher the condensation risk state the current environment is usually in; on the contrary, when the reference value is small or changes slowly, it indicates that the environment is relatively stable, the humidity in the air changes smoothly, and there is no trend of rapidly approaching the condensation critical point, and it can be determined that the environment has not reached the condensation risk.

[0072] The rapid drop in the air flow temperature in a local area inside the chamber in a very short period of time is usually an important physical sign that the environmental conditions are about to reach or have reached the condensation risk. This rapid temperature drop phenomenon will cause the water vapor carrying capacity in the air flow to drop rapidly, making the air that was originally close to the saturated state instantaneously exceed the saturation point in the local area, thereby inducing the condensation of water vapor to form condensate droplets. Especially in a closed detection chamber, when cold air or a low-temperature wall surface causes a sharp drop in the local air mass temperature, so-called "temperature difference-driven condensation" will occur. Even if the overall environmental parameters are still within the safety threshold range, a local condensation microenvironment may appear due to thermal inhomogeneity. Therefore, the non-linear and drastic downward trend of the air flow temperature in a very short period is one of the sensitive leading signals indicating the presence of condensation risk in the detection chamber. The monitoring of this phenomenon can be quantified by calculating the "critical air flow cooling reference value" to dynamically identify the condensation induction factors existing in the system operation, trigger the authenticity recognition mechanism of the atomized particles in advance, and effectively avoid the calibration distortion problem caused by misjudgment of condensate droplets.

[0073] The specific steps for comprehensively analyzing the air flow temperature drop amplitude in a local area inside the chamber under the detection window to generate the critical air flow cooling reference value are as follows:

[0074] The air flow temperature data is collected in real time by temperature sensors arranged in a local area of the detection chamber. Based on the collected temperature change information, the air flow temperature change rate factor is calculated, and the calculation expression is as follows:

[0075]

[0076] , where ΔT is the amplitude of the air flow temperature change, which reflects the severity of the temperature change of the air flow in a short period and is a direct physical quantity for identifying local cooling phenomena (i.e., condensation induction points); Δt is the sampling time interval, which serves as the time reference for the temperature change rate and is used to quantify the degree of temperature change per unit time; β is the temperature change response adjustment factor, which is used for the attenuation degree of the air flow temperature change amplitude ΔT relative to the reference temperature T0 in the exponential function, giving a higher response weight to a larger temperature drop and enhancing the sensitivity of the system to sharp temperature drop phenomena; T0 is the initial temperature reference value of the chamber; e is the natural base; Λ is the air flow temperature change rate factor, which enhances the system's perception ability of local air flow abnormal cooling phenomena by dynamically amplifying the physical response of sharp temperature drop behavior;

[0077] By introducing an exponential decay term, a higher weight is given to a larger temperature change, thereby enhancing the sensitive recognition ability of the air flow temperature sudden drop behavior and effectively highlighting the potential condensation formation conditions in the detection environment.

[0078] After obtaining the temperature change rate factor Λ, combined with the humidity change characteristics in the detection environment, an air flow critical cooling reference value is generated, and the generation formula is as follows:

[0079]

[0080] , where Ψ is the air flow critical cooling reference value, H is the current humidity value of the detection environment, H max is the set maximum humidity reference value, is the thermo-humid coupling adjustment factor, which is used to control the contribution sensitivity of temperature change to the final air flow critical cooling reference value Ψ, and T env is the real-time average temperature of the detection environment.

[0081] This formula comprehensively considers the combined effects of the air flow temperature drop amplitude, humidity level, and environmental thermal state, and realizes the accurate quantitative evaluation of the condensation risk state in the detection environment through the weighted processing of the humidity adjustment factor and the temperature difference index.

[0082] Through the above steps, the condensation risk caused by the abnormal change of the local air flow temperature in the chamber can be quantified in real time and dynamically, forming a discriminant basis that can be used to drive subsequent ultrasonic excitation identification and environmental condition intervention, effectively improving the stability and detection accuracy of the medical nebulizer calibration system in a complex environment.

