IoT-based intelligent airflow control method for cleanrooms
By acquiring the real-time operating status of production equipment through IoT technology and combining it with the partial differential equation of turbulent diffusion in cleanroom airflow organization, the airflow in the cleanroom can be predicted and adjusted in advance, solving the problem of particulate matter diffusion lag caused by sudden actions of production equipment in the cleanroom and achieving stable control of airflow in the cleanroom.
Patent Information
- Application Number
- CN202611316057.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-08-28
AI Technical Summary
Existing cleanroom airflow control methods suffer from delayed adjustment response due to gas transmission and sensor sampling delays when production equipment suddenly operates, which fails to suppress particulate matter diffusion in a timely manner and disrupts the stability of the unidirectional flow pattern in the cleanroom.
By acquiring the real-time operating cycle state sequence of production equipment through the Internet of Things, mapping it to the spatiotemporal function of particulate matter release source intensity, and coupling it with the pre-built cleanroom airflow organization turbulent diffusion partial differential equation to solve the spatial evolution distribution field of particulate matter concentration in future periods, generating pre-adjustment control vectors in advance, and changing the local unidirectional flow direction and wind speed.
It achieves proactive feedforward prediction based on equipment cycle events, eliminates gas transmission and sensor sampling delays, maintains the stability of unidirectional airflow patterns in key areas of the cleanroom, and improves the predictive accuracy and physical feasibility of control execution.
Smart Images

Figure CN122834975B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of control and regulation technology, specifically to an intelligent airflow control method for cleanrooms based on the Internet of Things. Background Technology
[0002] Cleanrooms have stringent requirements for micro-contamination control, and existing airflow control systems mostly employ closed-loop control architectures based on environmental parameter feedback. Particulate matter concentration sensors and micro-differential pressure sensors are deployed at the top and return air side of key areas in the cleanroom to continuously collect environmental data and transmit it to the controller. The controller has internally set safety thresholds for concentration and differential pressure. It compares the real-time collected values with these safety thresholds. When the measured concentration exceeds the safety threshold, the controller calculates the required airflow compensation value based on a proportional-integral-derivative control law, outputting an analog electrical signal to the variable air volume (VAV) terminal unit and the return air valve actuator. This changes the supply air velocity and airflow direction, attempting to dilute and remove particulate matter by increasing local exhaust airflow or increasing supply air velocity, trying to restore a local unidirectional flow pattern. This feedback mechanism relies on the sensors' physical perception of existing contamination; particulate matter must move with the airflow to the sensor sampling port and be detected and analyzed before subsequent adjustment actions can be triggered.
[0003] In semiconductor manufacturing processes, production equipment undergoes periodic or random start-ups, shutdowns, and operational switching. Friction in moving parts or the instantaneous opening and closing of process chambers can suddenly release large amounts of particulate matter. When this sudden equipment action causes particulate release, the airflow carries the particles and diffuses them within the cleanroom space. Under existing feedback control architectures, the particles need to undergo physical transport from the release point to the sensor location via airflow, and the sensor itself also requires processing time to complete particle sampling and data conversion. During this transport and sampling delay, the controller maintains the original airflow parameters, allowing the particles to continue diffusing within the space. This makes them highly susceptible to breaching the boundaries of local unidirectional flow and intruding into downstream critical process areas. By the time the sensor detects excessive concentration and triggers fan adjustment, the contamination has already spread, meaning the control action is essentially delayed beyond the time of the contamination event.
[0004] Existing cleanroom airflow control methods based on sensor feedback suffer from a core technical problem when dealing with particulate matter releases caused by sudden equipment operation. This problem stems from a lag in the adjustment response due to delays in gas transmission and sensor sampling. This lag prevents the control system from intervening in the initial stages of particulate matter release, forcing it to passively wait for contamination to spread before taking remedial measures. Consequently, temporary contamination inevitably occurs in localized areas at the moment of equipment operation, disrupting the stability of the unidirectional flow pattern in the cleanroom. Simply optimizing the parameters of the feedback control law cannot fundamentally eliminate this lag defect caused by physical delay. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent airflow control method for cleanrooms based on the Internet of Things, which can solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The method for intelligent airflow control in cleanrooms based on the Internet of Things includes: acquiring the operating cycle state sequence of production equipment in the cleanroom in real time through the Internet of Things, and mapping the operating cycle state sequence to a spatiotemporal function of particulate matter release source intensity; The spatiotemporal function of the particulate matter emission source intensity is coupled with the pre-established partial differential equation of cleanroom airflow organization turbulent diffusion to calculate the spatial evolution distribution field of particulate matter concentration in future time periods. Based on the location and intensity of the predicted concentration extreme points in the spatial evolution distribution field, pre-adjustment control vectors are generated for each air supply terminal and return air inlet. Before the particulate matter generated by the operation of the production equipment diffuses to the critical area, the local unidirectional flow direction and wind speed are changed in advance.
[0007] Preferably, the step of mapping the operating cycle state sequence to a spatiotemporal function of particulate matter emission source intensity includes: extracting the device start-stop timing and action type identifier from the operating cycle state sequence; Based on the action type identifier, a preset release source feature map is retrieved to obtain the baseline release intensity and spatial distribution profile corresponding to the action type identifier; The reference release intensity and the spatial distribution profile are aligned on the time axis and mapped to spatial coordinates according to the device start-up and shutdown sequence to generate a spatiotemporal function of the particulate matter release source intensity that varies with time and spatial location.
[0008] Preferably, the step of coupling the spatiotemporal function of the particulate matter emission source intensity with the pre-established partial differential equation of cleanroom airflow organization turbulence diffusion includes: obtaining the current real-time pressure difference distribution and supply air velocity of the cleanroom as initial boundary conditions; Substitute the spatiotemporal function of the particulate matter emission source intensity as a source term into the partial differential equation of turbulent diffusion for cleanroom airflow organization. The finite volume method is used to perform spatiotemporal discretization and iterative calculation on the partial differential equation of turbulent diffusion in the cleanroom airflow organization with source terms until the residual converges to a preset threshold, and the spatial evolution distribution field of particulate matter concentration in the future time period is output.
[0009] Preferably, the step of generating pre-adjustment control vectors for each air supply terminal and return air inlet based on the location and intensity of the predicted concentration extreme points in the spatial evolution distribution field includes: extracting the spatial coordinates and extreme values of the predicted concentration extreme points that exceed the cleanliness threshold in the spatial evolution distribution field; Calculate the weighting coefficients of the spatial coordinates within the flow field influence domain of each air supply terminal and return air inlet; Based on the extreme values and the weighting coefficients, an air volume adjustment objective function is constructed. The objective function is constrained and solved to obtain the target air volume values of each air supply terminal and return air inlet. These values are then combined to form the pre-adjustment control vector.
[0010] Preferably, after generating the spatiotemporal function of the particulate matter release source intensity, the method further includes: identifying the time interval between adjacent action events in the running beat state sequence; If the time interval is less than a preset time window threshold, the spatiotemporal functions of particulate matter release source intensity corresponding to adjacent action events are superimposed and merged, and the time step of the cleanroom airflow organization turbulent diffusion partial differential equation is dynamically adjusted during coupled solution, wherein the time step is negatively correlated with the rate of change of the release source intensity after superposition and merging.
[0011] Preferably, before the step of retrieving a preset release source feature map based on the action type identifier, the method further includes: collecting historical particulate matter concentration time-series data of the cleanroom production equipment when it historically performed each action type; Cluster analysis was performed on the historical particulate matter concentration time series data according to action type, and the concentration transient response curve and spatial diffusion attenuation characteristics corresponding to each action type were extracted; The peak value of the concentration transient response curve is fitted with the envelope of the spatial diffusion attenuation feature to form a new benchmark release intensity and spatial distribution profile, and the release source feature map is replaced and updated.
