A cooking oil fume generating device and method based on multi-modal fusion

Abstract

The present application relates to the technical field of kitchen air conditioner detection, and particularly relates to a cooking oil fume generating device and a generating method based on multi-modal fusion, which comprises: a multi-component oil fume generation simulation module, a multi-modal data acquisition module and an AI decision engine module. The present application can precisely reproduce the oil fume components generated by five cooking methods by means of accurate simulation and reproduction, and can precisely reproduce the transient oil fume characteristics of five Chinese cooking methods, namely, frying, frying, frying, stewing and stir-frying, to provide reliable oil fume gas for oil fume purification equipment such as kitchen air conditioners, range hoods, kitchen air purifiers and the like; a multi-modal fusion mechanism is adopted, an AI model of an improved CNN convolutional neural network model is used to monitor parameters such as the temperature, humidity and particle size distribution of oil fume gas in real time, has the ability to respond to environmental disturbances in real time, can automatically adjust PID control parameters according to external environmental changes, ensures the stability of the generated oil fume gas, and improves the identification precision of small smoke particles.

Description

Technical Field

[0001] This invention relates to the field of kitchen air conditioner testing technology, specifically to a cooking fume generating device and method based on multimodal fusion, and particularly to a cooking fume generating device and method for testing kitchen air conditioners based on multimodal fusion. Background Technology

[0002] The adoption of kitchen air conditioners is steadily increasing, with more and more families and restaurant owners recognizing the importance of installing air conditioning in their kitchens. This not only improves the working environment for chefs and other staff, increasing work efficiency, but also helps extend the lifespan of kitchen equipment. At the same time, kitchen air conditioners can effectively control the spread of cooking fumes, providing users with a more comfortable cooking and dining experience.

[0003] The main chemical components of cooking fumes generated in kitchens are alkanes, ketones, aldehydes, volatile organic compounds (VOCs), and polycyclic aromatic hydrocarbons (PAHs). An effectively controlled experimental environment is needed to simulate and analyze the fume generation process and its impact on the environment and human health. Research on cooking fume generating devices is of great significance for testing fume purification equipment such as kitchen air conditioners, range hoods, and kitchen air purifiers. The existing technology mainly achieves this research goal through two devices: (1) Commonly used commercial fume generators generally use single-channel atomization technology to atomize oil into tiny droplets to generate oil mist aerosols. However, they can only output premixed solution components and generate oil mist aerosols with fixed components. They cannot dynamically adjust the chemical composition of oil fume as needed, nor can they simulate the dynamic component changes in actual cooking, which limits the flexibility of the experiment and the depth of research; (2) Laboratory simulation systems with higher flexibility. The related fume generators often use electric heating plates to heat oil to generate oil fume, simulating the generation of oil fume in cooking. Although the system has certain simulation advantages, it is difficult to effectively control the oil fume components during the oil fume generation process, and it is impossible to accurately control the chemical composition, which affects the accuracy and repeatability of the experimental results.

[0004] Therefore, this application proposes a cooking fume generating device and method based on multimodal fusion. Summary of the Invention

[0005] Therefore, the present invention aims to provide a cooking fume generating device and method based on multimodal fusion, which features independent supply of multiple substances and deep learning dynamic control.

[0006] According to a first aspect of the present invention, a cooking fume generating device based on multimodal fusion is provided, comprising: a multi-component fume generation simulation module, used to execute PID initial control commands and PID adjustment commands, mix different fume components in proportion, adjust the particle size of atomized particles through an atomization integration module, and dynamically adjust the fume generation concentration according to test requirements; a multimodal data acquisition module, used to acquire multimodal real-time detection data during the test process and send it to an AI decision engine module, and to receive PID initial control commands and PID adjustment commands, and control the multi-component fume generation simulation module to execute PID initial control commands and / or PID adjustment commands; and an AI decision engine module, used to generate PID initial control commands based on user-inputted initial fume mist-related parameters, send the PID initial control commands to the multimodal data acquisition module, and perform fusion processing on the received multimodal real-time detection data through an AI model constructed by an improved CNN convolutional neural network model, compare preset target features, output fume state identification results and PID adjustment parameter values, generate PID adjustment commands, and send them to the multimodal data acquisition module.

[0007] Optionally, the multi-component oil fume generation simulation module includes: a five-channel atomization module, including an aldehyde channel, a polycyclic aromatic hydrocarbon channel, a vegetable oil channel, a volatile organic compound channel, and an inorganic salt suspension channel, wherein each of the aldehyde channel, polycyclic aromatic hydrocarbon channel, vegetable oil channel, volatile organic compound channel, and inorganic salt suspension channel is equipped with an independent storage tank and a peristaltic pump for separately and quantitatively delivering oil fume component precursor liquids of aldehydes, polycyclic aromatic hydrocarbons, vegetable oils, volatile organic compounds, and inorganic salt suspensions; and a conical premixing chamber for receiving and initially converging and mixing the components from the... Five precursor liquids of cooking fume components are transported through aldehyde channels, polycyclic aromatic hydrocarbon channels, vegetable oil channels, volatile organic compound channels, and inorganic salt suspension channels to obtain a premixed liquid. A static mixer with a built-in helical blade structure is connected to the output end of the conical premixing chamber to homogenize the premixed liquid to obtain a mixed liquid. An atomization integrated module is used to atomize the mixed liquid into cooking fume mist under high temperature and high pressure airflow shearing action, and to complete pyrolysis and aerosol generation under gradient temperature control to simulate the physicochemical properties of real cooking fumes.

[0008] Optionally, the atomization integrated module includes a Venturi nozzle, comprising a converging section, a throat, and a diffuser section; the converging section adopts a continuous and smooth arc transition profile with an inlet converging angle of 30°, guiding the high-temperature and high-pressure airflow from the inlet end to stably accelerate into the throat through a continuous and smooth curvature transition, and reducing flow separation and vortex losses; the throat is a cylindrical structure of equal diameter, used to ensure that the high-temperature and high-pressure airflow reaches its maximum velocity and forms a negative pressure core region in the throat, and the throat side is provided with an inlet hole, which is connected to the output end of the static mixer through a connecting pipe, for use The mixture is driven to self-intake, and under the shearing action of high temperature and high pressure airflow, the mixture is broken down to 0.1-10μm to form oil mist droplets; the diffusion section is a gradually expanding diffusion structure with a diffusion angle of 20°, and the diffusion section has a built-in cyclone separator for further breaking down and homogenizing the oil mist droplets to form oil mist aerosol; the high pressure air source is used to deliver room temperature high pressure airflow; the cylindrical heater is arranged between the high pressure air source and the Venturi nozzle to heat the room temperature high pressure airflow to a preset temperature to form a high temperature high pressure airflow, which is then injected into the converging section of the Venturi nozzle.

[0009] Optionally, the multimodal data acquisition module includes: a temperature sensor, a dew point sensor, a pressure sensor, a flow rate sensor, a laser scattering sensor, a PID-VOC sensor, an infrared spectrometer, an RGB industrial camera, an infrared thermal imaging sensor, and a control module; the probe structures of the temperature sensor, pressure sensor, and flow rate sensor are all embedded inside the throat for real-time acquisition of real-time temperature data, real-time airflow pressure data, and real-time flow rate data of the high-temperature, high-pressure airflow; the dew point sensor is disposed on the outer wall of the diffuser section for acquiring dew point temperature data of the mist at the diffuser section wall; the laser scattering sensor, PID-VOC sensor, infrared spectrometer, RGB industrial camera, and infrared thermal imaging sensor are all integrated and deployed downstream of the Venturi nozzle outlet. The pipeline section includes a laser scattering sensor for collecting particulate matter distribution data of oil fume mist, a PID-VOC sensor for testing the oil fume components of aldehydes and PAHs, an infrared spectrometer for identifying characteristic functional groups of organic matter in oil fume mist, an RGB industrial camera for acquiring RGB images of oil fume mist, and an infrared thermal imaging sensor for acquiring infrared thermal imaging data of oil fume mist. The control module includes a touchscreen, a PLC controller, and an AD conversion module. The PLC controller is electrically connected to a five-channel atomization module, a static mixer, a high-pressure gas source, a cylindrical heater, a multimodal data acquisition module, and the touchscreen via the AD conversion module. The PLC controller is connected to an AI decision engine module via wired or wireless means.