[0083] The larger the airflow critical cooling reference value generated after comprehensively analyzing the temperature drop amplitude of the local airflow in the cabin body under the detection window, the closer or the more likely the environmental conditions have reached the condensation risk state. Conversely, it indicates a lower condensation risk. This reference value is a dynamic risk characterization index formed by quantitatively analyzing the short-term temperature drop amplitude of the local airflow in the cabin body within the monitoring window. When it is detected that the local airflow temperature drops sharply within an extremely short period, it indicates that the moisture-carrying capacity of the gas drops sharply, and the water vapor in the air is extremely likely to reach saturation and form condensate droplets in the local area. At this time, the airflow critical cooling reference value will show a high value, thus reflecting that the current environment has entered or is approaching the condensation critical state. On the contrary, if this reference value is small, it indicates that the airflow temperature changes smoothly and there is no sharp temperature drop sufficient to trigger the condensation condition, and the environment is still in the thermo-hygroscopic stable range.

[0084] The intelligent condensation risk determination module driven by deep learning inputs the key risk indicators after comprehensive analysis into a deep learning model pre-trained based on historical data, and through the deep learning model, it intelligently evaluates the environmental conditions to determine whether the current environment has reached the condensation risk critical point.

[0085] Input the comprehensively analyzed reference value of the relative humidity rate change and the airflow critical cooling reference value into a deep learning model pre-trained based on historical data. Through the deep learning model, generate a condensation risk coefficient, and through the condensation risk coefficient, intelligently evaluate the environmental conditions to determine whether the current environment has reached the condensation risk critical point.

[0086] The deep learning model pre-trained based on historical data refers to an intelligent discrimination model that uses a large amount of historical detection data of the cabin body environment monitoring, performs offline modeling and optimization through deep learning algorithms, and can accurately predict the condensation risk state. Specifically, in the modeling stage, first, it is necessary to collect the original historical data covering various environmental conditions (including different temperatures, humidities, airflow changes, dew point conditions, etc.) and form a standardized sample set through preprocessing. Then, extract the key feature indicators from each sample, such as the relative humidity rate change per unit time and the local airflow temperature drop amplitude, and construct the training feature vectors accordingly. At the same time, pair each group of feature vectors with the corresponding actual condensation occurrence state (i.e., whether condensate droplets actually appear) as labels to form training data with supervision signals. Based on this, use deep learning models such as LSTM (Long Short-Term Memory Network), GRU (Gated Recurrent Unit), or Transformer structure, and through a large number of iterative trainings, optimize the model parameters to enable the model to accurately learn the complex, non-linear, multi-dimensional mapping relationship between the feature indicators and the condensation risk state.

[0087] The trained deep learning model has a powerful ability to understand environmental dynamic changes and identify risk patterns. During subsequent actual applications, it can receive new real-time detection data inputs (i.e., the reference value of relative humidity rate change and the reference value of critical air flow cooling after comprehensive analysis), and automatically infer the current environmental condensation risk level based on the laws learned internally. The condensation risk coefficient output by the model is a quantitative expression of the degree to which the current environmental conditions approach the condensation critical state, usually represented by a continuous value (such as 0 to 1), where approaching 1 indicates a very high condensation risk and approaching 0 indicates environmental stability. The system can set a threshold according to the condensation risk coefficient. When the coefficient exceeds the set threshold, it can intelligently determine that the detection chamber has reached the condensation trigger condition and promptly initiate subsequent dynamic identification measures such as ultrasonic vibration and spectrum analysis. By introducing a pre-trained deep learning model, not only the accuracy and response speed of environmental condensation risk identification are greatly improved, but also the intelligence and adaptive evolution of the detection system are realized, overcoming the limitation of the traditional fixed-threshold judgment method's poor adaptability to complex environmental changes.

[0088] The deep learning model is not limited here. Any deep learning model that can comprehensively analyze the reference value γ of relative humidity rate change RH and the reference value Ψ of critical air flow cooling to generate the condensation risk coefficient CRC can be used. To implement the technical solution of the present invention, the present invention provides a specific implementation method:

[0089] The formula for generating the condensation risk coefficient CRC is as follows: CRC = P1·γ RH + P2·Ψ, where P1 and P2 are the preset proportionality coefficients of the reference value γ of relative humidity rate change RH and the reference value Ψ of critical air flow cooling respectively, and both P1 and P2 are greater than 0.