[0012] Preferably, the step of obtaining the current real-time differential pressure distribution and air supply velocity of the cleanroom as initial boundary conditions includes: obtaining the original differential pressure value and original air velocity value collected by the sensor array distributed in the cleanroom; Kalman filtering is used to fuse and smooth the original pressure difference value and the original wind speed value to obtain the filtered current state parameter. The current state parameters are weighted and fused with the predicted state parameters from the previous moment, and the weighted fusion result is used as the initial boundary condition and substituted into the cleanroom airflow organization turbulent diffusion partial differential equation.
[0013] Preferably, in the step of constraining and solving the objective function of air volume adjustment, the constraint conditions of the objective function of air volume adjustment include: the adjusted air volume value of each air supply terminal and return air inlet does not exceed the maximum rated air volume of the corresponding air inlet; The differential pressure gradient between adjacent zones remains within the tolerance range of the design differential pressure threshold after the pre-adjustment control vector is executed; The sum of the absolute values of the changes in each air volume value in the pre-adjustment control vector is minimized, and the concentration value of the predicted concentration extreme point in the spatial evolution distribution field is lower than the cleanliness threshold after adjustment.
[0014] Preferably, after changing the local unidirectional flow direction and wind speed in advance, the method further includes: collecting the actual particulate matter concentration value of the key area in real time, and calculating the residual between the actual particulate matter concentration value and the predicted concentration value at the corresponding location in the spatial evolution distribution field; Determine whether the residual exceeds the preset allowable deviation range. If it does, calculate the compensation air volume correction value based on the residual, superimpose the compensation air volume correction value onto the current pre-adjustment control vector, correct the air volume of the air supply terminal and return air inlet based on the superimposed control vector, and use the actual particulate matter concentration value to correct the diffusion coefficient in the turbulent diffusion partial differential equation of the cleanroom airflow organization.
[0015] Preferably, when multiple production devices in the cleanroom simultaneously generate operating cycle state sequences, an independent initial pre-adjustment control vector is generated for each production device. The system identifies whether the airflow adjustment directions of the initial pre-adjustment control vectors for the same air supply terminal or return air inlet are opposite. If they are opposite, the extreme values of the predicted concentration extreme points corresponding to each production equipment are extracted. For each air supply terminal or return air inlet with opposite airflow adjustment directions, only the adjustment instructions in the initial pre-adjustment control vector corresponding to the production equipment with the largest predicted concentration extreme value are retained. The adjustment instructions of the other production equipment for the air supply terminal or return air inlet with opposite airflow adjustment directions are discarded. The final global pre-adjustment control vector is generated based on the adjustment instructions.
[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By acquiring the operating cycle state sequence of production equipment through the Internet of Things (IoT), this sequence is mapped to a spatiotemporal function of particulate matter release source intensity. This function is then coupled with a pre-built partial differential equation for turbulent diffusion in cleanroom airflow organization to calculate the spatial evolution distribution field of particulate matter concentration in future time periods. Based on the predicted location and intensity of concentration extreme points, a pre-adjustment control vector is generated. This vector alters the local unidirectional flow direction and velocity before particulate matter generated by equipment actions diffuses into critical areas. This processing logic shifts the control mode from passive feedback after an event occurs to active feedforward prediction driven by equipment cycle events. This eliminates the inherent physical delays caused by gas transmission and sensor sampling, constructs an airflow barrier before actual particulate matter diffusion, suppresses local particulate matter concentration peaks caused by sudden equipment actions, and maintains the stability of the unidirectional airflow pattern in critical areas. By extracting action type identifiers and retrieving the feature spectrum of release sources to obtain the baseline release intensity, and combining real-time pressure difference and supply air velocity as boundary conditions for finite volume method iterative calculation, the accuracy of the evolution distribution field prediction is improved, ensuring that the generation of the pre-adjustment control vector is based on a rigorous physical model.
[0017] 2. During operation, when the time interval between adjacent action events is less than the time window threshold, the spatiotemporal functions of the corresponding particulate matter release source intensity are superimposed and merged, and the time step of the partial differential equation is dynamically adjusted to avoid repeated calculations and actuator oscillations under high-frequency actions. By collecting historical particulate matter concentration time-series data, the feature map of the release source is clustered and updated to ensure that the setting of the release source intensity is consistent with the aging and wear state of the equipment. Kalman filtering is used to fuse the original pressure difference and wind speed values as initial boundary conditions, reducing the interference of sensor noise on the solution of the partial differential equation. When generating the pre-adjustment control vector, the maximum rated air volume, pressure difference gradient tolerance, and minimization of air volume change are introduced as constraints to ensure the physical feasibility of the actuator action. The compensation air volume correction value is calculated based on the residual between the actual concentration and the predicted concentration, and the diffusion coefficient of the partial differential equation is corrected to eliminate the accumulation of model prediction errors. When multiple devices generate operating cycles simultaneously, conflict identification is performed on the commands with opposite air volume adjustment directions at the same air outlet, and the adjustment command of the device with the maximum extreme value is retained to avoid control confrontation under multi-source disturbances and ensure the consistency of global flow field adjustment. Attached Figure Description
[0018] Figure 1 A flowchart illustrating the overall implementation of IoT-based intelligent airflow control in cleanrooms; Figure 2 Flowchart for converting equipment operation cycle sequences into a spatiotemporal function of particulate matter emission source intensity; Figure 3 Flowchart for solving the spatial evolution distribution field of particulate matter concentration using the source-term coupled turbulent diffusion equation; Figure 4Flowchart for generating pre-adjustment control vectors for supply and return air terminals based on concentration extreme value information; Figure 5 Flowchart for verifying the residual of measured concentration after airflow pre-regulation and correcting closed-loop model parameters; Figure 6 A flowchart for handling multiple control vector conflicts and generating global control vectors under synchronous operation of multiple devices. Detailed Implementation
[0019] This specific embodiment provides an intelligent airflow control method for cleanrooms based on the Internet of Things (IoT). This method uses IoT technology to achieve real-time perception of the operating status of production equipment, transforms equipment action events into quantifiable mathematical models of particulate matter release sources, and combines computational fluid dynamics methods to predict the diffusion process of particulate matter in the cleanroom in advance. Based on the prediction results, feedforward control commands are generated to pre-adjust the airflow organization before the particulate matter actually diffuses to the critical area.
[0020] Appendix Figure 1 In one embodiment, the cleanroom airflow intelligent control system includes an IoT sensing layer, a data processing layer, a model calculation layer, and an execution control layer. The IoT sensing layer consists of status acquisition terminals deployed on various production equipment, a sensor array distributed throughout the cleanroom space, and an edge computing gateway. The status acquisition terminals communicate with the programmable logic controllers of the production equipment via an industrial bus to acquire real-time operating parameters and action commands. The sensor array includes a micro-differential pressure sensor, a wind speed sensor, and an optical particulate counter, used to collect differential pressure, wind speed, and particulate matter concentration data at different locations within the cleanroom, respectively. The edge computing gateway is responsible for preprocessing and format conversion of the raw data collected by the sensing layer, and transmitting the processed data to the data processing layer via Ethernet. The data processing layer receives the data stream from the edge computing gateway, performs timestamp alignment, outlier removal, and time sequence reassembly on the data, generating a standardized operating cycle state sequence. The model computation layer is equipped with a pre-trained database of emission source feature maps and a pre-built solver for the partial differential equations of cleanroom airflow organization turbulence diffusion. It receives the operating cycle state sequence output from the data processing layer and completes the construction of the spatiotemporal function of particulate matter emission source intensity, the solution of the spatial evolution distribution field of particulate matter concentration, and the generation of pre-adjustment control vectors. The execution control layer consists of variable air volume terminal controllers, return air valve actuators, and a central controller. It receives the pre-adjustment control vectors output from the model computation layer, converts them into analog electrical signals or digital control commands, and drives the actuators of each air supply terminal and return air inlet to complete the airflow adjustment action.