[0010] Optionally, an annular Peltier condenser is also provided on the outer wall of the diffusion section. The installation position of the cooling end face of the annular Peltier condenser is aligned with the installation position of the probe sensing end face of the dew point sensor on the outer wall of the diffusion section to form a humidity closed-loop control circuit.

[0011] Optionally, the AI ​​decision engine module includes: an initial control command generation module, used to generate a PID initial control command based on the initial oil fume mist related parameters input by the user; an adjustment command generation module, used to fuse the received multimodal real-time detection data through an AI model constructed by an improved CNN convolutional neural network model, compare preset target features, output the oil fume state recognition result and PID adjustment parameter value, and generate a PID adjustment command; and a command sending module, used to send the PID initial control command and / or PID adjustment command to the PLC controller, and execute the PID initial control command and / or PID adjustment command through the PLC controller.

[0012] Optionally, the adjustment instruction generation module includes: a data preprocessing module, used to preprocess the RGB image of the oil fume and the infrared thermal imaging data of the oil fume to obtain RGB image preprocessed data and infrared thermal imaging preprocessed data; a data enhancement module, used to randomly rotate the RGB image preprocessed data and add brightness jitter to the rotated RGB image preprocessed data to obtain RGB image enhanced data, and inject Gaussian noise into the infrared thermal imaging preprocessed data to obtain infrared thermal imaging enhanced data; a feature extraction module, used to extract the spatial texture features of the RGB image enhanced data through the k-th layer convolutional network to capture the morphological distribution of oil fume, and to extract the temperature gradient features of the infrared thermal imaging enhanced data through the l-th layer convolutional network to reflect the trend of hot airflow movement; and an attention mechanism introduction module, used to introduce the extracted... Temperature gradient features are projected into a low-dimensional space, and the projections of query vectors and key vectors are calculated to construct a dynamic temperature similarity matrix of infrared temperature of oil fume mist. Temperature gradient features are then injected into the standard dot product attention to enhance the response of the oil fume mist flow region. A cross-modal attention fusion module is used to enhance the model's motion mechanism by normalizing attention weights, generating a spatial attention map, identifying key high-temperature regions, and outputting weighted value vectors. Based on the attention weights, the fusion features enhance the dynamic oil fume mist region, completing the multimodal feature fusion. A state recognition and decision output module is used to input the fused multimodal features into the decision output layer, output the oil fume state recognition result, and determine whether the oil fume state recognition result meets the set requirements. If the oil fume state recognition result does not meet the set requirements, PID adjustment parameters are output, and PID adjustment instructions are generated.

[0013] According to a second aspect of the present invention, a method for generating cooking fumes based on multimodal fusion is provided, comprising the following steps: executing PID initial control commands and PID adjustment commands through a multi-component fume generation simulation module, mixing different fume components in proportion, adjusting the atomization particle size through an atomization integration module, and dynamically adjusting the fume generation concentration according to test requirements; acquiring multimodal real-time detection data during the test process through a multimodal data acquisition module and sending it to an AI decision engine module, and receiving PID initial control commands and PID adjustment commands, controlling the multi-component fume generation simulation module to execute PID initial control commands and / or PID adjustment commands; generating PID initial control commands through the AI ​​decision engine module based on user-inputted initial fume mist-related parameters, and sending the PID initial control commands to the multimodal data acquisition module; and performing fusion processing on the received multimodal real-time detection data through an AI model constructed by an improved CNN convolutional neural network model, comparing preset target features, outputting fume state identification results and PID adjustment parameter values, generating PID adjustment commands and sending them to the multimodal data acquisition module.

[0014] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method.

[0015] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method.

[0016] The present invention provides a cooking fume generating device and method based on multimodal fusion, which has the following beneficial effects: (1) By accurately simulating and reproducing the components of oil fumes generated by five cooking methods, it can accurately reproduce the transient oil fume characteristics of five types of Chinese cooking methods such as stir-frying, pan-frying, deep-frying, stewing, and stir-frying, providing reliable oil fume mist for oil fume purification equipment such as kitchen air conditioners, range hoods, and kitchen air purifiers; (2) By adopting a multimodal fusion mechanism, through an AI model with an improved CNN convolutional neural network model that introduces an attention mechanism, it can monitor parameters such as temperature, humidity, and particle size distribution of oil fume mist in real time, and has the ability to respond to environmental disturbances in real time. It can adjust the oil fume mist according to the external environment. (2) Automatically adjust PID control parameters according to environmental changes to ensure the stability of generated oil fume mist and improve the accuracy of fine smoke particle identification; (3) Combine traditional control theory with deep learning to realize dynamic adjustment of equipment PID and online optimization of parameters; (4) Use artificial intelligence technology to build a model of complex nonlinear generation relationship, and obtain the input material scheme of cooking oil fume generator according to the oil fume composition and particle size distribution, which can meet the customized needs of different test conditions; (5) Through AI model calculation and analysis, the solution ratio scheme can be given according to the temperature, humidity and particle size distribution parameters of oil fume mist set by the user, and the solution ratio can be dynamically adjusted during the generation process. Attached Figure Description

[0017] Figure 1 This is a structural block diagram of a cooking fume generator based on multimodal fusion according to an embodiment of the present invention.

[0018] Figure 2 This is a structural block diagram of the AI ​​decision engine module in an embodiment of the present invention.

[0019] Figure 3 This is a structural block diagram of the adjustment instruction generation module according to an embodiment of the present invention.

[0020] Figure 4 This is a flowchart of a cooking fume generation method based on multimodal fusion according to an embodiment of the present invention.

[0021] Figure 5 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0022] The embodiments of this application will now be described in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. Furthermore, the following embodiments and features can be combined with each other unless otherwise specified. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0023] Example 1

[0024] Figure 1 This is a structural block diagram of a cooking fume generator based on multimodal fusion provided in Embodiment 1 of the present invention. (Refer to the attached diagram.) Figure 1 The cooking fume generator includes: a multi-component fume generation simulation module, a multi-modal data acquisition module, and an AI decision engine module.

[0025] In some optional implementations of this application, a multi-component oil fume generation simulation module is used to execute PID initial control commands and PID adjustment commands, mix different oil fume components in proportion, adjust the atomization particle size through an atomization integration module, and dynamically adjust the oil fume generation concentration according to test requirements.

[0026] In some optional implementations of this application, the multi-component oil fume generation simulation module includes: a five-channel atomization module, including an aldehyde channel, a polycyclic aromatic hydrocarbon channel, a vegetable oil channel, a volatile organic compound channel, and an inorganic salt suspension channel, wherein each of the aldehyde channel, polycyclic aromatic hydrocarbon channel, vegetable oil channel, volatile organic compound channel, and inorganic salt suspension channel is equipped with an independent storage tank and a peristaltic pump for separately and quantitatively delivering the oil fume component precursor liquids of aldehydes, polycyclic aromatic hydrocarbons, vegetable oils, volatile organic compounds, and inorganic salt suspensions; and a conical premixing chamber for receiving and initially mixing the premixed oil fume components. The process involves a confluence mixing unit that delivers five precursor liquids of cooking fumes via aldehyde channels, polycyclic aromatic hydrocarbon channels, vegetable oil channels, volatile organic compound channels, and inorganic salt suspension channels to obtain a premixed liquid. A static mixer, equipped with a built-in helical blade structure, is connected to the output end of the conical premixing chamber to homogenize the premixed liquid, resulting in a mixed liquid. An integrated atomization module is used to atomize the mixed liquid into cooking fume aerosols under high-temperature and high-pressure airflow shearing, and to complete pyrolysis and aerosol generation under gradient temperature control, simulating the physicochemical properties of real cooking fumes.

