Intelligent constant humidity control FFU integrated system and dust-free workshop application method thereof

Abstract

The application discloses an intelligent constant-humidity control FFU integrated system and a dust-free workshop application method thereof, and relates to the technical field of workshop application.The method comprises the following steps: an environment sensing unit is used for collecting the internal and external environment parameters and the equipment running state in the workshop in real time; a prediction unit is used for training a model with time series prediction ability by fusing historical production data and real-time environment fluctuation characteristics, so as to output the temperature and humidity change trend of each partition in the workshop in a future time period and a key adjustment time window; a decision unit is used for dynamically selecting a target operation strategy from a process priority mode, an energy-saving mode and an emergency mode based on real-time production process requirements, environment sensitivity thresholds and energy consumption constraints; and a parameter optimization unit is used for adjusting the wind field parameters of the FFU unit under the target operation strategy, so as to maintain the uniform distribution of temperature and humidity and minimize the system energy consumption; the adjustment comprises fan rotating speed gradient configuration, air outlet angle collaborative control and air circulation path optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of workshop application, in particular to an intelligent constant humidity control FFU integrated system and a dust-free workshop application method thereof. BACKGROUND

[0002] In modern industrial production, special environments such as dust-free workshops and constant temperature and humidity workshops have strict requirements for humidity control. For example, in places such as pharmaceutical dust-free workshops, food dust-free workshops, and medical device dust-free workshops, accurate control of humidity is of great significance to ensure product quality and prevent the growth of microorganisms.

[0003] However, existing humidity control technology often has some shortcomings when applied to dust-free workshops. On the one hand, traditional humidity control equipment and FFU are usually operated independently, lacking effective cooperation, resulting in less than ideal accuracy and stability of humidity control; on the other hand, monitoring and adjusting temperature and humidity often rely on manual control, which is not timely enough to meet the requirements of some production environments for humidity.

[0004] To address the above problems, no effective solutions have been proposed so far. SUMMARY

[0005] Embodiments of the present application provide an intelligent constant humidity control FFU integrated system and a dust-free workshop application method thereof to solve the above technical problems.

[0006] The present application provides an intelligent constant humidity control FFU integrated system, comprising:

[0007] An environment perception unit composed of a temperature and humidity sensor, a production equipment heat load monitoring module, and a meteorological data interface, for real-time collection of internal and external environmental parameters and equipment operating status in the workshop;

[0008] A prediction unit for training a model with time series prediction capability by fusing historical production data and real-time environmental fluctuation characteristics to output temperature and humidity change trends and key adjustment time windows for each partition of the workshop in the future time period;

[0009] A decision unit for dynamically selecting a target operating strategy from a process priority mode, an energy saving mode, and an emergency mode based on real-time production process requirements, environmental sensitivity thresholds, and energy consumption constraints;

[0010] A parameter optimization unit for adjusting wind field parameters of the FFU unit under the target operating strategy using an optimization algorithm to minimize system energy consumption while maintaining uniform distribution of temperature and humidity; the adjustment includes fan speed gradient configuration, outlet angle coordinated control, and airflow circulation path optimization.

[0011] Further, the intelligent constant humidity control FFU integrated system further comprises:

[0012] The linkage unit integrates the UWB personnel positioning system and the infrared heat source detection device, and is used for triggering the FFU power step attenuation and the temperature and humidity adjusting device hibernation instruction when it is detected that there is no personnel activity in the target partition for 10 minutes continuously and the equipment heat radiation intensity is lower than a threshold value;

[0013] The feedback unit collects the environment response data and the equipment execution state of each partition, and constructs a deviation tracing model through a Bayesian network, and feeds back to the prediction unit and the parameter optimization unit for model iterative updating.

