Self-adaptive adjusting system and method for air circulation function of clean workshop in electronic industry

Abstract

The invention discloses an electronic industry clean workshop air circulation function self-adaptive adjusting system and method. The adjusting system comprises a three-dimensional grid monitoring module, a data processing module, a visualization module, a computer vision module, an intelligent decision module and an air circulation control module. The adjusting method comprises the steps of three-dimensional grid sensing arrangement and monitoring, data processing, data visualization, image feature extraction, abnormal area early warning, air circulation parameter decision making, self-adaptive adjusting execution and closed-loop feedback optimization. A continuous space distribution model is constructed through real-time monitoring data of a three-dimensional grid sensor and is converted into a dynamic visual three-dimensional heat map, correlation between environment characteristics and air circulation control quantity is further established, and air flow control parameters are output based on local pollution abnormal conditions. The adjusting accuracy of the system on complex working conditions can be remarkably improved, and self-adaptive adjustment and sustainable evolution of the air circulation function are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of control systems, and in particular to a self-adaptive regulating system and method for air circulation functions in clean workshops in the electronics industry. Background Art

[0002] Electronics cleanrooms are specially designed and constructed to meet the cleanliness requirements of modern high-precision electronic products. Their core goal is to control airborne particulate matter below a specific concentration to prevent particles from landing on products like integrated circuits, leading to circuit shorts, performance degradation, and even product failure. Therefore, the cleanliness of electronics cleanrooms directly impacts the performance and yield of chips, integrated circuits, and other products. Electronics cleanrooms must maintain ISO Class 5 or higher cleanliness standards, and achieving this through traditional air circulation systems often relies on high airflow rates and extremely high fan energy consumption.

[0003] Air circulation systems in traditional electronics industry cleanrooms are often set up according to fixed operating conditions, or rely on simple sensors for monitoring and manual control of the air circulation system. This approach has significant drawbacks: First, single-point monitoring cannot accurately reflect the distribution and dynamic changes of particle concentration in three-dimensional space, making it difficult to detect pollution anomalies in local areas. Second, air circulation systems with operating parameters set according to fixed operating conditions cannot promptly and accurately adjust to changes in particle concentration caused by factors such as equipment load changes, process flow differences, and the number of different personnel activities, often resulting in wasted fan energy. Third, air circulation systems with operating parameters set according to fixed operating conditions cannot quickly eliminate interference caused by sudden increases in particle concentration and fluctuations in the production environment, which may affect the production quality of high-precision electronic products and even contaminate production lines. Fourth, traditional air circulation systems usually use the entire factory as a control unit, making it difficult to accurately match the cleanliness requirements of local areas. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention applies for an adaptive adjustment system and method for the air circulation function of clean workshops in the electronics industry. Through three-dimensional grid monitoring, three-dimensional visualization analysis, computer vision feature extraction and artificial intelligence decision-making, it realizes precise and adaptive adjustment of the air circulation in clean workshops in the electronics industry, and solves the problems of insufficient monitoring accuracy, lack of adaptive ability and delayed emergency response in traditional air circulation systems.

[0005] In a first aspect, the present application discloses an adaptive regulation system for air circulation function in a clean workshop of the electronics industry, comprising a three-dimensional grid monitoring module, a data processing module, a visualization module, a computer vision module, an intelligent decision-making module, and an air circulation control module;

[0006] The three-dimensional grid monitoring module is used to obtain real-time environmental data in the electronic industry clean room by arranging sensors in the form of a three-dimensional grid in the electronic industry clean room, including data on particle concentration, wind speed and air pressure changes over space and time;

[0007] The data processing module converts the environmental data acquired by the three-dimensional grid monitoring module into a continuous spatial distribution model according to an interpolation method, including a particle concentration field model, a wind speed field model, and an air pressure field model, and obtains a particle concentration change rate field model by calculating the partial derivative of the particle concentration field model with respect to time t;

[0008] The visualization module generates a three-dimensional visual heat map based on the continuous spatial distribution model and displays it to the monitoring personnel through a display device;

[0009] The computer vision module establishes a correlation between the visualized three-dimensional heat map and environmental features by using convolutional neural network model technology; the environmental features include high particle concentration areas, particle generation sources, air pressure zone boundaries, airflow dead spots and eddy current areas;

[0010] The intelligent decision-making module uses deep neural network model technology to establish a correlation between the environmental characteristics and the air circulation control variables based on the cleanliness control requirements and the principle of reducing energy consumption. The air circulation control variables include the fan supply air volume and direction, the fan exhaust air volume and the air valve opening;

[0011] The air circulation control module adjusts parameters of the air circulation equipment in the clean workshop of the electronics industry according to the air circulation control amount.