[0021] In one embodiment, the operational cycle state sequence of production equipment within a cleanroom is acquired in real time via the Internet of Things (IoT). The operational cycle state sequence consists of a series of discrete action events, each containing a timestamp, a unique device identifier, an action type identifier, and an action duration. A status acquisition terminal reads the register values inside the programmable logic controller (PLC) of the production equipment at a fixed sampling frequency. When a register value changes, the status acquisition terminal generates a new action event record. The status acquisition terminal adds a local timestamp to the generated action event record and sends it to an edge computing gateway via the MQTT protocol. The edge computing gateway performs time synchronization processing on the received action event records from multiple production devices, using a network time protocol to align the local clocks of each status acquisition terminal with the system's global clock, eliminating clock discrepancies between different devices. The edge computing gateway sorts the time-synchronized action event records in chronological order, generating a global operational cycle state sequence containing action information for all production equipment.
[0022] The operating cycle state sequence is mapped to a spatiotemporal function of particulate matter release source intensity. This function describes the particulate matter release rate at any spatial location within the cleanroom at any given time, with spatial coordinates and time as independent variables and the number of particles released per unit volume per unit time as the dependent variable. Each action event in the operating cycle state sequence corresponds to an independent particulate matter release source; different types of equipment actions have different release intensities and spatial distribution characteristics. By extracting the action type identifier from the action event and retrieving a preset release source feature map, the baseline release intensity time curve and spatial distribution profile function corresponding to that action type are obtained. The baseline release intensity time curve is translated and stretched along the time axis according to the occurrence time and duration of the action event, and the spatial distribution profile function is transformed according to the equipment's installation location within the cleanroom, yielding the local release source intensity spatiotemporal function corresponding to that action event. The local release source intensity spatiotemporal functions corresponding to all action events are superimposed to obtain the global particulate matter release source intensity spatiotemporal function.
[0023] The spatiotemporal function of particulate matter emission source intensity is coupled with a pre-established partial differential equation for cleanroom airflow organization and turbulent diffusion. This equation, based on the Reynolds-averaged Navier-Stokes equations and component transport equations, describes the physical laws governing airflow motion and particulate matter diffusion within the cleanroom. The pre-established equations include continuity, momentum, energy, and particulate matter transport equations, with coefficients predetermined based on the cleanroom's geometry, building material properties, and air physics parameters. During the coupled solution process, the current real-time pressure differential distribution and supply air velocity are first obtained as initial boundary conditions. The spatiotemporal function of particulate matter emission source intensity is substituted as a source term into the particulate matter transport equations. The finite volume method is used to spatially discretize the entire computational domain, dividing the continuous domain into several non-overlapping control volumes. For each control volume, integral form of the control equations is applied, transforming the partial differential equations into a system of algebraic equations. The SIMPLE algorithm is used to iteratively solve the algebraic equations until the residuals of each variable converge to a preset threshold. The spatial distribution data of particulate matter concentration at different times in the future period are output, forming a spatial evolution distribution field of particulate matter concentration.
[0024] Based on the location and intensity of the predicted concentration extreme points in the spatial evolution distribution field, pre-regulation control vectors are generated for each air supply terminal and return air vent. The pre-regulation control vector is a one-dimensional array, with the number of elements equal to the total number of air supply terminals and return air vents in the cleanroom. Each element corresponds to a target airflow regulation value for one vent. First, all grid nodes in the spatial evolution distribution field are traversed, and grid nodes with concentration values exceeding the cleanliness threshold are extracted. The point with the largest concentration value among these nodes is identified as the predicted concentration extreme point. The spatial coordinates of the predicted concentration extreme point and the spatial distance between each air supply terminal and return air vent are calculated. Combined with the flow field influence range of each vent, the influence weight coefficient of each vent on the extreme point is determined. An airflow regulation objective function is constructed with the goal of minimizing the concentration at the predicted concentration extreme point, using the maximum rated airflow of each vent, the pressure gradient limit between adjacent zones, and the airflow change limit as constraints. A sequential quadratic programming algorithm is used to solve the constrained objective function, obtaining the target airflow values for each air supply terminal and return air vent. These target airflow values are then combined to form the pre-regulation control vector.
[0025] The equipment start-up and shutdown sequence and action type listed in the column are used to preemptively change the local unidirectional airflow direction and velocity before particulate matter generated by the production equipment diffuses to the critical area. The critical area refers to the process operation area within the cleanroom where particulate matter concentration requirements are most stringent; its spatial scope is predefined in the system. The shortest time required for particulate matter to diffuse from the release source to the boundary of the critical area is calculated. This time is obtained by calculating the particulate matter diffusion velocity and the shortest distance from the release source to the critical area. The execution time of the pre-adjustment control vector is set to this shortest time minus the response time of the actuator, ensuring that the airflow organization adjustment is completed before the particulate matter reaches the critical area. The central controller receives the pre-adjustment control vector and decomposes it into independent control commands for each variable air volume (VAV) terminal and return air valve. The VAV terminal controller adjusts the opening of the air valve according to the control commands, changing the airflow volume and angle at the supply air terminal. The return air valve actuator adjusts the opening of the return air valve according to the control commands, changing the return air volume at the return air inlet. By adjusting the airflow ratio at different locations of the air supply terminals and return air inlets, a local airflow barrier is formed around the predicted concentration extreme point, altering the diffusion path of particulate matter and preventing it from spreading to critical areas.
[0026] Appendix Figure 2 In one embodiment, the operating cycle state sequence is extracted. The operating cycle state sequence is stored in time series format, with each data point containing a timestamp, device ID, action code, and state value. Device start-up and stop timing is extracted by analyzing changes in state values; a change from 0 to 1 indicates device startup, and a change from 1 to 0 indicates device shutdown. Action type identifiers are determined by action codes; different action codes correspond to different device action types, such as robotic arm movement, wafer transfer, cavity opening and closing, and valve switching. The system pre-establishes a mapping table between action codes and action type identifiers, converting action codes into standardized action type identifiers through table lookup operations.
[0027] Based on the action type identifier, a preset release source feature map is retrieved to obtain the baseline release intensity and spatial distribution profile corresponding to the action type identifier. The release source feature map is a database storing particulate matter release characteristic data corresponding to different equipment types and action types. Each entry includes an action type identifier, equipment model, baseline release intensity time curve, spatial distribution profile function, and applicable conditions. The baseline release intensity time curve is a discrete time-intensity data pair sequence, describing the change in the number of particulate matter released per unit time under standard operating conditions for that action type. The spatial distribution profile function is a three-dimensional function describing the spatial distribution characteristics of particulate matter release intensity in a local coordinate system with the equipment action point as the origin. During the retrieval, the equipment model is first determined based on the equipment ID, and then a matching entry is searched in the release source feature map based on the action type identifier. If a completely matching entry exists, the baseline release intensity time curve and spatial distribution profile function of that entry are directly extracted; if no completely matching entry exists, entries with the same equipment model and similar action types are searched, and an approximate baseline release intensity time curve and spatial distribution profile function are generated using an interpolation method.