[0090] The "preset proportionality coefficient" refers to two weight coefficients used to control or adjust the influence degrees of the reference value γ of relative humidity rate change RH and the reference value Ψ of critical air flow cooling on the finally generated condensation risk coefficient CRC, that is, P1 and P2 in the formula. These two coefficients are fixed parameters preset when constructing the condensation risk model, usually determined based on a large amount of historical data analysis or system tuning experiments, aiming to reflect the sensitivity weights of these two indicators to condensation risk under different working conditions. For example, if historical data shows that the drastic fluctuation of relative humidity is more sensitive to condensation formation, then P1 > P2 can be set to enhance the weight of γ RH in the calculation of CRC. By setting P1 and P2, the model can be flexibly adapted to the dominant factors of different environmental characteristics, making the output of the condensation risk coefficient more targeted and accurate, thereby improving the intelligence level and environmental response ability of the entire detection system.

[0091] From the condensation risk coefficient, it can be seen that the greater the reference value of the relative humidity rate change generated by comprehensively analyzing the sudden change rate of relative humidity per unit time under the detection window, and the greater the reference value of the critical air flow cooling generated by comprehensively analyzing the air flow temperature drop amplitude in the local area of the cabin under the detection window, the greater the condensation risk coefficient generated when the intelligent evaluation of environmental conditions is carried out through a deep learning model pre-trained based on historical data, indicating that the probability of the current environmental conditions triggering the condensation risk critical point is greater. Conversely, the probability of the current environmental conditions triggering the condensation risk critical point is smaller.

[0092] Compare and analyze the condensation risk coefficient generated when the intelligent evaluation of environmental conditions is carried out through a deep learning model pre-trained based on historical data with the preset reference threshold of the condensation risk coefficient to determine whether the current environment reaches the condensation risk critical point. The judgment logic is as follows:

[0093] If the condensation risk coefficient is greater than the preset reference threshold of the condensation risk coefficient, it is determined that the current environment reaches the condensation risk critical point; if the condensation risk coefficient is less than or equal to the preset reference threshold of the condensation risk coefficient, it is determined that the current environment does not reach the condensation risk critical point.

[0094] The particle dynamic recognition and condensate droplet exclusion module, when the environmental conditions trigger the condensation risk critical point, dynamically adjusts the frequency and amplitude of the ultrasonic generator to make the drug particles and condensate droplets vibrate slightly, captures the particle dynamic vibration spectrum information through a highly sensitive laser interferometer, analyzes the spectrum difference of particle vibration in real time, and uses machine learning algorithms to quickly identify and exclude non-drug condensate droplets;

[0095] When the environmental conditions trigger the condensation risk critical point, dynamically adjust the frequency and amplitude of the ultrasonic generator to cause the drug particles and the condensate droplets to produce small-amplitude vibrations. Capture the spectral information of the dynamic vibrations of the particles through a highly sensitive laser interferometer, analyze the spectral differences in the particle vibrations in real time, and use machine learning algorithms to quickly identify and exclude the non-drug condensate droplets. The role of this step is to actively identify and intelligently eliminate potential condensation errors in the atomization detection process, ensuring the authenticity and reliability of the detection data from the source. Specifically, applying an ultrasonic field can stimulate the particles to produce small-amplitude vibrations. Due to the essential differences in physical properties such as density, surface tension, and internal structure between drug particles and condensate droplets, their vibration responses (such as frequency, amplitude, and vibration stability) under ultrasonic excitation will also be significantly different. A highly sensitive laser interferometer can capture the vibration characteristics of the particles in real time with nanometer-level precision, forming fine-grained and high-resolution vibration spectral data. These spectral data carry the essential information of the physical properties of the particles. By analyzing the spectral characteristics in real time, the differences in the dynamic vibration behaviors between drug aerosol particles and non-drug condensate droplets can be revealed. Subsequently, using a pre-trained machine learning classification model (such as support vector machines, convolutional neural networks, etc.), quickly classify and identify the captured vibration characteristics to accurately exclude the condensate droplet particles. Through this dynamic identification and exclusion mechanism, it is possible to effectively prevent condensate droplets from being wrongly included in the detection results of effective drug particles, prevent the overestimation of the performance of the nebulizer, and thus ensure the accuracy, scientificity of the nebulizer calibration results and the safety of clinical applications. In addition, this step achieves adaptive response and intelligent correction under changing environmental conditions, breaking through the limitation of the poor adaptability of the traditional static detection mode to complex environmental changes, greatly enhancing the robustness and intelligent decision-making ability of the calibration system against extreme environmental interference, and is an important technical support for ensuring the stable operation of the medical nebulizer calibration system under high-precision and high-reliability requirements.