[0012] Preferably, the sensors of the three-dimensional grid monitoring module include a particle concentration sensor, a wind speed sensor and an air pressure sensor, which are respectively used to obtain data on the changes of particle concentration, wind speed and air pressure in space and time; the three-dimensional grid divides the electronic industry clean workshop into multiple rectangular or polygonal areas in the horizontal direction, and is layered according to height in the vertical direction. The sensors in each layer are distributed in an array, and the spacing between sensors is reduced for areas sensitive to the production environment; the production environment sensitive areas include areas where high-precision equipment operates, key process flow areas, areas with intensive personnel activities, air pressure zone boundaries, areas with frequent historical pollution or areas prone to exceeding concentration standards, and local microenvironments with the highest cleanliness level.

[0013] Preferably, the visualization module adopts a differentiated visualization coding method based on the visualized three-dimensional heat map, and extracts the environmental features from the visualized three-dimensional heat map through the computer vision module, including: adopting a color gradient visualization coding method for the particulate matter concentration field model, assigning differentiated color marks according to the size of the particulate matter concentration value, and highlighting the high-concentration particulate matter area exceeding the threshold; adopting a vector arrow visualization coding method for the wind speed field model, assigning the length and direction of the vector arrow according to the size and direction of the wind speed, and then identifying the airflow dead corner and eddy current area through the computer vision module and highlighting them; adopting a contour line visualization coding method for the air pressure field model, drawing contour lines according to the spatial distribution of air pressure, and then identifying the air pressure partition boundary through the computer vision module and highlighting it; adopting a color gradient visualization coding method for the particulate matter concentration change rate field model, assigning differentiated color marks according to the size of the particulate matter concentration change rate value, and then identifying the particulate matter generation source through the computer vision module and highlighting it.

[0014] A second aspect of the present application discloses a method for adaptively adjusting the air circulation function of a clean workshop in the electronics industry, comprising the following steps:

[0015] S1. 3D Grid Sensor Arrangement and Monitoring: Sensors are deployed in a horizontally partitioned and vertically layered manner within the electronics industry cleanroom to construct a 3D grid monitoring network. The sensors are used to obtain real-time environmental data within the electronics industry cleanroom, including data on particle concentration variations over space and time (c(x, y, z, t), wind speed variations over space and time (w(x, y, z, t), and air pressure variations over space and time (p(x, y, z, t)). For production environment-sensitive areas, the spacing between sensors is reduced. The sensors transmit the collected environmental data to the data processing module via wireless communication.

[0016] S2. Data processing: Based on the environmental data obtained by the sensor and the interpolation method, the particle concentration field model C(x, y, z, t), wind speed field model W(x, y, z, t) and pressure field model P(x, y, z, t) are constructed; by calculating the partial derivative of the particle concentration field model C(x, y, z, t) with respect to time t, the particle concentration change rate field model V is constructed. c (x,y,z,t);

[0017] S3. Data visualization: The particle concentration field model C(x, y, z, t) and the particle concentration change rate field model V c(x, y, z, t), wind speed field model W(x, y, z, t) and pressure field model P(x, y, z, t) are converted into a three-dimensional visualization heat map, and the three-dimensional visualization heat map is mapped as a texture onto the three-dimensional model of the electronic industry clean room;

[0018] S4. Image feature extraction: Analyze the three-dimensional visualized heat map using computer vision technology to extract environmental features, including areas with high particle concentrations, particle generation sources, air pressure zone boundaries, airflow blind spots, and eddy current areas;

[0019] S5. Abnormal area warning: For the particle concentration field model C(x,y,z,t) and the particle concentration change rate field model V c Areas where the (x, y, z, t) values ​​exceed the preset threshold are highlighted and an audible and visual alarm is triggered;