[0027] In some optional implementations of this application, the five-channel atomization module involved in this application is divided into an aldehyde channel, a polycyclic aromatic hydrocarbon (PAHs) channel, a vegetable oil channel, a volatile organic compound (VOCs) channel, and an inorganic salt suspension channel. Each channel is equipped with an independent solution storage tank and a peristaltic pump. The independent control of the five channels can precisely adjust the mixing ratio of different substances.

[0028] In some optional implementations of this application, the peristaltic pump uses a multi-channel planetary roller pump head, with a flow rate adjustment range of 0.01-10 L / min, a flow rate accuracy of 0.01 mL / min, a speed accuracy of ±0.5% (matching PLC controller instructions), an IP65 protection rating (protection against oil fumes and water vapor intrusion), support for external analog input (0-10V to receive AI instructions), and stepper motor microstepping drive (16000 steps / revolution) to achieve 0.001 mL / min fine adjustment. It features a flow adaptive mechanism and viscosity compensation algorithm, dynamically increasing the speed in the vegetable oil channel (speed compensation +5% for every 10 cP increase in viscosity); pulsation suppression technology, with an 8-roller design and a hose damper, suppresses VOCs channel pulsation to <2%.

[0029] In some optional implementations of this application, the aldehyde solution channel involved in this application corresponds to solutions of aldehyde and ketone compounds such as acrolein, glutaraldehyde, and nonanal, which can simulate aldehydes produced by the oxidative decomposition of oils during high-temperature cooking (such as stir-frying and deep-frying). An independent solution storage tank and peristaltic pump are configured, with a flow rate adjustment range of 0.01-10 L / min, a flow rate accuracy of 0.01 mL / min, and an error of ±0.5%; corrosion-resistant fluororubber hoses are used to transport the liquid.

[0030] In some optional implementations of this application, the polycyclic aromatic hydrocarbon (PAH) channels involved in this application correspond to organic solvents (such as xylene solution) containing PAHs such as benzo[a]pyrene, naphthalene, and anthracene. This allows for the reproduction of highly carcinogenic PAHs generated during high-temperature carbonization processes such as grilling and stir-frying.

[0031] In some optional implementations of this application, the vegetable oil channel involved in this application corresponds to refined soybean oil, corn oil, or rapeseed oil (simulating actual cooking oil), generating oil fume particulate matter (PM2.5). 2.5 It serves as the core carrier and adsorption medium for other pollutants; it uses a platinum-cured silicone tubing to resist grease adsorption.

[0032] In some optional implementations of this application, the volatile organic compound (VOCs) channel involved in this application corresponds to a mixed solution of volatile organic compounds (VOCs) such as toluene, xylene, and acetone, simulating gaseous organic compounds in cooking fumes. These substances are precursors to ozone and secondary particulate matter.

[0033] In some optional implementations of this application, the inorganic salt suspension channel involved in this application corresponds to a suspension of inorganic salt particles such as sodium chloride and sulfate, reproducing the approximately 12% inorganic salt particles (such as salt volatilized from heated food) in cooking fumes, and forming aggregated particles with other components; a reinforced TPE flexible hose with a wall thickness of 1.6 mm is used to resist particle abrasion. The channel is equipped with an ultrasonic oscillator to prevent inorganic salt sedimentation.

[0034] In some alternative implementations of this application, the conical premixing chamber involved in this application is made of 316L stainless steel coated with tungsten carbide, and initially mixes five oil fume component precursor liquids, which are connected to a static mixer.

[0035] In some alternative implementations of this application, the static mixer involved in this application is made of 316L stainless steel coated with tungsten carbide and has a built-in helical blade structure to enhance the mixing effect at low flow rates and prevent the solution from separating due to density differences.

[0036] In some optional implementations of this application, the atomizing integrated module includes: a Venturi nozzle, comprising a contraction section, a throat, and a diffuser section; the contraction section adopts a continuous and smooth arc transition profile, with an inlet contraction angle of 30°, guiding the high-temperature and high-pressure airflow from the inlet end to stably accelerate into the throat through the continuous and smooth curvature transition, and reducing flow separation and vortex losses; the throat is a cylindrical structure of equal diameter, used to ensure that the high-temperature and high-pressure airflow reaches the maximum flow velocity in the throat and forms a negative pressure core area; the throat has an inlet hole on its side, and the inlet hole is connected to the static mixer through a connecting pipe. The outlet connection is used to drive the mixture to self-inhale, and under the shearing action of high temperature and high pressure airflow, the mixture is broken down to 0.1-10μm to form oil mist droplets; the diffusion section is a gradually expanding diffusion structure with a diffusion angle of 20°. The diffusion section has a built-in cyclone separator for further breaking down and homogenizing the oil mist droplets to form oil mist aerosol; the high-pressure air source is used to deliver room temperature high-pressure airflow; the cylindrical heater is arranged between the high-pressure air source and the Venturi nozzle to heat the room temperature high-pressure airflow to a preset temperature to form a high temperature high-pressure airflow, which is then injected into the converging section of the Venturi nozzle.

[0037] In some optional implementations of this application, the Venturi nozzle involved in this application consists of three independently machined high-precision components: a converging section, a throat, and a diffuser section. The entire assembly is made of Inconel 625 high-temperature alloy, and the inner wall is coated with a 50±5μm thick tungsten carbide wear-resistant and corrosion-resistant coating. This coating can withstand corrosion from aldehydes and PAHs, and also reduces the wear of the nozzle by inorganic salt particles. Its surface is mirror-polished, resulting in a wall roughness Ra<0.2μm, which significantly reduces flow resistance and carbon buildup tendency. The Venturi nozzle meets the requirements of ISO 5167-3:2022 standard, with a nominal pressure of 4MPa, and is suitable for Reynolds numbers of 1.5×10⁻⁶. 5 ~2×10 6 The high-speed airflow conditions within the range can ensure the turbulent atomization effect of the oil fume mist.

[0038] In some optional implementations of this application, the contraction section employs a continuous and smooth curvature transition structure with an inlet contraction angle of 30°, guiding the high-temperature, high-pressure airflow to smoothly accelerate into the throat. This structure effectively suppresses boundary layer separation and vortex shedding, and CFD simulations show that the total pressure loss coefficient is less than 0.08. Its length-to-inlet diameter ratio (L / D) is optimized to 0.6 (e.g., 7.2 mm length when the inlet diameter is 12 mm), balancing acceleration efficiency and pressure drop control.

[0039] In some optional implementations of this application, the throat is a high-precision, uniform-diameter cylindrical structure with an inner diameter manufacturing tolerance controlled within ±0.02 mm. Its length is 1.8 times the throat diameter, ensuring that the high-temperature, high-pressure airflow reaches its maximum velocity and maintains strong turbulence in this region. The high-temperature, high-pressure airflow generates a significant negative pressure in the throat (measured to over -80 kPa), forming a stable and repeatable self-inhalation drive zone. Its sidewall has a 4 mm diameter inlet hole, which is connected to the static mixer via an Inconel 625 alloy tube coated with tungsten carbide. Preferably, the inlet hole is precisely located 1.5 mm downstream of the lowest point of negative pressure to maximize the mixture intake efficiency. When the heated high-speed airflow passes through the throat, a local vacuum is generated. This negative pressure draws the mixture from the static mixer into the main vapor stream (i.e., the high-temperature, high-pressure airflow). The high-speed shearing of the high-temperature, high-pressure air breaks the mixture to 0.1–10 μm, forming oil mist droplets, which are further homogenized by the built-in cyclone separator in the diffuser section. In addition, a flow rate sensor is integrated inside the throat, which can monitor the flow rate of the high-temperature and high-pressure airflow inside the throat in real time, and dynamically adjust the heater power of the cylindrical heater through the PLC controller to ensure that the temperature fluctuation is less than ±0.1m / s, thereby realizing closed-loop control of the high-temperature and high-pressure airflow rate.