[0096] When the environmental conditions trigger the condensation risk critical point, the specific steps for dynamically adjusting the frequency and amplitude of the ultrasonic generator to cause the drug particles and the condensate droplets to produce small-amplitude vibrations, capturing the spectral information of the dynamic vibrations of the particles through a highly sensitive laser interferometer, analyzing the spectral differences in the particle vibrations in real time, and using machine learning algorithms to quickly identify and exclude the non-drug condensate droplets are as follows:

[0097] When the environmental conditions trigger the condensation risk critical point, start the ultrasonic excitation device and adjust the ultrasonic excitation parameters in real time according to the condensation risk coefficient CRC deviation value. The adjustment formulas for the frequency and amplitude are as follows:

[0098]

[0099] , where f usis the ultrasonic excitation frequency after dynamic adjustment, which is the ultrasonic frequency applied to the atomized particles when the condensation risk is triggered, with the unit of Hertz. By adjusting the frequency, a distributive difference in the acoustic response between the drug particles and the condensate droplets can be created, facilitating identification. It is usually adjusted within the range of 20 - 100 kHz. The high-frequency band excites the response of finer particles. f0 is the reference ultrasonic frequency, and α is the frequency adjustment coefficient, which controls the response intensity of the excitation frequency increase after the CRC exceeds the threshold and takes a positive value. CRC is the condensation risk coefficient, CRC th is the reference threshold of the condensation risk coefficient. θ is the frequency adjustment index factor, which controls the non-linear growth curve of the frequency with the risk level, improves the sensitivity of the frequency response to high-risk areas, and magnifies small risk changes in the high range. The value range is 1.5 - 3, A us is the ultrasonic excitation amplitude after dynamic adjustment. The amplitude determines the sound pressure intensity, which in turn affects the vibration amplitude of the particles and regulates their physical behavior in space. A0 is the reference ultrasonic amplitude, and μ is the amplitude adjustment coefficient, which controls the proportional factor of the excitation amplitude increment after the CRC exceeds the threshold and affects the enhancement rate of the excitation sound pressure, used to enhance the response difference between particles. η is the amplitude adjustment index factor, which enhances the response ability of the system to distinguish droplets in the high condensation risk section;

[0100] Precise excitation of the micro-amplitude vibration of drug particles and condensate droplets is achieved through the above dynamic control formula, providing a physical response basis for subsequent identification.

[0101] Under the action of ultrasonic excitation, the particles will generate a micro-amplitude vibration response. A high-sensitivity laser interferometer is used to collect the particle vibration spectrum signal in real time, and a spectrum difference characteristic index is constructed. The formula is as follows:

[0102]

[0103] , where VSI is the vibration spectrum difference index, which is used to quantify the dynamic vibration behavior difference between the currently to-be-identified particles and the drug particles under acoustic excitation conditions. The larger the value, the more obvious the deviation of the particles from the standard drug particles in the spectrum response, and the higher the possibility of being identified as "non-drug particles" (such as condensate droplets), Ω P (j) is the vibration amplitude of the standard drug particle at the j-th frequency point, indicating the vibration intensity measured for the known standard drug particle at the j-th frequency point under acoustic excitation conditions, Ω D (j) is the vibration amplitude of the currently to-be-identified particle at the j-th frequency point, indicating the vibration response amplitude generated by the currently to-be-identified particle at the j-th frequency point under acoustic excitation conditions during the actual detection process. M is the total number of frequency points, and κ is the non-linear amplification factor, indicating the non-linear amplification degree of the spectrum difference term, usually a real number greater than 1, and the common values are 2, 3 or higher,

[0104] The above spectral difference index is used to capture the dynamic response differences between drug particles and condensate droplets under acoustic excitation conditions and has good class discrimination ability.