[0020] S6. Decision on air circulation parameters: Input the environmental characteristics into the intelligent decision module, and based on the cleanliness control requirements and the principle of reducing energy consumption, output air circulation control parameters, including fan supply air volume and direction, fan exhaust air volume, and air valve opening;

[0021] S7, adaptive adjustment execution: adjusting parameters of the air circulation equipment of the electronics industry clean workshop according to the air circulation control amount;

[0022] S8. Closed-loop feedback optimization: When the duration of the high particulate matter concentration area exceeds the preset time length T, the closed-loop feedback optimization is triggered, specifically: repeat steps S1 to S4, and input the adjusted environmental characteristics into the intelligent decision-making module. The intelligent decision-making module is based on a deep neural network model and uses an incremental learning algorithm to iteratively optimize the association model between the environmental characteristics and the air circulation control quantity. The incremental learning algorithm updates the hidden layer weights of the neural network through real-time collected environmental data, and the update frequency is automatically triggered after each adjustment cycle ends.

[0023] Compared with the prior art, the beneficial effects of the present invention are: in response to the problems of insufficient monitoring accuracy, lack of adaptive ability and delayed emergency response in the existing air circulation system, an adaptive adjustment system and method for the air circulation function of the clean workshop of the electronics industry is proposed, the adjustment system includes a three-dimensional grid monitoring module, a data processing module, a visualization module, a computer vision module, an intelligent decision-making module and an air circulation control module; the adjustment method includes three-dimensional grid sensor layout and monitoring, data processing, data visualization, image feature extraction, abnormal area warning, air circulation parameter decision, adaptive adjustment execution and closed-loop feedback optimization; a continuous spatial distribution model is constructed through the real-time monitoring data of the three-dimensional grid sensor to realize the distribution of particulate matter in the workshop, High-precision prediction of airflow status effectively reduces quality risks caused by pollution blind spots such as pollution sources, airflow dead corners and eddy current areas; the visualization module converts the particulate matter concentration field, wind speed field, etc. into a dynamic visual three-dimensional heat map, and combines computer vision technology to extract environmental characteristics, so that technicians can intuitively locate pollution sources and airflow anomalies, improving operation and maintenance efficiency; through artificial intelligence technology, the relationship between environmental characteristics and air circulation control quantities is established, and airflow control parameters are output based on local pollution anomalies. Compared with traditional artificial intelligence technology that outputs airflow control parameters based on single-point monitoring data, it can significantly improve the system's adjustment accuracy for complex working conditions, and achieve continuous evolution of adaptability through a closed-loop feedback optimization mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a module diagram of an adaptive regulation system for air circulation in a clean workshop in the electronics industry according to an embodiment of the present invention;

[0025] Figure 2 This is a flow chart of a method for adaptively adjusting the air circulation function of a clean workshop in the electronics industry according to an embodiment of the present invention;

[0026] Figure numerals: 1-three-dimensional grid monitoring module, 11-particle concentration sensor, 12-wind speed sensor, 13-air pressure sensor, 2-data processing module, 3-visualization module, 4-computer vision module, 5-intelligent decision-making module, 6-air circulation control module. DETAILED DESCRIPTION

[0027] The following is a more detailed description of the embodiments of the present invention with reference to the accompanying drawings and reference numerals, so that those skilled in the art can implement the invention after studying this specification. It should be understood that the specific embodiments described herein are only used to illustrate the invention and are not intended to limit the invention.

[0028] The first aspect of this application discloses Figure 1The self-adaptive regulation system for air circulation function of clean workshops in the electronics industry shown includes a three-dimensional grid monitoring module 1, a data processing module 2, a visualization module 3, a computer vision module 4, an intelligent decision-making module 5 and an air circulation control module 6.