[0040] In some optional implementations of this application, the diffuser section is a gradually expanding diffuser structure with a diffusion angle of 20°, within the optimal diffusion angle range (18-22°), effectively restoring static pressure while suppressing airflow separation and recirculation vortex formation. The built-in axial vortex consists of six circumferentially distributed guide vanes at a 15° installation angle, inducing a moderate circumferential velocity component and extending the gas-liquid two-phase contact time to over 0.15s, promoting further droplet breakup and homogenization. The outlet Mach number is controlled below 0.3 to avoid shock wave disturbances affecting the downstream aging process. An annular Peltier condenser and a high-precision dew point sensor are integrated into the outer wall, enabling real-time monitoring of the humidity of the oil fume mist. The power of the Peltier condenser is dynamically adjusted via a PLC controller to ensure humidity fluctuations are less than ±3%RH. After aerodynamic optimization, the overall structure achieves a measured total pressure loss coefficient of 0.14, meeting the high-efficiency and low-consumption design goals.

[0041] In some optional implementations of this application, the Venturi nozzle is connected to a high-pressure air source, and a cylindrical heater is installed between the outlet of the high-pressure air source and the Venturi nozzle (i.e., the contraction section), forming an integrated "heating-atomization" structure. An internal resistance wire or ceramic heating element is arranged, and an external insulation layer is used to reduce heat loss, heating the room-temperature compressed air to a set temperature adjustable from 30-450℃ with an accuracy of ±1℃. Furthermore, four high-temperature platinum resistance thermocouples are evenly arranged on the surface of the cylindrical heater, and the signals are fed back to the PLC controller. Combined with the real-time temperature data of the high-temperature, high-pressure airflow within the throat, feedforward compensation is performed (specifically, the PLC controller dynamically adjusts the power of the cylindrical heater) to ensure that the outlet air temperature fluctuation is less than ±1℃.

[0042] In some optional implementations of this application, the multimodal data acquisition module includes: a temperature sensor, a dew point sensor, a pressure sensor, a flow rate sensor, a laser scattering sensor, a PID-VOC sensor, an infrared spectrometer, an RGB industrial camera, an infrared thermal imaging sensor, and a control module; the probe structures of the temperature sensor, pressure sensor, and flow rate sensor are all embedded inside the throat for real-time acquisition of real-time temperature data, real-time airflow pressure data, and real-time flow rate data of the high-temperature, high-pressure airflow; the dew point sensor is disposed on the outer wall of the diffuser section for acquiring dew point temperature data of the mist at the wall of the diffuser section; the laser scattering sensor, PID-VOC sensor, infrared spectrometer, RGB industrial camera, and infrared thermal imaging sensor are all integrated and deployed in the Venturi nozzle. In the downstream straight pipe section, the laser scattering sensor is used to collect particulate matter distribution data of the oil fume mist; the PID-VOC sensor is used to test the oil fume components of aldehydes and PAHs; the infrared spectrometer is used to identify the characteristic functional groups of organic matter in the oil fume mist; the RGB industrial camera is used to collect RGB images of the oil fume mist; and the infrared thermal imaging sensor is used to collect infrared thermal imaging data of the oil fume mist. The control module includes a touch screen, a PLC controller, and an AD conversion module. The PLC controller is electrically connected to the five-channel atomization module, static mixer, high-pressure gas source, cylindrical heater, multimodal data acquisition module, and touch screen through the AD conversion module. The PLC controller is connected to the AI ​​decision engine module via wired or wireless means.

[0043] In some optional implementations of this application, an annular Peltier condenser is also provided on the outer wall of the diffusion section. The installation position of the cooling end face of the annular Peltier condenser is aligned with the installation position of the probe sensing end face of the dew point sensor on the outer wall of the diffusion section to form a humidity closed-loop control circuit.

[0044] In some optional implementations of this application, the velocity, pressure, and temperature sensors are embedded in the throat region of the Venturi nozzle via a multi-parameter integrated probe structure. The probes are mounted at 120° intervals along the circumference of the throat at the front section, 1.5 mm from the end of the contraction section. This location is at the lowest negative pressure point determined by CFD simulation and in the core region where the airflow is most stable, thus avoiding the impact of boundary layer disturbances on measurement accuracy. A miniature thermal velocity sensor employs MEMS thin-film thermistor technology, directly exposed to the airflow, with a measurement range of 0-150 m / s and a response time <10 ms, reflecting the dynamic changes of the high-temperature, high-pressure airflow (i.e., the main airflow) in real time. A piezoresistive pressure sensor is connected to the annular pressure equalization chamber in the throat via four sidewall pressure-sensing holes, achieving four-point average pressure difference measurement to eliminate instantaneous fluctuations caused by local vortices. Its range is 0-1 MPa, with an accuracy of ±0.1%FS. The pressure-sensing holes have a diameter of 1 mm, an axial length of 5 mm, and are tilted at 15° towards the incoming flow direction to reduce the risk of particle blockage. The platinum resistance temperature sensor (Pt100) is encapsulated in a ceramic protective sleeve, with the insertion depth controlled at 1 / 3 of the throat's inner diameter (approximately 2 mm, corresponding to a 6 mm throat diameter) to avoid significant disturbance to the main field. The temperature measurement range is -50 to 500℃, with a resolution of 0.1℃. All sensing element signal lines are led out via Inconel 625 alloy sealed connectors, featuring IP65 protection and double-layer shielding to effectively suppress electromagnetic interference and thermal noise in high-temperature environments, ensuring long-term operational stability.

[0045] In some optional implementations of this application, the dew point sensor employs a high-precision capacitive humidity probe, installed close to the middle section of the outer wall of the diffuser section, with its axial position located at 50% of the total length of the diffuser section (e.g., 30 mm from the throat outlet when the diffuser section is 60 mm long). This area is where the airflow has completed initial atomization and begun to cool, making it a critical region for the initiation of water vapor condensation. The sensor base is fully bonded to the metal wall surface using thermally conductive silicone with a thermal conductivity of 3.5 W / (m·K), ensuring rapid heat transfer from the wall surface to the sensing end. A miniature heat-insulating shield is also added to block external radiant heat (such as heater heat radiation) and prevent measurement deviation. The probe's sensing end faces the inner side of the wall surface, utilizing the principle of heat conduction to monitor the dew point temperature of the vapor near the wall surface in real time, achieving a measurement accuracy of ±0.5℃ and a response time of <3 s. Its installation position is aligned with the cooling surface of the Peltier condenser integrated into the outer wall of the diffuser section, forming a closed-loop humidity control circuit. This accurately captures the initial condensation point during programmed cooling, providing critical boundary conditions for simulating oil fume aging.

[0046] In some optional implementations of this application, the laser scattering sensor and the PID-VOC sensor are uniformly installed in a straight pipe section 100-150 mm downstream of the Venturi nozzle outlet. This location is after the diffuser outlet, where the airflow tends to stabilize after rectification, and the turbulence intensity is significantly reduced (measured TKE < 0.5 m). 2 / s 2 To avoid interference from strong vortices on optical measurements and gas sampling, the laser scattering sensor measures particle size in the range of 0.1-10 μm, with a minimum resolution particle size of 0.01 μm and a response time of less than 1 second. It can be used with a pump-type sensor with PID flow control (such as a cleanroom particle counting sensor), with an error ≤ ±3%, for collecting particulate matter distribution data in oil fume mist. It has an IP65 protection rating and an anti-grease coating. The PID-VOC sensor probe is inserted axially into the pipe at 1 / 4 radius (i.e., 1 / 4 R from the pipe wall) through a stainless steel sampling nozzle. This position balances representativeness and low risk of particle deposition. A bypass flow controller maintains a constant flow of 1 L / min for continuous online detection. The PID-VOC sensor can test for aldehydes and PAHs in oil fumes. Low range (0-500 ppm): for detecting trace aldehyde leaks (such as PID-AH); high range (0-2000 ppm): suitable for PAHs and mixed VOCs. It uses a stainless steel chamber with an anti-grease film to prevent grease from clogging the gas path.