[0105] Input the constructed vibration spectral difference index VSI into the trained machine learning model (the model can be a support vector machine SVM, random forest RF, or one-dimensional convolutional neural network 1D-CNN, etc.). By comparing the discrimination score output by the model with the decision threshold, determine the category of the particle. The judgment logic is as follows:

[0106]

[0107] , where Class particle is the particle category determination result. This variable is the classification output result of particle recognition, representing the system's final judgment on the current input particle type. σ(VSI,Θ ML ) is the output classification score of the machine learning model. This function means that the input particle spectral difference index VSI is input into the machine learning model (such as a support vector machine, random forest, or neural network), and the particle classification score value obtained after calculation based on the model parameters Θ ML . Θ ML is the set of machine learning model parameters, and its content varies according to the model type. σ(·) is a function symbol, indicating "substitute VSI into the model, combine with the model parameters, and obtain an output score". H is the particle category determination threshold. This value is the decision critical value used to map the model output score σ to a specific category, converting the continuous output score σ into a binary classification label (drug particle or condensate droplet), directly determining the classification output result, and having the functional role of a decision boundary;

[0108] If the recognition result is a condensate droplet, then remove the condensate particles from the detection data and do not participate in the calculation of the atomizer performance evaluation to ensure the authenticity and effectiveness of the final drug output particle concentration evaluation.

[0109] The role of this step is to achieve rapid, accurate identification and dynamic exclusion of particle categories, remove condensate droplets from the metering and efficacy evaluation data in real time, ensure the reliability of the final evaluation data and the accuracy of the drug dose, and fundamentally eliminate the technical errors caused by condensate interference.

[0110] The present invention monitors the changes in environmental conditions in real time through a highly sensitive multi-sensor array, extracts the key features of the condensation risk by combining feature engineering, and uses a deep learning model to achieve intelligent prediction of the environmental state, enabling the system to actively identify and trigger an intervention mechanism before the condensation risk appears; further, by dynamically adjusting the ultrasonic excitation and using a laser interferometer to capture the particle vibration characteristics in real time, and combining machine learning algorithms to quickly distinguish drug particles from condensate droplets, non-drug components are accurately removed to ensure the accuracy and reliability of the verification data. Overall, this solution significantly improves the adaptability and intelligent level of the medical nebulizer verification system under complex environmental conditions, greatly reduces the risk of detection misjudgment caused by environmental condensation interference, and thus provides a solid guarantee for equipment quality control and clinical treatment safety.

[0111] The above formulas are all dimensionless and take their numerical values for calculation. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0112] Only some exemplary embodiments of the present invention have been described by way of illustration above. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0113] It should be noted that in this article, if there are relational terms such as first and second, they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or device including the element.

[0114] It should be understood that in various embodiments of the present application, the magnitude of the sequence numbers of the above processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0115] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0116] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0117] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0118] In addition, the functional units in each embodiment of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0119] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0120] Only some exemplary embodiments of the present invention have been described above by way of illustration. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