[0029] The three-dimensional grid monitoring module 1 utilizes sensors arranged in a three-dimensional grid within the electronics cleanroom to acquire real-time environmental data within the cleanroom, including data on particle concentration, wind speed, and air pressure as they vary over space and time. Specifically, the sensors in the three-dimensional grid monitoring module 1 include a particle concentration sensor 11, a wind speed sensor 12, and an air pressure sensor 13, respectively used to acquire data on particle concentration, wind speed, and air pressure as they vary over space and time. The three-dimensional grid divides the cleanroom into multiple rectangular or polygonal areas horizontally and layers them vertically by height, with sensors arranged in an array within each layer. For conventional production areas, horizontal spacing ranges from 5m to 10m, and vertical spacing ranges from 3m to 5m. For sensitive production environment areas, sensor spacing is reduced based on sensitivity, with horizontal spacing ranging from 1m to 5m and vertical spacing from 1m to 3m. These sensitive production environment areas include areas where high-precision equipment operates, critical process flow areas, areas with high human activity, air pressure zone boundaries, areas with historically high pollution levels or prone to exceeding pollution standards, and local microenvironments with the highest cleanliness levels. The initial sensor layout plan refers to the simulation results of the computational fluid dynamics model of the electronic industry clean workshop; the sensor layout plan during the continuous operation period is adjusted according to the environmental characteristics obtained by the computer vision module 4.

[0030] The data processing module 2 converts the environmental data obtained by the three-dimensional grid monitoring module 1 into a continuous spatial distribution model according to the interpolation method, including a particle concentration field model, a wind speed field model and an air pressure field model, and obtains a particle concentration change rate field model by calculating the partial derivative of the particle concentration field model with respect to time t.

[0031] The visualization module 3 generates a visualized three-dimensional heat map based on the continuous spatial distribution model and displays it to monitoring personnel via a display device. The computer vision module 4 uses convolutional neural network model technology to establish a correlation between the visualized three-dimensional heat map and environmental characteristics; the environmental characteristics include areas with high particle concentrations, particle generation sources, air pressure zone boundaries, airflow blind spots, and vortex areas. In a specific implementation, the visualization module 3 adopts a differentiated visualization coding method based on the visualized three-dimensional heat map, and extracts the environmental features from the visualized three-dimensional heat map through the computer vision module 4, including: adopting a color gradient visualization coding method for the particle concentration field model, assigning differentiated color marks according to the particle concentration value, and highlighting the high particle concentration area exceeding the threshold; adopting a vector arrow visualization coding method for the wind speed field model, assigning the vector arrow length and direction according to the magnitude and direction of the wind speed, and then identifying the airflow dead corner and eddy current area through the computer vision module 4 and highlighting them; adopting a contour line visualization coding method for the pressure field model, drawing contour lines according to the spatial distribution of the air pressure, and then identifying the pressure partition boundary through the computer vision module 4 and highlighting it; adopting a color gradient visualization coding method for the particle concentration change rate field model, assigning differentiated color marks according to the particle concentration change rate value, and then identifying the particle generation source through the computer vision module 4 and highlighting it.

[0032] The intelligent decision module 5 establishes the association between the environmental characteristics and the air circulation control quantity based on the cleanliness control requirements and the principle of reducing energy consumption by adopting deep neural network model technology. The air circulation control quantity includes the fan supply volume and direction, the fan exhaust volume and the air valve opening. In the specific implementation, the input layer of the deep neural network model is a 20-dimensional environmental feature vector, which specifically includes: the three-dimensional coordinates of the high-concentration particulate matter area, the particulate matter concentration value and the area type mark, a total of 5 dimensions; the three-dimensional coordinates of the particulate matter generation source, the particulate matter concentration change rate and the source type label, a total of 5 dimensions; the coordinates of the pressure partition boundary, the pressure gradient value and the partition number, a total of 5 dimensions; the center coordinates of the airflow dead corner or vortex area, the wind speed vector and the vortex intensity index, a total of 5 dimensions. The output layer of the deep neural network model is a (3k+m+n)-dimensional air circulation control quantity matrix, which specifically includes: the three-dimensional vector of the supply volume of k fans, the exhaust volume of m fans, and the opening of n air valves. Based on the cleanliness control requirements and the principle of reducing energy consumption, the intelligent decision module adopts the loss function as shown below:

[0033] L=w1·L c +w2·L e (1)

[0034] Among them, L cis the cleanliness loss, L e is the energy loss; w1 and w2 are both weight coefficients. When the duration of the high concentration area of ​​particulate matter exceeds the preset time length T, w1 = 1 and w2 = 0, otherwise it automatically switches to the mode of w1 = 0.8 and w2 = 0.2.

[0035] The air circulation control module 6 adjusts the parameters of the air circulation equipment of the electronic industry clean workshop according to the air circulation control amount. The air circulation equipment includes a fan and a damper.