[0028] The baseline release intensity and spatial distribution profile are aligned on the time axis and mapped to spatial coordinates according to the equipment start-up and shutdown sequence, generating a spatiotemporal function of particulate matter release source intensity that varies with time and spatial location. The time axis alignment operation aligns the origin of the baseline release intensity time curve with the occurrence time of the equipment action event, and translates the time axis of the baseline release intensity time curve so that the start point of the curve corresponds to the start time of the action event, and the end point of the curve corresponds to the end time of the action event. If the duration of the action event differs from the duration of the baseline release intensity time curve, the time axis of the baseline release intensity time curve is stretched or compressed to match the duration of the action event. The spatial coordinate mapping operation transforms the local coordinate system with the equipment action point as the origin into the global coordinate system of the cleanroom. First, the installation position coordinates and attitude angles of the equipment in the cleanroom's global coordinate system are obtained, and a transformation matrix from the local coordinate system to the global coordinate system is established. Each local coordinate point in the spatial distribution profile function is multiplied by the transformation matrix to obtain the corresponding global coordinate point. Multiplying the time-axis-aligned baseline release intensity with the spatial distribution contour function mapped from spatial coordinates yields the spatiotemporal function of the local release source intensity corresponding to the action event, whose expression is:
[0029] in, Indicates the first An action event in global coordinates place, time The intensity of particulate matter emission sources, This represents the baseline release strength-time curve after time axis alignment. The spatial distribution contour function represents the spatial coordinate mapping. Indicates the first The coordinates of the device action point corresponding to each action event in the global coordinate system.
[0030] By superimposing the spatiotemporal functions of the local emission source intensity corresponding to all action events, the global particulate matter emission source intensity spatiotemporal function is obtained, and its expression is:
[0031] in, Represents the spatiotemporal function of global particulate matter emission source intensity. This represents the total number of action events in the running beat state sequence.
[0032] In one embodiment, after generating the spatiotemporal function of particulate matter release source intensity, the time interval between adjacent action events in the operating cycle state sequence is identified. Adjacent action events refer to two consecutive action events occurring on the same production equipment. The time interval between adjacent action events is obtained by calculating the difference between the start time of the latter action event and the end time of the former action event. The calculated time interval is compared with a preset time window threshold, which is pre-set based on the airflow characteristics and particulate matter diffusion velocity of the cleanroom.
[0033] If the time interval is less than a preset time window threshold, the spatiotemporal functions of particulate matter emission source intensities corresponding to adjacent action events are superimposed and merged. The superposition and merging operation involves splicing the spatiotemporal functions of local emission source intensities corresponding to two adjacent action events on the time axis, and adding the intensity values of the two functions within the overlapping time period. The time range of the merged emission source intensity spatiotemporal function is from the start time of the first action event to the end time of the second action event. During coupled solution, the time step of the cleanroom airflow organization turbulent diffusion partial differential equation is dynamically adjusted. The time step is negatively correlated with the rate of change of the superimposed and merged emission source intensity. The rate of change of emission source intensity is obtained by calculating the first derivative of the merged emission source intensity spatiotemporal function with respect to time. The formula for adjusting the time step is:
[0034] in, Indicates the adjusted time step. Indicates the initial time step. This represents the proportionality coefficient. This represents the absolute value of the maximum rate of change of the spatiotemporal function of the merged release source intensity within the current calculation period. (Scale factor) The value ranges from 0.001 to 0.01, and is adjusted according to the size of the cleanroom and the required calculation accuracy. When the intensity of the release source changes rapidly, the time step automatically decreases to improve calculation accuracy; when the intensity of the release source changes slowly, the time step automatically increases to improve calculation efficiency.
[0035] In one embodiment, before retrieving a preset release source feature map based on the action type identifier, historical particulate matter concentration time-series data is collected from the production equipment in the cleanroom during the execution of various action types. Multiple optical particulate matter counters are deployed at different locations within the cleanroom, with a sampling frequency set to 1Hz. When the production equipment performs a certain action type, all optical particulate matter counters are simultaneously triggered to begin data collection, which lasts for 30 minutes after the action ends. The collected particulate matter concentration data is associated and stored with the corresponding action type identifier, equipment ID, collection time, and collection location to form a historical particulate matter concentration time-series database.
[0036] Historical particulate matter concentration time-series data were clustered based on action type. The K-means clustering algorithm was used, with action type identifiers as cluster labels, grouping all historical particulate matter concentration time-series data corresponding to the same action type into one class. For each class, characteristic parameters of the concentration time-series curves were extracted, including peak concentration, peak occurrence time, rise time, fall time, and half-width at half-maximum (WHM). The mean and standard deviation of the characteristic parameters for all concentration time-series curves within the same class were calculated, and outliers exceeding the range of the mean ± 3 standard deviations were removed.
[0037] The concentration transient response curve and spatial diffusion decay characteristics corresponding to each action type were extracted. The concentration transient response curve was obtained by arithmetically averaging all historical particulate matter concentration time-series data corresponding to the same action type. The spatial diffusion decay characteristics were obtained by analyzing the variation of particulate matter concentration with distance at different locations. For each sampling location, the ratio of the peak concentration at that location to the peak concentration at the location closest to the release source was calculated to obtain the concentration decay coefficient at that location. An exponential function was used to fit the concentration decay coefficients at different distances to obtain the spatial diffusion decay function.
[0038] The peak value of the concentration transient response curve is fitted with the envelope of the spatial diffusion attenuation characteristics to form a new baseline release intensity and spatial distribution profile, thus updating the release source feature map. The peak value of the concentration transient response curve is used as the peak value of the new baseline release intensity, and the shape of the concentration transient response curve is used as the shape of the new baseline release intensity time curve. The spatial diffusion attenuation function is used as the radial component of the new spatial distribution profile function, and a three-dimensional spatial distribution profile function is constructed by combining the directional characteristics of the equipment action. The newly generated baseline release intensity time curve and spatial distribution profile function replace the old data corresponding to the action type identifier in the release source feature map, completing the update of the release source feature map. The release source feature map is updated monthly, or immediately after equipment maintenance. The baseline release intensity parameters for different action types are shown in Table 1 below.
[0039] Table 1 Examples of baseline release strength parameters for different motion types Action type identifier Equipment Model Peak release intensity (cells / s) Rise time (s) Descent time (s) Spatial attenuation coefficient (m⁻¹) 001 A-100 5.2e+05 0.3 2.1 1.8 002 A-100 3.7e+05 0.5 1.8 1.5 003 B-200 8.9e+05 0.2 3.5 2.2 004 B-200 2.1e+05 0.8 1.2 1.3 005 C-300 1.5e+06 0.1 4.7 2.7 Table 1 shows examples of baseline release intensity parameters for different action types. Action type identifiers include 001 for rapid robotic arm movement, 002 for wafer transfer, 003 for process chamber opening and closing, 004 for valve switching, and 005 for material handling. Peak release intensity represents the maximum number of particles released per unit time for this action type. Rise time represents the time required from the start of the action to reach the peak release intensity. Fall time represents the time required for the release intensity to decrease from the peak to 10% of the peak. Spatial decay coefficient represents the rate of decay of particle concentration with increasing distance.
[0040] Appendix Figure 3 In one embodiment, the current real-time differential pressure distribution and supply air velocity of the cleanroom are obtained as initial boundary conditions. The differential pressure distribution of the cleanroom is measured using micro differential pressure sensors distributed between areas of different cleanliness levels, with at least one micro differential pressure sensor deployed between each adjacent area. The supply air velocity is measured using hot-wire anemometers installed at the outlets of each air supply terminal, with one hot-wire anemometer deployed at each air supply terminal. The sampling frequency of the sensors is set to 10Hz, and the collected data is transmitted to the data acquisition module via a 4-20mA analog signal.