[0047] In some optional implementations of this application, the infrared spectrometer (i.e., Fourier transform infrared spectrometer, FT-IR) has its sampling point located 120 mm downstream of the Venturi nozzle outlet in a straight tube section, coplanar with the laser scattering sensor but circumferentially offset by 90° to avoid spatial interference and airflow obstruction. The sampling tube employs a double-layer heating structure: the inner tube is made of fused silica, and the outer layer is a tungsten carbide-coated Inconel 625 alloy, heated to 180°C to prevent high-boiling-point organic compounds (such as PAHs) from condensing and precipitating on the tube wall. The sampling flow rate is controlled collaboratively by the sonic nozzle and a proportional valve, dynamically adjusted based on the measured main airflow velocity (from a flow velocity sensor at the throat) to achieve constant-velocity sampling (deviation controlled within ±4%). For example, when the main airflow velocity is 85 m / s, the control module adjusts the bypass pump to maintain a consistent sampling airflow velocity, avoiding particle classification deviations due to velocity mismatch. The spectral range of the infrared spectrometer covers the CH bond (2900 cm⁻¹). -1 ), C=O bond (1700cm) -1 Characteristic peaks of organic matter in oil fume, etc. Resolution ≤ 4 cm⁻¹ -1 (Fourier transform type FT-IR) to distinguish structurally similar PAHs. Scanning speed ≥ 1 scan / second to match the transient diffusion characteristics of oil fume. A transmission + ATR (attenuated total reflection) composite probe is selected, which can withstand oil fume temperature of 200℃.

[0048] In some optional implementations of this application, the RGB industrial camera and infrared thermal imaging sensor are positioned directly opposite the oil fume / mist outlet, with an installation distance of 500 mm. The field of view completely covers the entire evolution region from the nozzle outlet to the initial diffusion zone (axial 0-300 mm). The RGB industrial camera employs a global shutter CMOS sensor with a resolution of 1920×1080 and a frame rate of 50 fps. It is equipped with a 50 mm equivalent focal length fixed-focus lens and a 45° field of view, and uses a backlit LED array (wavelength 630 nm) to achieve high-contrast imaging, used to capture the morphology, boundary sharpness, and aggregation behavior of the fog. Infrared thermal imaging sensors (such as the testo890) can be simultaneously deployed on the same optical platform, with their optical axes parallel to the RGB camera and a spacing of approximately 150 mm to avoid obstruction. They feature a resolution of 1280×960, autofocus, 1-8x zoom, a spectral range of 7.5μm-14μm, a temperature measurement range of -30-650℃, and a thermal sensitivity of <0.04℃, enabling real-time acquisition of the three-dimensional temperature field distribution of the fog flow. The two devices synchronize data acquisition via hardware triggering, with timestamp alignment accuracy ≤10 ms. Image data, after non-local mean denoising and SIFT feature registration, is input into the AI ​​model to achieve morphology-temperature coupled feature extraction and dynamic evaluation of fogging quality.

[0049] In some optional implementations of this application, the control module consists of a touch screen, a PLC controller, and an AD conversion module. The PLC controller is electrically connected to the five-channel atomizing module, static mixer, high-pressure gas source, cylindrical heater, multimodal data acquisition module, and touch screen through the AD conversion module. The PLC controller can be connected to the AI ​​decision engine module via serial port, Ethernet, USB, wireless, etc. The touch screen is connected to the PLC controller, and the PLC controller can be controlled through the touch screen. The control module can perform the following functions: (1) control the flow rate of the peristaltic pump and the power of the ultrasonic oscillator of each channel according to the PID initial control command; (2) control the flow rate, flow rate, and power of the high-temperature and high-pressure airflow at the Venturi nozzle and the heater according to the PID initial control command, and adjust the pollutant particle size distribution range, humidity, and temperature; (3) control the touch screen; (4) collect and summarize the data of each sensor and transmit the data to the AI ​​decision engine module; (5) perform corresponding control adjustments according to the PID adjustment command issued by the AI ​​decision engine module.

[0050] In some optional implementations of this application, see Appendix Figure 2The AI ​​decision engine module includes: an initial control command generation module, used to generate PID initial control commands based on the initial oil fume mist parameters input by the user; an adjustment command generation module, used to fuse the received multimodal real-time detection data through an AI model constructed by an improved CNN convolutional neural network model, compare preset target features, output the oil fume state recognition result and PID adjustment parameter value, and generate PID adjustment commands; and a command sending module, used to send the PID initial control commands and / or PID adjustment commands to the PLC controller, and execute the PID initial control commands and / or PID adjustment commands through the PLC controller.

[0051] In some optional implementations of this application, the AI ​​decision engine module uses an Intel i9-13900K processor and an NVIDIA RTX 6000 Ada GPU, connects to an AI model, generates PID initial control instructions based on the user-inputted initial oil fume and mist parameters, and sends them to the PLC controller. It also compares the received multimodal real-time detection data from various sensors with preset target features, outputs the oil fume state identification result and PID adjustment parameter values, generates PID adjustment instructions, and sends these instructions to the PLC controller. The PLC controller then adjusts the control of the corresponding modules according to the PID adjustment instructions.

[0052] In some optional implementations of this application, see Appendix Figure 3The adjustment instruction generation module includes: a data preprocessing module, used to preprocess the RGB image and infrared thermal imaging data of the oil fume mist to obtain RGB image preprocessed data and infrared thermal imaging preprocessed data; a data enhancement module, used to randomly rotate the RGB image preprocessed data and add brightness jitter to the rotated RGB image preprocessed data to obtain RGB image enhanced data, and inject Gaussian noise into the infrared thermal imaging preprocessed data to obtain infrared thermal imaging enhanced data; a feature extraction module, used to extract the spatial texture features of the RGB image enhanced data through the k-th layer convolutional network to capture the morphological distribution of oil fume mist, and to extract the temperature gradient features of the infrared thermal imaging enhanced data through the l-th layer convolutional network to reflect the trend of hot airflow movement; and an attention mechanism introduction module, used to introduce the extracted temperature gradient features into the infrared thermal imaging enhanced data. The gradient features are projected onto a low-dimensional space, and the projections of the query vector and key vector are calculated to construct a dynamic temperature similarity matrix of the infrared temperature of oil fume mist. Temperature gradient features are then injected into the standard dot product attention to enhance the response of the oil fume mist flow region. The cross-modal attention fusion module enhances the model's motion mechanism by normalizing attention weights, generates a spatial attention map, identifies key high-temperature regions, and outputs a weighted value vector. Based on the attention weights, the fusion features enhance the dynamic oil fume mist region, completing the multimodal feature fusion. The state recognition and decision output module inputs the fused multimodal features into the decision output layer, outputs the oil fume state recognition result, and determines whether the oil fume state recognition result meets the set requirements. If the oil fume state recognition result does not meet the set requirements, PID adjustment parameters are output, and PID adjustment instructions are generated.

[0053] In some optional implementations of this application, the AI ​​model involved in this application is constructed from an improved CNN convolutional neural network model, and the specific process of generating PID adjustment instructions is as follows.

[0054] The first step is to perform dual-branch data preprocessing using the data preprocessing module.