[0036] The second aspect of this application discloses Figure 2 The self-adaptive adjustment method for the air circulation function of the clean workshop of the electronics industry shown includes the following steps:

[0037] S1. Three-dimensional grid sensor layout and monitoring: Sensors are arranged in the electronic industry clean room in a horizontal partition and vertical layered manner to construct a three-dimensional grid monitoring network; the sensors include a particle concentration sensor 11, a wind speed sensor 12, and an air pressure sensor 13. In the specific implementation, the sensors are integrated into non-critical parts of the production equipment as much as possible, while ensuring that the sensor measurement port is flush with or exposed to the surface of the equipment to avoid data deviation due to obstruction of the equipment structure. The environmental data in the electronic industry clean room is obtained in real time through sensors, including data c(x, y, z, t) on particle concentration changes with space and time, data w(x, y, z, t) on wind speed changes with space and time, and data p(x, y, z, t) on air pressure changes with space and time; for areas sensitive to the production environment, the spacing between sensors is reduced; the sensors transmit the collected environmental data to the data processing module through wireless communication.

[0038] S2. Data processing: Based on the environmental data obtained by the sensors and the interpolation method, the particle concentration field model C(x,y,z,t), wind speed field model W(x,y,z,t) and pressure field model P(x,y,z,t) are constructed. By calculating the partial derivative of the particle concentration field model C(x,y,z,t) with respect to time t, the particle concentration change rate field model V is constructed. c In specific implementation, the Kriging interpolation method is used for particulate matter concentration data to increase sensitivity to areas with high particulate matter concentrations or sources of particulate matter; the spline interpolation method is used for wind speed and air pressure data to ensure smooth changes in the wind speed and air pressure fields.

[0039] S3. Data visualization: The particle concentration field model C(x, y, z, t) and the particle concentration change rate field model V c(x, y, z, t), wind speed field model W(x, y, z, t) and air pressure field model P(x, y, z, t) are converted into a three-dimensional visualization heat map, and the three-dimensional visualization heat map is mapped as a texture onto the three-dimensional model of the electronics industry clean room.

[0040] S4. Image feature extraction: Use computer vision technology to analyze the three-dimensional visualized heat map and extract environmental features, including areas with high particle concentrations, particle generation sources, air pressure zone boundaries, airflow blind spots, and eddy current areas.

[0041] S5. Abnormal area warning: For the particle concentration field model C(x,y,z,t) and the particle concentration change rate field model V c Areas where the (x, y, z, t) values ​​exceed the preset thresholds are highlighted and trigger audible and visual alarms.

[0042] S6. Decision on air circulation parameters: The environmental characteristics are input into the intelligent decision module 5, and based on the cleanliness control requirements and the principle of reducing energy consumption, the air circulation control parameters are output, including the fan supply volume and direction, the fan exhaust volume and the air valve opening.

[0043] S7. Adaptive adjustment execution: Parameter adjustment is performed on the air circulation equipment of the electronic industry clean workshop according to the air circulation control amount.

[0044] S8, closed-loop feedback optimization: When the duration of the high-concentration area of ​​particulate matter exceeds the preset duration T, the closed-loop feedback optimization is triggered, specifically: repeat steps S1 to S4, and input the adjusted environmental characteristics into the intelligent decision-making module. The intelligent decision-making module is based on a deep neural network model and uses an incremental learning algorithm to iteratively optimize the correlation model between the environmental characteristics and the air circulation control quantity. The incremental learning algorithm updates the hidden layer weights of the neural network through real-time collected environmental data, and the update frequency is automatically triggered after each adjustment cycle. In a specific implementation, the preset duration T is 10 minutes, and the goal of the iterative optimization is to reduce the fan energy consumption by no less than 15% while meeting the cleanliness level requirements. The learning rate of the incremental learning algorithm is 0.01.