[0041] Kalman filtering is used to fuse and smooth the original pressure difference and wind speed values to obtain the filtered current state parameters. The state vector of the Kalman filter includes the pressure difference and wind speed values at each measurement point, while the observation vector consists of the original pressure difference and wind speed values collected by the sensors. The prediction step of the Kalman filter predicts the prior state estimate and error covariance matrix at the current time based on the state estimate from the previous time step and the system state transition matrix. The update step of the Kalman filter calculates the Kalman gain based on the current observation values and the observation matrix, and then corrects the prior state estimate to obtain the posterior state estimate at the current time step. The system state transition matrix uses an identity matrix, assuming that the system state remains unchanged within one sampling period. The observation matrix also uses an identity matrix, assuming a linear relationship between the sensor observations and the system state. The process noise covariance matrix and the observation noise covariance matrix are predetermined based on the sensor's technical parameters and the statistical characteristics of historical data.
[0042] The current state parameters are weighted and fused with the predicted state parameters from the previous time step. The weighted fusion result is then used as the initial boundary condition and substituted into the partial differential equation for turbulent diffusion in the cleanroom airflow organization. The predicted state parameters from the previous time step refer to the predicted state values for the current time step obtained when solving the partial differential equation for turbulent diffusion in the cleanroom airflow organization in the previous step. The weighting coefficients for the weighted fusion are determined based on the posterior error covariance matrix of the Kalman filter and the model prediction error covariance matrix. The weighted fusion formula is as follows:
[0043] in, This represents the initial boundary conditions after weighted fusion. This represents the current state parameter after Kalman filtering. This represents the current state parameter predicted from the previous time step. This represents the weighting coefficient. Weighting coefficient The value ranges from 0.6 to 0.9. When the sensor has high measurement accuracy, Take the larger value; when the model's prediction accuracy is high. Take the smaller value.
[0044] In one embodiment, the spatiotemporal function of particulate matter release source intensity is substituted as a source term into the partial differential equation for turbulent diffusion in cleanroom airflow organization. The expression for the particulate matter transport equation in the partial differential equation for turbulent diffusion in cleanroom airflow organization is as follows:
[0045] in, This indicates particulate matter concentration, expressed in particles / m³. Indicates time, in seconds; , , These represent the airflow at... , , The velocity component in the direction, in m / s; This represents the effective diffusion coefficient, with units of m² / s; The spatiotemporal function representing the intensity of particulate matter emission sources, expressed in particles / (m³•s). Effective diffusion coefficient. It consists of the molecular diffusion coefficient and the turbulent diffusion coefficient, which is calculated based on the turbulence model.
[0046] The finite volume method was used to perform spatiotemporal discretization and iterative calculation of the partial differential equations for turbulent diffusion in cleanroom airflow organization with source terms. Spatial discretization employed a structured mesh, dividing the computational domain of the cleanroom into a uniform hexahedral mesh with mesh sizes ranging from 0.1m to 0.5m based on the required computational accuracy. Integral form of the governing equations was applied to each control volume, transforming the partial differential equations into a system of algebraic equations concerning the mesh node variables. Temporal discretization used a first-order implicit scheme to ensure computational stability. The SIMPLE algorithm was employed to couple the velocity and pressure fields, and a successive over-relaxation iterative method was used to solve the system of algebraic equations. During the iteration process, the residual changes of each variable were monitored, and the computation was considered convergent when the residuals of all variables were less than a preset threshold. The preset threshold was set to... .
[0047] Output the spatial evolution distribution field of particulate matter concentration for future time periods. The length of the future time period is determined based on the operating cycle of the production equipment and the particulate matter diffusion rate, typically set to 10 to 60 seconds. The spatial evolution distribution field is stored as a three-dimensional array, where the array dimensions correspond to spatial coordinates and time, and the values of the array elements are the particulate matter concentrations at the corresponding locations and times. The temporal resolution of the spatial evolution distribution field is consistent with the computation time step, and the spatial resolution is consistent with the grid size.
[0048] Appendix Figure 4 In one embodiment, the spatial coordinates and extreme values of predicted concentration extreme points exceeding a cleanliness threshold are extracted from the spatial evolution distribution field. The cleanliness threshold is preset according to the cleanroom grade standard; for example, the cleanliness threshold for an ISO 5 cleanroom is 3520 particles / m³ (particle size ≥ 0.5 μm). All time steps and all grid nodes in the spatial evolution distribution field are traversed, and grid nodes with concentration values exceeding the cleanliness threshold are marked as exceeding nodes. For each time step, the node with the largest concentration value among all exceeding nodes is found, and this node is determined as the predicted concentration extreme point for that time step. The spatial coordinates of each predicted concentration extreme point are recorded. Extreme values and corresponding time .
[0049] Calculate the weighting coefficients for spatial coordinates within the flow field influence domains of each air supply terminal and return air inlet. Each air supply terminal and return air inlet has a corresponding flow field influence domain, which refers to the spatial area where changes in airflow at that inlet can significantly affect the internal flow field. The extent of the flow field influence domain is determined through prior numerical simulation or experimental measurements, typically a spherical region with a radius of 3m to 5m centered on the inlet. For each predicted concentration extreme point, calculate its spatial distance from each air supply terminal and return air inlet. If the distance is less than the radius of the flow field influence domain of that inlet, then the extreme point is within the flow field influence domain of that inlet. The weighting coefficient is inversely proportional to the square of the distance, and its calculation formula is:
[0050] in, Indicates the first The first opportunity for the first The weighting coefficients for each predicted concentration extreme point Indicates the first The first trend and the first Spatial distance between predicted concentration extreme points This represents the total number of air vents whose flow field influence domains include the extreme point. If the extreme point is not within the flow field influence domain of a certain air vent, then the weight coefficient of that air vent for that extreme point is 0.
[0051] An objective function for airflow regulation is constructed based on extreme values and weighting coefficients. The goal of this objective function is to ensure that the adjusted concentration values at all predicted concentration extreme points are below the cleanliness threshold, while minimizing the sum of airflow changes at each air outlet. The expression for the objective function is:
[0052] in, Represents the objective function value. This represents the total number of predicted concentration extreme points. Indicates the first Changes in air volume at each vent Indicates the first The rated air volume of each air outlet This represents the regularization coefficient. The value ranges from 0.01 to 0.1, and is used to balance the concentration control target and the target of minimizing air volume change.
[0053] The objective function for airflow regulation is constrained and solved to obtain the target airflow values for each supply air terminal and return air inlet, which are then combined to form a pre-regulation control vector. The constraints include: the regulated airflow values for each supply air terminal and return air inlet do not exceed the maximum rated airflow of the corresponding inlet; the differential pressure gradient between adjacent zones after executing the pre-regulation control vector remains within the tolerance range of the design differential pressure threshold; the sum of the absolute values of the changes in airflow values in the pre-regulation control vector is minimized; and the predicted concentration extreme points in the spatial evolution distribution field have regulated concentration values lower than the cleanliness threshold. A sequential quadratic programming algorithm is used to transform the constrained nonlinear optimization problem into a series of quadratic programming subproblems for solution. In each iteration, a quadratic approximation of the objective function and a linear approximation of the constraints are constructed, the quadratic programming subproblems are solved to obtain the search direction, and then a line search is performed to determine the step size and update the design variables. The iteration terminates when the change in the objective function value is less than the preset precision, yielding the optimal airflow change. The current airflow value of each inlet is added to the optimal airflow change to obtain the target airflow value for each inlet. The target airflow values for all air outlets are arranged in a preset order to form a pre-adjustment control vector. The pre-adjustment control vector is shown in Table 2 below.