[0055] The RGB image of the oil fume mist is normalized according to the following formula (1): In the formula, This is a normalized RGB image of the cooking fumes. This is the original RGB image of the cooking fumes (size H×W×3), where H is the height, W is the width, 3 is the number of channels corresponding to the R, G, and B components, and (x, y) are the pixel coordinates. The average of the three RGB channels of the training set can eliminate the lighting bias of RGB images; The standard deviation of the RGB values ​​in the training set can eliminate contrast differences.

[0056] The infrared thermal imaging data of oil fume mist is matrixed according to the following formula (2): In the formula, A matrixed infrared thermal image of cooking fumes; The original infrared temperature matrix (size H×W×1) of the oil fume mist, a single-channel thermal map; , These are the maximum and minimum values ​​of the temperature measurement range of the infrared spectrometer, normalized to the [0, 1] interval, and aligned with the RGB image size of the oil fume mist.

[0057] The second step is to perform data augmentation (training phase) using the data augmentation module.

[0058] The RGB image preprocessing data is randomly rotated according to the following formula (3), and brightness jitter is added to the rotated image data according to the following formula (4) to improve the robustness of the model to changes in viewing angle and illumination fluctuations, and to prevent overfitting. Formula (3) is: Formula (4) is: In the formula, These are the normalized RGB image pixels after random rotation; The rotation angle is random, ranging from -15° to 15°; To add brightness jitter and random rotation to a normalized RGB image; This is a normalized RGB image after random rotation; Added brightness jitter.

[0059] Gaussian noise is injected into the original infrared temperature of the oil fume mist according to the following formula (5) to simulate the infrared sensor noise and improve the model's anti-interference ability. Formula (5) is: In the formula, Infrared temperature of oil fume mist after adding noise; The original infrared temperature of the oil fume mist; This is temperature noise, which follows a Gaussian distribution.

[0060] The third step is to perform dual-branch feature extraction using the feature extraction module.

[0061] The spatial texture features of the RGB image enhancement data are extracted using the k-th layer convolutional network according to the following formula (6) to capture the morphological distribution of oil fumes. Formula (6) is: In the formula, Output the value of the feature map at the (x, y) position for the spatial texture; The RGB image is used as the convolution input with index (i, j); The weights are the first convolutional kernel weights, the number of input channels is 3, the number of output channels is 32, and the convolutional size is 7×7. This can capture the local texture and morphological features of oil fume particles. The first bias vector has a length of 32, which can enhance the nonlinear expressive power of the model; the activation function ReLU(x)=max(0,z) can solve the gradient vanishing problem; z is the sum of the weighted input and bias received by the neuron.

[0062] The temperature gradient features of the infrared thermal imaging enhancement data are extracted through the l-th layer convolutional network according to the following formula (7) to reflect the trend of thermal airflow. Formula (7) is: In the formula, Output the value of the feature map at the (x, y) position for the temperature gradient; Infrared imaging convolution input indexed by (m, n); The weights are the second convolution kernel weights. The number of input channels is 1, the number of output channels is 32, and the convolution size is 5×5. This can capture the local texture and morphological features of oil fume particles. The second bias vector, with a length of 32, enhances the model's nonlinear expressive power.

[0063] The fourth step is to introduce the attention mechanism through the attention mechanism introduction module.

[0064] The projections of the query vector Q and the key vector K are calculated according to the following formula (8), projecting the temperature gradient features into a low-dimensional space to reduce computation. Formula (8) is: In the formula, The feature map output for the RGB branch; The output feature map of the infrared branch; It is a 1×1 convolution used for channel transformation and dimensionality reduction of feature maps.

[0065] The dynamic temperature similarity matrix of the infrared temperature of oil fume mist is generated according to the following formula (9), and temperature gradient features are injected into the standard dot product attention to enhance the response of the oil fume mist flow region. Formula (9) is: In the formula, S is the dynamic temperature similarity matrix of the infrared temperature of oil fume mist; T is the temperature of oil fume mist. This is the motion enhancement factor, with a value of 0.3, used to weight temperature change information; The temperature gradient matrix is ​​calculated according to the following formula (10), which can discretize the approximate temperature gradient and quantify the direction of heat conduction. Formula (10) is: In the formula, The value represents the direction of heat conduction at the position (x, y); The temperature gradient convolution input is indexed to (i, j); It is a 3×3 learnable convolution with a size of (3×3×2×32), inputting a 2-channel gradient and outputting a 32-channel gradient. This is the bias vector.

[0066] The fifth step is to perform cross-modal attention fusion through the cross-modal attention fusion module.

[0067] The attention weights are normalized according to the following formula (11) to enhance the model's motion mechanism and output a spatial attention map (with dimensions of H×W×H×W) to identify key high-temperature regions. Formula (11) is: In the formula, A is the normalized attention weight matrix; S is the dynamic temperature similarity matrix of infrared temperature of oil fume mist. Let be the element in the i-th row and j-th column of the attention weight matrix A; Let be the element in the i-th row and j-th column of the dynamic temperature similarity matrix S of the infrared temperature of oil fume mist.

[0068] The weighted output of the value vector V is calculated according to the following formula (12), and the dynamic oil fume mist region is enhanced by fusing features based on attention weights. Formula (12) is: In the formula, V is the final output feature map after attention weighting; V is the value vector matrix.

[0069] The sixth step is to output the PID adjustment parameters through the state recognition and decision output module.

[0070] The probability of oil fume mist is output according to the following formula (13), and it is determined whether the state of oil fume mist meets the set requirements (i.e., the preset target characteristics). Formula (13) is: In the formula, This represents the probability predicted by the model that "the oil fume mist meets the set requirements"; The weight matrix (size L×1) is used to learn the key features of oil fume mist; The vector after fusing features ( In the formula The height of the feature map; The width of the feature map; (Number of fused feature channels); This is the bias vector.

[0071] The PID tuning parameters are output according to the following formula (14). Formula (14) is: In the formula, This is the vector for adjusting the proportional gain. The integral gain adjustment vector; This is the vector for the differential gain adjustment. The weight matrix (size L×3) outputs the adjustment values ​​of the three PID parameters. To fuse feature vectors; It is the bias vector; This is a scaling factor that controls the adjustment range and prevents oscillations. It is the z-th power of the natural constant e; The natural constant e raised to the power of -z, where z is... The results are obtained. The PID adjustment command is generated using the output PID three-parameter adjustment values.

[0072] In some optional implementations of this application, the specific application scenario of a cooking fume generator based on multimodal fusion is used as an example. For example: (1) Simulate the characteristics of a surge in aldehydes at high temperature when stir-frying vegetables. The specific implementation steps include: a. Initial stage: The flow rate of the vegetable oil channel (soybean oil) is set to 2.5L / min, and the flow rate of the aldehyde channel (acrylaldehyde solution) is 0.8L / min; b. Stir-frying heating stage (5th second): The AI ​​decision engine module detects the feedback from the temperature sensor (heating temperature > 300℃), and dynamically increases the flow rate of the aldehyde channel to 1.2L / min (+50%) according to formula (14), and simultaneously adjusts the power of the cylindrical heater to 400℃ (driven by the temperature gradient enhancement mechanism according to formula (9); c. Data results: The PID-VOC sensor detects that the aldehyde concentration increases from 35ppm to 182ppm (which meets the characteristics of a surge in aldehydes within 5 seconds in the initial stage of stir-frying), and the particle size distribution is: 0.5~2.0μm particles account for 23.1% (simulation target 23±3%). For example: (2) PAHs accumulation simulation in stewing scenario (environmental disturbance adaptive), simulates the continuous accumulation of PAHs in the later stage of stewing meat, and encounters the interference of cold air from the air conditioner. The specific implementation steps include: a. Steady state stage: PAHs channel (benzo[a]pyrene solution) flow rate 1.0L / min, temperature maintained at 200℃; b. Cold air interference (120th second): the infrared thermal imager detects a sudden drop in temperature of 10℃, the AI ​​decision engine module calculates the temperature gradient ∇T through formula (10), generates PID compensation instructions, the PLC dynamically increases the power of the cylindrical heater by 15%, and restores the target temperature within 2 seconds; c. Data results: temperature fluctuation < ±0.7℃, humidity fluctuation ±2.3%RH, PAHs concentration maintained at 205ppm. For example: multi-component synergy in frying scenario (AI dynamic ratio), VOCs and inorganic salt particles are generated synergistically when frying chicken legs. The specific implementation steps include: a. Initial parameters: VOCs channel (toluene solution): 0.5L / min, inorganic salt channel: 1.2L / min (ultrasonic oscillator to prevent sedimentation); b. Component abnormality (45th second): the laser scattering sensor detects that the proportion of 8~10μm particles is only 9% (< target 12±3%). The AI ​​decision engine module outputs Psmoke=0.92 through formula (13) (thickening is required). According to formula (14), the flow rate of inorganic salt channel is adjusted to 1.5L / min (+25%); c. Data results: the proportion of 8~10μm particles rebounds to 11.5% (close to the target value), and the particle aggregation degree is improved (RGB image shows improved particle size uniformity).