[0045] It can be seen that by constructing a continuous spatial distribution model through the real-time monitoring data of the three-dimensional grid sensor, high-precision prediction of the particle distribution and airflow status in the factory can be achieved, effectively reducing the quality risks caused by pollution blind spots such as pollution sources, airflow dead corners and vortex areas; the particle concentration field, wind speed field, etc. are converted into dynamic visual three-dimensional heat maps through the visualization module, and combined with the extraction of environmental characteristics using computer vision technology, technical personnel can intuitively locate pollution sources and airflow anomalies, thereby improving operation and maintenance efficiency; through artificial intelligence technology, the relationship between environmental characteristics and air circulation control quantities is established, and airflow control parameters are output based on local pollution anomalies. Compared with traditional artificial intelligence technology that outputs airflow control parameters based on single-point monitoring data, it can significantly improve the system's adjustment accuracy for complex working conditions, and achieve continuous evolution of adaptability through a closed-loop feedback optimization mechanism.

[0046] The above is a description of one or more embodiments of the present invention, and while the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present invention, and these modifications and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. The air circulation function adaptive adjustment system of the electronic industry clean room is characterized by: The system comprises a three-dimensional grid monitoring module, a data processing module, a visualization module, a computer vision module, an intelligent decision-making module and an air circulation control module; the three-dimensional grid monitoring module arranges sensors in the form of a three-dimensional grid in the electronic industry clean room to obtain environmental data in real time in the electronic industry clean room, including data on particle concentration, wind speed and air pressure changing with space and time; the data processing module converts the environmental data obtained by the three-dimensional grid monitoring module into a continuous spatial distribution model according to an interpolation method, including a particle concentration field model, a wind speed field model and an air pressure field model, and obtains a particle concentration change rate field model by calculating the partial derivative of the particle concentration field model with respect to time t; the visualization module calculates the particle concentration change rate field model according to the continuous spatial distribution The model generates a visual three-dimensional heat map and displays it to monitoring personnel through a display device; the computer vision module establishes a correlation between the visual three-dimensional heat map and environmental characteristics by adopting convolutional neural network model technology; the environmental characteristics include areas with high particle concentration, sources of particle generation, air pressure zone boundaries, airflow dead corners and vortex areas; the intelligent decision-making module establishes a correlation between the environmental characteristics and air circulation control quantities by adopting deep neural network model technology based on cleanliness control requirements and the principle of reducing energy consumption, and the air circulation control quantities include fan air supply volume and direction, fan exhaust volume and air valve opening; the air circulation control module adjusts the parameters of the air circulation equipment in the clean workshop of the electronics industry according to the air circulation control quantities.

2. The electronic industry clean room air circulation function adaptive adjustment system according to claim 1 is characterized in that: The sensors of the three-dimensional grid monitoring module include a particle concentration sensor, a wind speed sensor and an air pressure sensor, which are respectively used to obtain data on the changes of particle concentration, wind speed and air pressure in space and time; the three-dimensional grid divides the electronic industry clean workshop into multiple rectangular or polygonal areas in the horizontal direction, and is layered by height in the vertical direction. The sensors in each layer are distributed in an array, and the layout spacing of sensors is reduced for areas sensitive to the production environment; the production environment sensitive areas include areas where high-precision equipment operates, key process flow areas, areas with intensive personnel activities, air pressure zone boundaries, areas with frequent historical pollution or areas prone to exceeding concentration standards, and local microenvironments with the highest cleanliness level.

3. The self-adaptive air circulation control system for clean workshops in the electronics industry according to claim 1 is characterized in that: The visualization module adopts a differentiated visualization coding method based on the visualized three-dimensional heat map, and extracts the environmental features from the visualized three-dimensional heat map through the computer vision module, including: adopting a color gradient visualization coding method for the particle concentration field model, assigning differentiated color marks according to the size of the particle concentration value, and highlighting the high particle concentration area exceeding the threshold; adopting a vector arrow visualization coding method for the wind speed field model, assigning the length and direction of the vector arrow according to the size and direction of the wind speed, and then identifying the airflow dead corners and eddy current areas through the computer vision module and highlighting them; adopting a contour line visualization coding method for the air pressure field model, drawing contour lines according to the spatial distribution of air pressure, and then identifying the air pressure partition boundaries through the computer vision module and highlighting them; adopting a color gradient visualization coding method for the particle concentration change rate field model, assigning differentiated color marks according to the size of the particle concentration change rate value, and then identifying the particle generation source through the computer vision module and highlighting it.