Claims

1. A medical nebulizer verification system, characterized in that, It includes an environmental perception and acquisition module, an environmental data preprocessing and structuring module, a condensation risk feature extraction and risk value quantification module, a deep learning-driven intelligent condensation risk determination module, and a particle dynamic identification and condensate droplet exclusion module: The environmental perception and acquisition module, during the verification process of a medical nebulizer, uses a multi-type high-sensitivity sensor array deployed inside the detection chamber to continuously obtain multi-dimensional physical parameters of the environmental conditions; The environmental data preprocessing and structuring module preprocesses the raw environmental data collected in real time and constructs a standardized environmental condition dataset according to the sampling period; The condensation risk feature extraction and risk value quantification module extracts key feature indicators reflecting the condensation risk of environmental conditions from the preprocessed dataset through feature engineering techniques, and comprehensively analyzes the extracted key feature indicators to quantify the environmental condition risk value; The deep learning-driven intelligent condensation risk determination module inputs the comprehensively analyzed risk key indicators into a deep learning model pre-trained based on historical data, and uses the deep learning model to intelligently evaluate the environmental conditions to determine whether the current environment reaches the trigger point of the condensation risk; The particle dynamic identification and condensate droplet exclusion module, when the environmental conditions trigger the condensation risk critical point, dynamically adjusts the frequency and amplitude of the ultrasonic generator to cause slight vibrations of the drug particles and condensate droplets, captures the particle dynamic vibration spectrum information through a high-sensitivity laser interferometer, analyzes the spectrum differences of the particle vibrations in real time, and uses machine learning algorithms to quickly identify and exclude non-drug condensate droplets.

2. The medical atomizer calibration system according to claim 1, characterized in that During the verification process of a medical nebulizer, the multi-dimensional physical parameters are continuously obtained through a multi-type high-sensitivity sensor array deployed inside the detection chamber. The specific steps are as follows: First, according to the structure of the detection chamber and the characteristics of the airflow distribution, various sensor nodes are arranged to ensure that all key areas are covered; Second, a unified sampling period and timestamp mechanism are set to synchronously collect and control various sensors to ensure the consistency of the data time sequence; Third, the data stream collected by each sensor node is read and cached in real time to form high-frequency continuous raw physical parameter data; Finally, the obtained data is preliminarily integrated and labeled through the central control processing module or the data acquisition card, providing a structured and multi-dimensional data input basis for subsequent data preprocessing, feature extraction, and condensation risk judgment.

3. The medical atomizer calibration system according to claim 1, wherein Key feature indicators reflecting the condensation risk of environmental conditions are extracted from the preprocessed dataset through feature engineering techniques. The extracted key feature indicators include the sudden change rate of relative humidity per unit time and the temperature drop amplitude of the local airflow inside the chamber. The sudden change rate of relative humidity per unit time and the temperature drop amplitude of the local airflow inside the chamber are comprehensively analyzed under the detection window to generate a relative humidity rate change reference value and an airflow critical cooling reference value respectively, and the environmental condition risk value is quantified through the relative humidity rate change reference value and the airflow critical cooling reference value.

4. The medical atomizer calibration system according to claim 3, wherein, The specific steps for comprehensively analyzing the sudden change rate of relative humidity per unit time under the detection window to generate a relative humidity rate change reference value are as follows: Collect the relative humidity numerical data under adjacent sampling periods in real time. Based on the absolute difference between adjacent sampling points, extract the humidity mutation amplitude feature. The extraction formula is as follows: ΔRH i = |RH i+1 - RH i | where, RH i is the relative humidity value collected at the i-th sampling point, RH i+1 is the relative humidity value collected at the (i + 1)-th sampling point, and ΔRH i is the absolute value amplitude of the relative humidity change between the i-th and the (i + 1)-th sampling points; On the basis of completing the extraction of the humidity mutation amplitude feature, introduce a non-linear amplification and accumulation mechanism to generate a reference value for the relative humidity rate change. The generation formula is as follows: where γ RH is the reference value of the relative humidity rate change, N is the total number of sampling points, γ is the mutation amplification factor, and tanh(·) is the hyperbolic tangent function.

5. The medical nebulizer calibration system according to claim 3, characterized in that, The specific steps for comprehensively analyzing the airflow temperature drop amplitude in a local area of the cabin within the detection window to generate a reference value for the critical airflow cooling are as follows: Real-time collect the airflow temperature data through temperature sensors arranged in a local area of the detection cabin. Based on the collected temperature change information, calculate the airflow temperature change rate factor. The calculation expression is as follows: In the formula, ΔT is the airflow temperature change amplitude, Δt is the sampling time interval, β is the temperature change response adjustment factor, T0 is the initial temperature reference value of the cabin, e is the natural logarithm base, and Λ is the airflow temperature change rate factor; After obtaining the temperature change rate factor Λ, combine the humidity change characteristics in the detection environment to generate a reference value for the critical airflow cooling. The generation formula is as follows: Where Ψ is the reference value of the critical temperature drop of the air flow, H is the current detected environmental humidity value, and H max is the set maximum humidity reference value, is the heat and humidity coupling adjustment factor, and T env is the real-time average temperature of the detected environment.