[0054] Table 2 Examples of Pre-adjustment Control Vectors Wind vent number Type of air vent Current air volume (m³ / h) Target air volume (m³ / h) Air volume change (m³ / h) 001 air supply terminal 1200 1500 +300 002 air supply terminal 1200 1450 +250 003 air supply terminal 1200 1100 -100 004 return air vent 800 1000 +200 005 return air vent 800 950 +150 006 return air vent 800 700 -100 Table 2 shows an example of a pre-adjustment control vector. This example includes three supply air terminals and three return air vents, all with the current airflow at the rated volume. Based on the location and intensity of the predicted concentration extreme points, the system generates target airflow values for each vent. Specifically, the airflow at supply air terminals 001 and 002 is increased to create a stronger unidirectional flow near the extreme points; the airflow at supply air terminal 003 is decreased to adjust the airflow direction; the airflow at return air vents 004 and 005 is increased to accelerate particulate matter discharge; and the airflow at return air vent 006 is decreased to maintain pressure differential balance between zones.
[0055] Appendix Figure 5 In one embodiment, after pre-changing the local unidirectional flow direction and wind speed, the actual particulate matter concentration value of the key area is collected in real time. Multiple optical particulate matter counters are uniformly deployed within the key area, with a sampling frequency set to 1Hz. The actual particulate matter concentration data collected by the optical particulate matter counters is transmitted to the data processing layer via Ethernet. The data processing layer performs timestamp alignment and outlier removal on the data to obtain standardized time-series data of the actual particulate matter concentration.
[0056] Calculate the residual between the actual particulate matter concentration and the predicted concentration at the corresponding location in the spatial evolution distribution field. For each sampling time and each sampling location, extract the predicted concentration value for that time and location from the spatial evolution distribution field. Subtract the predicted concentration value from the actual particulate matter concentration value to obtain the residual value. The formula for calculating the residual value is:
[0057] in, Indicates time ,Location The residual value at that point, Indicates time ,Location The actual particulate matter concentration at that location, Represents the time in the spatial evolution distribution field ,Location The predicted concentration value at that location.
[0058] Determine if the residuals exceed the preset allowable deviation range. The preset allowable deviation range is set to ±20% of the predicted concentration value. If the residual values at all sampling locations are within the preset allowable deviation range, the model prediction is considered accurate and no correction is needed. If the residual value at any sampling location exceeds the preset allowable deviation range, the model prediction is considered to have a significant error and correction is required.
[0059] If the residual exceeds the preset allowable deviation range, a compensation airflow correction value is calculated based on the residual. The calculation of the compensation airflow correction value adopts a proportional control law, and the magnitude of the correction value is proportional to the residual value. For each supply air terminal and return air vent, the corresponding compensation airflow correction value is calculated based on its influence weight coefficient on the sampling location where the residual exceeds the limit. The compensation airflow correction value is superimposed on the current pre-adjustment control vector to obtain the corrected control vector. Based on the superimposed control vector, the airflow of the supply air terminal and return air vent is corrected, and the central controller sends the corrected control commands to each actuator to adjust the airflow of each vent.
[0060] The diffusion coefficient in the turbulent diffusion partial differential equation for cleanroom airflow organization is corrected using actual particulate matter concentration values. The correction employs the gradient descent method, with the objective function being minimizing the sum of squared residuals. At each correction step, the partial derivative of the sum of squared residuals with respect to the diffusion coefficient is calculated, and the value of the diffusion coefficient is adjusted along the negative gradient direction. The adjustment step size is adaptively determined based on the magnitude of the residuals; a larger step size is used when the residuals are large, and a smaller step size is used when the residuals are small. The corrected diffusion coefficient is used for the next solution of the spatial evolution distribution field of particulate matter concentration.
[0061] Appendix Figure 6In one embodiment, when multiple production devices in a cleanroom simultaneously generate operating cycle state sequences, an independent initial pre-adjustment control vector is generated for each device. For each device, its corresponding operating cycle state sequence is extracted to generate a spatiotemporal function of particulate matter release source intensity under the individual action of that device. This spatiotemporal function of release source intensity is coupled with the partial differential equation of turbulent diffusion in the cleanroom airflow organization to obtain the spatial evolution distribution field of particulate matter concentration under the individual action of that device. The corresponding initial pre-adjustment control vector is generated based on this spatial evolution distribution field. The initial pre-adjustment control vector for each production device only considers the particulate matter release generated by the action of that device and does not consider the influence of other devices.
[0062] Identify whether the airflow adjustment directions of the initial pre-conditioning control vectors for the same supply air terminal or return air vent are opposite. For each supply air terminal and return air vent, check the sign of the airflow change corresponding to that vent in all initial pre-conditioning control vectors. If there are two or more initial pre-conditioning control vectors with opposite signs of airflow change corresponding to the same vent, then a conflict is considered to exist between these initial pre-conditioning control vectors.
[0063] If a conflict exists, extract the extreme values of the predicted concentration extreme points for each production device. For each conflicting air outlet, identify all production devices requiring adjustment of that outlet and extract the maximum extreme value of the predicted concentration extreme points for these devices. For the same supply air terminal or return air outlet, retain only the adjustment command in the initial pre-adjustment control vector corresponding to the production device with the largest predicted concentration extreme value; discard the adjustment commands for the remaining production devices at the same supply air terminal or return air outlet, and set their airflow change to 0.
[0064] The final global pre-adjustment control vector is generated based on the processed adjustment instructions. The airflow changes after conflict resolution in all initial pre-adjustment control vectors are superimposed to obtain the total airflow change for each vent. The current airflow value of each vent is added to the total airflow change to obtain the target airflow value for that vent. The target airflow values of all vents are arranged in a preset order to form the final global pre-adjustment control vector. Multi-device conflict handling is shown in Table 3 below.
[0065] Table 3 Examples of Multi-Device Conflict Handling Wind vent number Airflow change of equipment 1 (m³ / h) Airflow variation of equipment 2 (m³ / h) Extreme values for device 1 (units / m³) Extreme values for device 2 (units / m³) Final air volume change (m³ / h) 001 +300 -200 8500 6200 +300 002 +250 +150 8500 6200 +400 003 -100 +200 8500 6200 -100 004 +200 +100 8500 6200 +300 005 +150 -150 8500 6200 +150 006 -100 +50 8500 6200 -100 Table 3 illustrates an example of multi-device conflict handling. In this example, two production devices simultaneously generate operating cycle state sequences, each producing an initial pre-adjustment control vector. Device 1 has a predicted concentration extreme value of 8500 particles / m³, while Device 2 has a predicted concentration extreme value of 6200 particles / m³. For vent 001, Device 1 requests an increase in airflow of 300 m³ / h, while Device 2 requests a decrease in airflow of 200 m³ / h, with opposite adjustment directions. Since Device 1 has a larger extreme value, its adjustment command is retained, resulting in a final airflow change of +300 m³ / h. For vent 003, Device 1 requests a decrease in airflow of 100 m³ / h, while Device 2 requests an increase in airflow of 200 m³ / h, with opposite adjustment directions. Again, Device 1's adjustment command is retained, resulting in a final airflow change of -100 m³ / h. For air outlet 005, device 1 requires an increase in airflow of 150 m³ / h, while device 2 requires a decrease in airflow of 150 m³ / h, with opposite adjustment directions. Retaining device 1's adjustment command, the final airflow change is +150 m³ / h. For air outlets 002, 004, and 006, the adjustment directions of the two devices are the same; therefore, the airflow changes are summed to obtain the final airflow change.