[0073] Among the optional implementations of this application, this application is significantly superior to traditional technologies in terms of dynamic component simulation, environmental disturbance suppression, and multimodal collaborative control, meeting the requirements for reproducing complex oil fume scenarios required for kitchen air conditioner testing, as detailed in Table 1 below.

[0074] Table 1 Performance comparison of the present invention with traditional cooking fume generators

[0075]

[0076] Example 2

[0077] This embodiment, based on Embodiment 1 above, provides a method for generating cooking fumes based on multimodal fusion. Please refer to [link to previous document]. Figure 4 The method, applied to a cooking fume generator based on multimodal fusion as described in Example 1, includes the following steps.

[0078] S1. The multi-component oil fume generation simulation module executes PID initial control commands and PID adjustment commands to mix different oil fume components in proportion, adjusts the atomization particle size through the atomization integration module, and dynamically adjusts the oil fume generation concentration according to test requirements.

[0079] S2. Collect multimodal real-time detection data during the test process through the multimodal data acquisition module and send it to the AI ​​decision engine module. Also receive PID initial control instructions and PID adjustment instructions, and control the multi-component oil fume generation simulation module to execute the PID initial control instructions and / or PID adjustment instructions.

[0080] S3. The AI ​​decision engine module generates an initial PID control command based on the initial oil fume and mist parameters input by the user, and sends the initial PID control command to the multimodal data acquisition module. The AI ​​model, constructed by an improved CNN convolutional neural network model, fuses the received multimodal real-time detection data, compares it with preset target features, outputs the oil fume state recognition result and PID adjustment parameter value, generates a PID adjustment command, and sends it to the multimodal data acquisition module.

[0081] Example 3

[0082] Based on Embodiment 2 described above, this embodiment also provides an electronic device, please refer to the appendix. Figure 5 , Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0083] like Figure 5 As shown, an electronic device may include a processing unit (such as a central processing unit, graphics processing unit, etc.) that can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) or a program loaded from a storage device into random access memory (RAM). The RAM also stores various programs and data required for the operation of the electronic device. The processing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0084] Typically, the following devices can be connected to an I / O interface: input devices such as touchscreens, touchpads, keyboards, mice, and cameras; output devices such as liquid crystal displays (LCDs) and speakers; storage devices such as magnetic tapes and hard drives; and communication devices. Communication devices allow electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0085] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0086] Example 4

[0087] Based on Embodiment 2 above, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above method.

[0088] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), or any suitable combination thereof.

[0089] In some alternative implementations of this application, the client and server may communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol), and may interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0090] The aforementioned computer-readable medium may be included in the aforementioned device or may exist independently without being assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: acquire training data and transform the training data to obtain initial data; determine an initial rule base based on the initial data and optimize the parameters of the initial rule base to obtain a target rule base; calculate activation weights for the rules in the target rule base according to a preset activation weight calculation formula; and determine abnormal information based on test data and the activation weights.

[0091] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0092] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0093] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including a data acquisition unit, a rule determination unit, a weight calculation unit, and an anomaly determination unit. The names of these units do not necessarily limit the specific unit; for example, a data acquisition unit may also be described as a "unit for acquiring training data."

[0094] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), etc.

[0095] Obviously, those skilled in the art will understand that the various steps of the present invention described above can be performed in a manner different from that described above, and the simulation methods and experimental equipment include, but are not limited to, the above description. The steps of the present invention described above can be performed in a different order in certain circumstances, and the steps shown or described above can be performed separately. Therefore, the present invention is not limited to any particular combination of hardware and software.

[0096] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered within the scope of protection of the present invention.

Claims

1. A cooking oil fume generating device based on multi-modal fusion, characterized in that, The method comprises the following steps: A multi-component oil smoke generation simulation module is used to execute PID initial control instructions and PID adjustment instructions, proportionally mix different oil smoke components, adjust the particle size of atomized particles through an atomization integrated module, and dynamically adjust the oil smoke generation concentration according to test requirements; The multi-component oil smoke generation simulation module comprises: A five-channel atomization module comprises an aldehyde channel, a polycyclic aromatic hydrocarbon channel, a vegetable oil channel, a volatile organic compound channel, and an inorganic salt suspension channel, and the aldehyde channel, the polycyclic aromatic hydrocarbon channel, the vegetable oil channel, the volatile organic compound channel, and the inorganic salt suspension channel are each provided with an independent liquid storage tank and a peristaltic pump for respectively quantitatively conveying oil smoke component precursor liquids of aldehydes, polycyclic aromatic hydrocarbons, vegetable oils, volatile organic compounds, and inorganic salt suspensions; A conical premixing chamber is used to receive and initially flow-mix the five kinds of oil smoke component precursor liquids conveyed by the aldehyde channel, the polycyclic aromatic hydrocarbon channel, the vegetable oil channel, the volatile organic compound channel, and the inorganic salt suspension channel, to obtain a premixing liquid; A static mixer is internally provided with a spiral blade structure, is connected with an output end of the conical premixing chamber, and is used to homogenize the premixing liquid to obtain a mixed liquid; An atomization integrated module is used to atomize the mixed liquid into oil smoke aerosol under the shearing action of high-temperature and high-pressure gas flow, and to complete pyrolysis and aerosol generation under gradient temperature control, so as to simulate the physical and chemical properties of real cooking oil smoke; A multi-modal data acquisition module is used to acquire multi-modal real-time detection data in the test process, and to send the multi-modal real-time detection data to an AI decision engine module, and to receive PID initial control instructions and PID adjustment instructions to control the multi-component oil smoke generation simulation module to execute the PID initial control instructions and / or the PID adjustment instructions; An AI decision engine module is used to generate PID initial control instructions according to initial oil smoke aerosol related parameters input by a user, to send the PID initial control instructions to the multi-modal data acquisition module, and to fuse the received multi-modal real-time detection data through an AI model constructed by an improved CNN convolutional neural network model, to compare preset target features, to output oil smoke state recognition results and PID adjustment parameter values, to generate PID adjustment instructions, and to send the PID adjustment instructions to the multi-modal data acquisition module.