6. The medical atomizer calibration system according to claim 3, characterized in that, Input the comprehensively analyzed reference value for the relative humidity rate change and the reference value for the critical airflow cooling into a deep learning model pre-trained based on historical data. Generate a condensation risk coefficient through the deep learning model, and use the condensation risk coefficient to conduct an intelligent evaluation of the environmental conditions to determine whether the current environment reaches the trigger point for the condensation risk.

7. The medical atomizer calibration system according to claim 6, wherein, Compare and analyze the condensation risk coefficient generated when the environmental conditions are intelligently evaluated by a deep learning model pre-trained based on historical data with a preset reference threshold for the condensation risk coefficient to determine whether the current environment reaches the trigger point for the condensation risk. The judgment logic is as follows: If the condensation risk coefficient is greater than the preset reference threshold for the condensation risk coefficient, it is determined that the current environment reaches the trigger point for the condensation risk; if the condensation risk coefficient is less than or equal to the preset reference threshold for the condensation risk coefficient, it is determined that the current environment does not reach the trigger point for the condensation risk.

8. A medical nebulizer calibration system according to claim 7, wherein, When the environmental conditions trigger the condensation risk critical point, dynamically adjust the frequency and amplitude of the ultrasonic generator to make the drug particles and the condensate droplets produce micro-amplitude vibrations. Capture the dynamic vibration spectrum information of the particles through a highly sensitive laser interferometer, analyze the spectrum difference of the particle vibrations in real time, and use machine learning algorithms to quickly identify and exclude the non-drug condensate droplets. The specific steps are as follows: When the environmental conditions trigger the condensation risk critical point, start the ultrasonic excitation device and adjust the ultrasonic excitation parameters in real time according to the condensation risk coefficient CRC deviation value. The adjustment formulas for the frequency and amplitude are as follows: Where, f us is the ultrasonic excitation frequency after dynamic adjustment, f0 is the reference ultrasonic frequency, α is the frequency adjustment coefficient, CRC is the condensation risk coefficient, CRC th is the reference threshold of the condensation risk coefficient, θ is the frequency adjustment index factor, A us is the ultrasonic excitation amplitude after dynamic adjustment, A0 is the reference ultrasonic amplitude, μ is the amplitude adjustment coefficient, η is the amplitude adjustment index factor; Under the action of ultrasonic excitation, the particles will produce a micro-amplitude vibration response. Use a highly sensitive laser interferometer to collect the particle vibration spectrum signal in real time and construct a spectrum difference characteristic index. The formula is as follows: where VSI is the vibration spectrum difference index, which is used to quantify the difference in the dynamic vibration behavior between the currently to-be-identified particle and the drug particle under acoustic excitation conditions, and Ω P (j) is the vibration amplitude of the standard drug particle at the j-th frequency point, and Ω D (j) is the vibration amplitude of the currently to-be-identified particle at the j-th frequency point, M is the total number of frequency points, and κ is the nonlinear amplification factor Intelligent identification of particle categories and exclusion of condensate droplets based on a machine learning model. Input the constructed vibration spectrum difference index VSI into a trained machine learning model, compare the discrimination score output by the model with the determination threshold, and determine the category of the particles. The judgment logic is as follows: where, Class particle is the particle category determination result, σ(VSI, Θ ML ) is the output classification score of the machine learning model, Θ ML is the set of machine learning model parameters, σ(·) is a function symbol, indicating "substitute VSI into the model and combine with the model parameters to obtain an output score", and H is the particle category determination threshold; If the recognition result is a condensate droplet, the condensate particles are removed from the detection data and do not participate in the calculation of the atomizer performance evaluation to ensure the authenticity and effectiveness of the final evaluation of the drug output particle concentration.

Images