[0066] In one embodiment, the cleanroom has geometric dimensions of 20m long, 15m wide, and 3m high, with a cleanliness level of ISO 5. The cleanroom is divided into three zones: a process zone, a buffer zone, and an equipment zone. The process zone, located in the center of the cleanroom, has an area of 10m × 10m and is the critical area with the most stringent requirements for particulate matter concentration. The buffer zone, located around the process zone, is 2m wide and is used to isolate the process zone from the equipment zone. The equipment zone, located at the edge of the cleanroom, is used to house production equipment and auxiliary facilities. Twenty air supply terminals are evenly distributed on the top of the cleanroom, each with a rated airflow of 1200m³ / h. Ten return air vents are deployed on the lower part of the side walls of the cleanroom, each with a rated airflow of 800m³ / h. Five production devices are installed in the cleanroom: two wafer transfer devices, two process handling devices, and one material handling device.
[0067] After system startup, the IoT sensing layer begins to collect the operating status of each production device in real time. When the wafer transfer device performs a wafer transfer action, the status acquisition terminal detects a change in the value of the register inside the device's programmable logic controller and generates an action event record, including a timestamp of 2024-05-20 10:30:00, device IDW-001, action type identifier 002, and action duration of 2 seconds. The status acquisition terminal sends this action event record to the edge computing gateway. The edge computing gateway performs time synchronization processing on the action event record and adds it to the global operating cycle state sequence.
[0068] After receiving the running cycle state sequence, the data processing layer extracts the device start / stop timing and action type identifier. Based on action type identifier 002, it retrieves the release source feature map to obtain the reference release intensity time curve and spatial distribution profile function corresponding to the wafer transfer action. The peak value of the reference release intensity time curve is 3.7e+05 units / s, the rise time is 0.5s, and the fall time is 1.8s. The spatial attenuation coefficient of the spatial distribution profile function is 1.5m⁻¹. The reference release intensity time curve is aligned with the time axis according to the occurrence time of the action event 2024-05-2010:30:00 and the duration 2s. The spatial distribution profile function is mapped to the spatial coordinates (5m, 5m, 1m) of the installation position of device W-001 to generate the spatiotemporal function of the local release source intensity corresponding to this action event.
[0069] The model calculation layer obtains the current real-time differential pressure distribution and supply air velocity of the cleanroom as initial boundary conditions. The differential pressure value between the process area and the buffer zone, collected by the micro differential pressure sensor, is 12 Pa, and the differential pressure value between the buffer zone and the equipment area is 8 Pa. The average air velocity at each air supply terminal, collected by the hot-wire anemometer, is 0.45 m / s. Kalman filtering is used to fuse and smooth the original differential pressure and air velocity values to obtain the filtered state parameters. The filtered state parameters are then weighted and fused with the predicted state parameters from the previous moment, with a weighting coefficient α of 0.8, to obtain the initial boundary conditions.
[0070] The spatiotemporal function of particulate matter emission source intensity was substituted as the source term into the partial differential equation for turbulent diffusion in cleanroom airflow organization. The computational domain was spatially discretized using the finite volume method, with a grid size of 0.2 m, resulting in a 100×75×15 grid. The initial time step was set to 0.1 s. The SIMPLE algorithm was used to iteratively solve the governing equations, with 200 iterations, until the residuals converged to... The spatial evolution distribution of particulate matter concentration over the next 30 seconds can be obtained by solving the problem.
[0071] Analysis of the spatial evolution distribution field revealed a predicted concentration extreme point at time 10:30:03, located at (6m, 6m, 1.2m), with an extreme value of 8500 particles / m³, exceeding the ISO Class 5 cleanroom cleanliness threshold of 3520 particles / m³. The spatial distances between this extreme point and each air supply terminal and return air vent were calculated, identifying air vents within the flow field's influence domain as air supply terminals 001, 002, and 003, and return air vents 004, 005, and 006. The weighting coefficients for each air vent to this extreme point were calculated: air supply terminal 001 = 0.35, air supply terminal 002 = 0.30, air supply terminal 003 = 0.15, return air vent 004 = 0.10, return air vent 005 = 0.07, and return air vent 006 = 0.03.
[0072] An objective function for airflow regulation is constructed, with a regularization coefficient λ of 0.05. A sequential quadratic programming algorithm is used to solve the objective function to obtain the target airflow value for each air outlet. The target airflow for supply air terminal 001 is 1500 m³ / h, for supply air terminal 002 it is 1450 m³ / h, for supply air terminal 003 it is 1100 m³ / h, for return air outlet 004 it is 1000 m³ / h, for return air outlet 005 it is 950 m³ / h, and for return air outlet 006 it is 700 m³ / h. These target airflow values are combined to form a pre-regulation control vector.
[0073] The shortest time required for particulate matter to diffuse from the release source location (5m, 5m, 1m) to the critical zone boundary (7m, 7m, 1m) was calculated. Based on an airflow velocity of 0.45m / s, the shortest time was calculated to be approximately 4.4s. The actuator response time is approximately 1s; therefore, the execution time of the pre-adjustment control vector was set to 10:30:02, meaning the adjustment action begins 3.4s before the particulate matter reaches the critical zone.
[0074] The central controller receives the pre-adjusted control vector and decomposes it into control commands for each actuator. The variable air volume (VAV) terminal controller adjusts the opening of the dampers at the supply air terminals 001, 002, and 003 to achieve supply air volumes of 1500 m³ / h, 1450 m³ / h, and 1100 m³ / h, respectively. The return air valve actuator adjusts the opening of the return air vents 004, 005, and 006 to achieve return air volumes of 1000 m³ / h, 950 m³ / h, and 700 m³ / h, respectively. By adjusting the airflow distribution at each vent, a local airflow barrier is formed around the predicted concentration extreme point, altering the diffusion path of particulate matter and preventing its diffusion into critical areas.
[0075] After executing the pre-adjusted control vector, the optical particulate counter in the critical area collects the actual particulate concentration value in real time. At 10:30:05, the maximum actual particulate concentration in the critical area is 2800 particles / m³, which is lower than the cleanliness threshold of 3520 particles / m³. The residual between the actual and predicted concentration values is calculated, and the residual value is -700 particles / m³, which is within the preset deviation allowable range of ±20%. Therefore, no compensation correction is required. The system continues to monitor the operating status of the production equipment, waiting for the next action event to occur.
[0076] In one embodiment, when two production devices perform actions simultaneously, the system generates an initial pre-adjustment control vector for each device. Device W-001 performs a wafer transfer action, and the generated initial pre-adjustment control vector requires air supply terminal 001 to increase airflow by 300 m³ / h, air supply terminal 002 to increase airflow by 250 m³ / h, air supply terminal 003 to decrease airflow by 100 m³ / h, return air outlet 004 to increase airflow by 200 m³ / h, return air outlet 005 to increase airflow by 150 m³ / h, and return air outlet 006 to decrease airflow by 100 m³ / h. Equipment P-001 performs the opening and closing action of the process chamber. The generated initial pre-adjustment control vector requires the air supply terminal 001 to reduce the air volume by 200 m³ / h, the air supply terminal 002 to increase the air volume by 150 m³ / h, the air supply terminal 003 to increase the air volume by 200 m³ / h, the return air outlet 004 to increase the air volume by 100 m³ / h, the return air outlet 005 to reduce the air volume by 150 m³ / h, and the return air outlet 006 to increase the air volume by 50 m³ / h.