2. The cooking oil fume generating device based on multi-modal fusion according to claim 1, characterized in that, The atomization integrated module comprises: A Venturi nozzle comprises a converging section, a throat section, and a diverging section; The converging section adopts a continuously smooth circular arc transition profile, and an inlet converging angle is 30°, so that the high-temperature and high-pressure gas flow is guided to stably accelerate from an inlet end into the throat section through continuously smooth curvature transition, and flow separation and vortex loss are reduced; The throat section is an equal-diameter cylindrical structure, which is used to ensure that the high-temperature and high-pressure gas flow reaches the maximum flow rate in the throat section and forms a negative pressure core area, and the throat section is provided with an introduction hole on a side surface, which is connected with an output end of the static mixer through a connecting pipe, and is used to drive the mixed liquid to be self-sucked in, so that the mixed liquid is broken to 0.1-10 μm under the shearing action of the high-temperature and high-pressure gas flow, and oil smoke droplets are formed; The diverging section is a gradually expanding diffusion structure, a diffusion angle of the diverging section is 20°, and the diverging section is internally provided with a cyclone, which is used to further break and homogenize the oil smoke droplets to form oil smoke aerosol. A high-pressure gas source is configured to deliver a high-pressure gas flow at normal temperature; A cylindrical heater is arranged between the high-pressure gas source and the Venturi nozzle, and is configured to heat the high-pressure gas flow at normal temperature to a preset temperature to form a high-temperature high-pressure gas flow, and inject the high-temperature high-pressure gas flow into the converging section of the Venturi nozzle.

3. The cooking oil fume generating device based on multi-modal fusion according to claim 2, characterized in that, The multi-modal data acquisition module comprises a temperature sensor, a dew point sensor, a pressure sensor, a flow rate sensor, a laser scattering sensor, a PID-VOC sensor, an infrared spectrometer, an RGB industrial camera, an infrared thermal imaging sensor, and a control module; The probe structures of the temperature sensor, the pressure sensor, and the flow rate sensor are embedded into the throat, and are configured to acquire real-time temperature data, real-time gas flow pressure data, and real-time flow rate data of the high-temperature high-pressure gas flow; The dew point sensor is arranged on the outer wall of the diffusion section, and is configured to acquire dew point temperature data of the gas mist at the wall surface of the diffusion section; The laser scattering sensor, the PID-VOC sensor, the infrared spectrometer, the RGB industrial camera, and the infrared thermal imaging sensor are integrated and arranged in a straight pipe section downstream of the outlet of the Venturi nozzle, the laser scattering sensor is configured to acquire particulate matter distribution data of the oil smoke mist, the PID-VOC sensor is configured to test aldehyde and PAHs components in the oil smoke mist, the infrared spectrometer is configured to identify characteristic functional groups of organic matter in the oil smoke mist, the RGB industrial camera is configured to acquire RGB images of the oil smoke mist, and the infrared thermal imaging sensor is configured to acquire infrared thermal imaging data of the oil smoke mist; The control module comprises a touch screen, a PLC controller, and an AD conversion module, the PLC controller is electrically connected to the five-channel atomization module, the static mixer, the high-pressure gas source, the cylindrical heater, the multi-modal data acquisition module, and the touch screen through the AD conversion module, and the PLC controller is connected to the AI decision engine module in a wired or wireless manner.

4. The cooking oil fume generating device based on multi-modal fusion according to claim 3, characterized in that, An annular Peltier condenser is further arranged on the outer wall of the diffusion section, a refrigeration end surface of the annular Peltier condenser is arranged on the outer wall of the diffusion section in alignment with a probe sensing end surface of the dew point sensor, and the annular Peltier condenser is configured to form a humidity closed-loop control circuit. 5.The cooking oil fume generating device based on multi-modal fusion of claim 2, wherein, The AI decision engine module comprises: An initial control instruction generation module configured to generate a PID initial control instruction according to initial oil smoke mist related parameters input by a user; An adjustment instruction generation module configured to fuse and process received multi-modal real-time detection data by an AI model constructed by an improved CNN convolutional neural network model, compare preset target features, output oil smoke state recognition results and PID adjustment parameter values, and generate a PID adjustment instruction; An instruction sending module configured to send the PID initial control instruction and / or the PID adjustment instruction to the PLC controller, and execute the PID initial control instruction and / or the PID adjustment instruction through the PLC controller.

6. The cooking oil fume generating device based on multi-modal fusion according to claim 5, characterized in that, The adjustment instruction generation module comprises: A data preprocessing module configured to preprocess RGB images of the oil smoke mist and infrared thermal imaging data of the oil smoke mist to obtain RGB image preprocessed data and infrared thermal imaging preprocessed data; The data enhancement module is configured to rotate the RGB image pre-processing data randomly, add brightness dithering to the rotated RGB image pre-processing data to obtain RGB image enhanced data, and inject Gaussian noise into the infrared thermal imaging pre-processing data to obtain infrared thermal imaging enhanced data. The feature extraction module is configured to extract spatial texture features of the RGB image enhanced data by a kth convolutional network to capture the distribution of oil smoke morphology, and extract temperature gradient features of the infrared thermal imaging enhanced data by an lth convolutional network to reflect the movement trend of hot air flow. The attention mechanism introduction module is configured to project the extracted temperature gradient features to a low-dimensional space, calculate the projection of a query vector and a key vector, construct a dynamic temperature similarity matrix of the infrared temperature of the oil smoke, and inject the temperature gradient features into the standard point product attention to strengthen the response of the oil smoke flow area. The cross-modal attention fusion module is configured to enhance the model motion mechanism by normalizing the attention weight, generate a spatial attention map to identify a key high-temperature area, calculate a value vector weighted output, fuse features according to the attention weight, enhance a dynamic oil smoke area, and complete multi-modal feature fusion. The state recognition and decision output module is configured to input the fused multi-modal features into a decision output layer, output an oil smoke state recognition result, determine whether the oil smoke state recognition result meets a set requirement, output a PID adjustment parameter if the oil smoke state recognition result does not meet the set requirement, and generate a PID adjustment instruction.

7. A cooking oil fume generation method based on multi-modal fusion, characterized in that, The method comprises the following steps: The multi-component oil smoke generation simulation module executes the PID initial control instruction and the PID adjustment instruction, proportionally mixes different oil smoke components, adjusts the particle size of the atomized particles through the atomization integrated module, and dynamically adjusts the oil smoke generation concentration according to the test requirements. The multi-component oil smoke generation simulation module comprises: The five-channel atomization module comprises an aldehyde channel, a polycyclic aromatic hydrocarbon channel, a vegetable oil channel, a volatile organic compound channel, and an inorganic salt suspension channel, and each of the aldehyde channel, the polycyclic aromatic hydrocarbon channel, the vegetable oil channel, the volatile organic compound channel, and the inorganic salt suspension channel is provided with an independent liquid storage tank and a peristaltic pump for quantitatively conveying oil smoke component precursor liquid of aldehydes, polycyclic aromatic hydrocarbons, vegetable oil, volatile organic compounds, and inorganic salt suspensions, respectively. The conical premixing chamber is used to receive and initially flow-mix the five oil smoke component precursor liquids conveyed by the aldehyde channel, the polycyclic aromatic hydrocarbon channel, the vegetable oil channel, the volatile organic compound channel, and the inorganic salt suspension channel to obtain a premixing liquid. The static mixer is internally provided with a spiral blade structure and is connected with the output end of the conical premixing chamber, and is used to homogenize the premixing liquid to obtain a mixed liquid. The atomization integrated module is used to atomize the mixed liquid into oil smoke mist under the shearing action of high-temperature and high-pressure gas flow, and complete pyrolysis and aerosol generation under gradient temperature control to simulate the physical and chemical properties of real cooking oil smoke. The multi-modal real-time detection data in the test process is collected by the multi-modal data acquisition module, and is sent to the AI decision engine module, and the PID initial control instruction and the PID adjustment instruction are received, and the multi-component oil fume generation simulation module is controlled to execute the PID initial control instruction and / or the PID adjustment instruction; The AI decision engine module generates the PID initial control instruction according to the initial oil fume gas related parameters input by the user, and sends the PID initial control instruction to the multi-modal data acquisition module, and the AI model constructed by the improved CNN convolutional neural network model is used to fuse the received multi-modal real-time detection data, compare the preset target characteristics, output the oil fume state recognition result and the PID adjustment parameter value, generate the PID adjustment instruction and send it to the multi-modal data acquisition module.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the method as claimed in claim 7.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to realize the steps of the method as claimed in claim 7.

Images