4. The self-adaptive air circulation control system for clean workshops in the electronics industry according to claim 1 is characterized in that: The input layer of the deep neural network model is a 20-dimensional environmental feature vector, specifically including: the three-dimensional coordinates of the particulate matter high concentration area, the particulate matter concentration value and the area type label, a total of 5 dimensions; the three-dimensional coordinates of the particulate matter generation source, the particulate matter concentration change rate and the source type label, a total of 5 dimensions; the coordinates of the pressure partition boundary, the pressure gradient value and the partition number, a total of 5 dimensions; the center coordinates of the airflow dead corner or eddy area, the wind speed vector and the eddy intensity index, a total of 5 dimensions; the output layer of the deep neural network model is a (3k+m+n)-dimensional air circulation control quantity matrix, specifically including: the three-dimensional vector of the air supply volume of k fans, the exhaust volume of m fans, and the opening of n air valves; based on the cleanliness control requirements and the principle of reducing energy consumption, the deep neural network model adopts the loss function as shown below: L=w1·L c +w2·L e Among them, L c is the cleanliness loss, L e is the energy loss; w1 and w2 are both weight coefficients. When the duration of the high concentration area of ​​particulate matter exceeds the preset time T, w1 = 1 and w2 = 0, otherwise it automatically switches to the normal mode of w1 = 0.8 and w2 = 0.

2.

5. The self-adaptive adjustment method of air circulation function in clean workshops of electronic industry is characterized by: The system is applied to the air circulation function adaptive adjustment system of the electronic industry clean workshop according to any one of claims 1 to 4, and comprises the following steps: S1. 3D Grid Sensor Arrangement and Monitoring: Sensors are deployed in a horizontally partitioned and vertically layered manner within the electronics industry cleanroom to construct a 3D grid monitoring network. The sensors are used to obtain real-time environmental data within the electronics industry cleanroom, including data on particle concentration variations over space and time (c(x, y, z, t), wind speed variations over space and time (w(x, y, z, t), and air pressure variations over space and time (p(x, y, z, t)). For production environment-sensitive areas, the spacing between sensors is reduced. The sensors transmit the collected environmental data to the data processing module via wireless communication. S2. Data processing: Based on the environmental data obtained by the sensor and the interpolation method, the particle concentration field model C(x, y, z, t), wind speed field model W(x, y, z, t) and pressure field model P(x, y, z, t) are constructed; by calculating the partial derivative of the particle concentration field model C(x, y, z, t) with respect to time t, the particle concentration change rate field model V is constructed. c (x,y,z,t); S3. Data visualization: The particle concentration field model C(x, y, z, t) and the particle concentration change rate field model V c (x, y, z, t), wind speed field model W(x, y, z, t) and pressure field model P(x, y, z, t) are converted into a three-dimensional visualization heat map, and the three-dimensional visualization heat map is mapped as a texture onto the three-dimensional model of the electronic industry clean room; S4. Image feature extraction: Analyze the three-dimensional visualized heat map using computer vision technology to extract environmental features, including areas with high particle concentrations, particle generation sources, air pressure zone boundaries, airflow blind spots, and eddy current areas; S5. Abnormal area warning: For the particle concentration field model C(x,y,z,t) and the particle concentration change rate field model V c Areas where the (x, y, z, t) values ​​exceed the preset threshold are highlighted and an audible and visual alarm is triggered; S6. Decision on air circulation parameters: Input the environmental characteristics into the intelligent decision module, and based on the cleanliness control requirements and the principle of reducing energy consumption, output air circulation control parameters, including fan supply air volume and direction, fan exhaust air volume, and air valve opening; S7, adaptive adjustment execution: adjusting parameters of the air circulation equipment of the electronics industry clean workshop according to the air circulation control amount; S8. Closed-loop feedback optimization: When the duration of the high particulate matter concentration area exceeds the preset time length T, the closed-loop feedback optimization is triggered, specifically: repeat steps S1 to S4, and input the adjusted environmental characteristics into the intelligent decision-making module. The intelligent decision-making module is based on a deep neural network model and uses an incremental learning algorithm to iteratively optimize the association model between the environmental characteristics and the air circulation control quantity. The incremental learning algorithm updates the hidden layer weights of the neural network through real-time collected environmental data, and the update frequency is automatically triggered after each adjustment cycle ends.

Images