[0077] The system detected a conflict in the adjustment directions of air supply terminals 001 and 003, and return air vents 005 and 006. The extreme values of the predicted concentration extreme points for the two devices were extracted: the extreme value for device W-001 was 8500 particles / m³, and the extreme value for device P-001 was 6200 particles / m³. Since the extreme value for device W-001 was larger, the adjustment command from device W-001 to the conflicting vent was retained. For air supply terminal 001, the final airflow change was +300 m³ / h; for air supply terminal 003, the final airflow change was -100 m³ / h; for return air vent 005, the final airflow change was +150 m³ / h; and for return air vent 006, the final airflow change was -100 m³ / h. For air vents with the same adjustment direction, the airflow changes are superimposed. The final airflow change for supply air terminal 002 is +400 m³ / h, and the final airflow change for return air vent 004 is +300 m³ / h. These airflow changes are combined to form the final global pre-adjustment control vector. The central controller adjusts the airflow of each air vent according to the final global pre-adjustment control vector, achieving effective control of particulate matter release generated by the simultaneous operation of multiple devices.
Claims
1. A method for intelligent airflow control in cleanrooms based on the Internet of Things, characterized in that, include: The operation cycle state sequence of cleanroom production equipment is acquired in real time through the Internet of Things, and the operation cycle state sequence is mapped to a spatiotemporal function of particulate matter release source intensity, including: extracting the equipment start-stop sequence and action type identifier from the operation cycle state sequence; Based on the action type identifier, a preset release source feature map is retrieved to obtain the baseline release intensity and spatial distribution profile corresponding to the action type identifier; The reference release intensity and the spatial distribution profile are aligned on the time axis and mapped to spatial coordinates according to the device start-up and shutdown sequence to generate a spatiotemporal function of the particulate matter release source intensity that varies with time and spatial location; The spatiotemporal function of the particulate matter emission source intensity is coupled with the pre-established partial differential equation of cleanroom airflow organization turbulence diffusion to solve the problem, including: obtaining the current real-time pressure difference distribution and supply air velocity of the cleanroom as initial boundary conditions. Substitute the spatiotemporal function of the particulate matter emission source intensity as a source term into the partial differential equation of turbulent diffusion for cleanroom airflow organization. The finite volume method is used to perform spatiotemporal discretization and iterative calculation on the partial differential equation of turbulent diffusion in the cleanroom airflow organization with source terms until the residual converges to a preset threshold, and the spatial evolution distribution field of particulate matter concentration in future time periods is output. Based on the location and intensity of the predicted concentration extreme points in the spatial evolution distribution field, a pre-adjustment control vector is generated for each air supply terminal and return air inlet, including: extracting the spatial coordinates and extreme values of the predicted concentration extreme points that exceed the cleanliness threshold in the spatial evolution distribution field; Calculate the weighting coefficients of the spatial coordinates within the flow field influence domain of each air supply terminal and return air inlet; Based on the extreme value and the weighting coefficient, an air volume adjustment objective function is constructed. The air volume adjustment objective function is constrained and solved to obtain the target air volume values of each air supply terminal and return air inlet. These values are then combined to form the pre-adjustment control vector. Before the particulate matter generated by the operation of the production equipment spreads to the critical area, the local unidirectional flow direction and wind speed are changed in advance.
2. The intelligent airflow control method for cleanrooms based on the Internet of Things according to claim 1, characterized in that, After generating the spatiotemporal function of the particulate matter emission source intensity, the method further includes: identifying the time interval between adjacent action events in the operating cycle state sequence; If the time interval is less than a preset time window threshold, the spatiotemporal functions of particulate matter release source intensity corresponding to adjacent action events are superimposed and merged, and the time step of the cleanroom airflow organization turbulent diffusion partial differential equation is dynamically adjusted during coupled solution, wherein the time step is negatively correlated with the rate of change of the release source intensity after superposition and merging.
3. The intelligent airflow control method for cleanrooms based on the Internet of Things according to claim 2, characterized in that, Before the step of retrieving a preset release source feature map based on the action type identifier, the method further includes: collecting historical particulate matter concentration time series data of the cleanroom production equipment when it historically performed each action type; Cluster analysis was performed on the historical particulate matter concentration time series data according to action type, and the concentration transient response curve and spatial diffusion attenuation characteristics corresponding to each action type were extracted; The peak value of the concentration transient response curve is fitted with the envelope of the spatial diffusion attenuation feature to form a new benchmark release intensity and spatial distribution profile, and the release source feature map is replaced and updated.
4. The intelligent airflow control method for cleanrooms based on the Internet of Things according to claim 3, characterized in that, The steps of obtaining the current real-time differential pressure distribution and air supply velocity of the cleanroom as initial boundary conditions include: obtaining the original differential pressure value and original air velocity value collected by the sensor array distributed in the cleanroom; Kalman filtering is used to fuse and smooth the original pressure difference value and the original wind speed value to obtain the filtered current state parameter. The current state parameters are weighted and fused with the predicted state parameters from the previous moment, and the weighted fusion result is used as the initial boundary condition and substituted into the cleanroom airflow organization turbulent diffusion partial differential equation.
5. The intelligent airflow control method for cleanrooms based on the Internet of Things according to claim 4, characterized in that, In the step of constraining and solving the objective function of air volume adjustment, the constraint conditions of the objective function of air volume adjustment include: the adjusted air volume value of each air supply terminal and return air inlet does not exceed the maximum rated air volume of the corresponding air inlet; The differential pressure gradient between adjacent zones remains within the tolerance range of the design differential pressure threshold after the pre-adjustment control vector is executed; The sum of the absolute values of the changes in each air volume value in the pre-adjustment control vector is minimized, and the concentration value of the predicted concentration extreme point in the spatial evolution distribution field is lower than the cleanliness threshold after adjustment.
6. The intelligent airflow control method for cleanrooms based on the Internet of Things according to claim 5, characterized in that, After changing the local unidirectional flow direction and wind speed in advance, the method also includes: collecting the actual particulate matter concentration value of the key area in real time, and calculating the residual between the actual particulate matter concentration value and the predicted concentration value at the corresponding position in the spatial evolution distribution field; Determine whether the residual exceeds the preset allowable deviation range. If it does, calculate the compensation air volume correction value based on the residual, superimpose the compensation air volume correction value onto the current pre-adjustment control vector, correct the air volume of the air supply terminal and return air inlet based on the superimposed control vector, and use the actual particulate matter concentration value to correct the diffusion coefficient in the turbulent diffusion partial differential equation of the cleanroom airflow organization.
7. The intelligent airflow control method for cleanrooms based on the Internet of Things according to claim 6, characterized in that, When multiple production devices in the cleanroom simultaneously generate operating cycle state sequences, an independent initial pre-adjustment control vector is generated for each production device. The system identifies whether the airflow adjustment directions of the initial pre-adjustment control vectors for the same air supply terminal or return air inlet are opposite. If they are opposite, the extreme values of the predicted concentration extreme points corresponding to each production equipment are extracted. For each air supply terminal or return air inlet with opposite airflow adjustment directions, only the adjustment instructions in the initial pre-adjustment control vector corresponding to the production equipment with the largest predicted concentration extreme value are retained. The adjustment instructions of the other production equipment for the air supply terminal or return air inlet with opposite airflow adjustment directions are discarded. The final global pre-adjustment control vector is generated based on the adjustment instructions.
Citation Information
Patent Citations
Sensor control method of air purification system
CN119393890A
Intelligent control system and method based on air purification
CN119755788A