[0014] Further, the production equipment heat load monitoring module in the environment perception unit is connected with the PLC controller through an industrial bus, and can analyze the equipment start-stop signal, the material processing amount and the heat load change curve in real time; the environment perception unit comprises:

[0015] The redundancy checking network adopts a hexagonal honeycomb topology structure to cover each partition of the workshop, and each partition is configured with three groups of cross-checked temperature and humidity sensing nodes;

[0016] The weather data cleaning module performs wavelet transform denoising and dynamic time alignment on the temperature and humidity, air pressure data of the external weather station;

[0017] The production equipment interface module collects the equipment running frequency, motor torque and cooling water flow data in the PLC controller in real time through the OPCUA protocol.

[0018] Further, the prediction unit comprises:

[0019] The space-time hypergraph network module models each partition of the workshop as a hypergraph node, connects the partitions having an environmental propagation relationship through a hyperedge, and extracts the temperature and humidity diffusion features of the cross-partitions by using a space-time convolution;

[0020] The time window predictor analyzes the time interval of the historical adjusting event and the Poisson distribution law of the production activity based on a hidden Markov jump diffusion model, and predicts the trigger probability and duration of the future key adjusting time window;

[0021] The feature fusion module performs tensor fusion on the spatial correlation features output by the hypergraph network and the time window prediction result, to generate a partition temperature and humidity evolution graph with an uncertainty interval.

[0022] Further, the prediction unit further comprises:

[0023] The physical constraint network generates extreme working condition training data with physical rationality by fusing the equipment thermodynamic equation and the fluid mechanics boundary condition, including the equipment sudden overload, the ventilation opening blockage and the material moisture absorption mutation scene.

[0024] a search module, based on real-time collected partition layout change data, reconstructs the hypergraph network topology, and a search strategy of the search module includes:

[0025] a differential architecture search algorithm is used to dynamically optimize the hyperedge connection mode, so that the hypergraph network adapts to the partition reorganization;

[0026] a topological persistence constraint is introduced to ensure smooth transition of the network structure between adjacent production batches;

[0027] a meta-knowledge distillation engine is introduced to encode the implicit response law of the workshop environment in the historical prediction model into a lightweight rule set as the model initialization prior in the new scene.

[0028] Further, the decision unit comprises:

[0029] an attention reinforcement learning framework, which inputs the hypergraph features output by the prediction unit and the time window probability as states, extracts environment sensitivity features of different spatial scales through a hierarchical attention mechanism;

[0030] a fuzzy differential game controller, which constructs a game model of the production department and the equipment department, and solves the Pareto optimal strategy by using a fuzzy differential equation;

[0031] a risk-constrained policy distillation module, which extracts time window constraint rules from the event-driven model of the prediction unit, encodes them into action masks of reinforcement learning, and limits the selection of high-risk strategies.

[0032] Further, the decision unit further comprises:

[0033] a heterogeneous policy fusion module, configured to: extract key propagation path features from the hypergraph network of the prediction unit, construct an environment fluctuation propagation probability graph; adopt a random walk graph attention mechanism to simulate the propagation influence of different decision strategies in the probability graph, and generate a risk heat map; and dynamically adjust the utility function of the game model by fusing the risk heat map and real-time production yield data through a double-flow gating network;

[0034] a causal reinforcement intervention module, which performs a closed-loop operation chain: when the feedback unit detects environmental deviation, an counterfactual causal reasoning algorithm is used to deduce an optimal intervention action set; a virtual twin decision environment is constructed to pre-act the long-term impact of the intervention action in the digital twin; and the optimized strategy in the virtual environment is migrated to the physical system by remodeling the strategy through a policy gradient.

[0035] Further, the linkage unit comprises:

[0036] a radar micro-motion perception system, which distinguishes between temporary stay and long-term departure of personnel through Doppler feature analysis;

[0037] a thermal inertia prediction module, which constructs a thermal radiation decay model after the device is shut down based on a material thermal conduction equation;

[0038] a power mapping module, which calculates an FFU power decay curve according to the thermal inertia model to control the temperature and humidity fluctuations to not exceed the process threshold.

[0039] Further, the deviation traceability analysis module of the feedback unit comprises:

[0040] a causal discovery engine, which adopts a PC algorithm based on conditional independence to identify key causal links of the environmental control deviation;

[0041] an incremental knowledge distillation component, which converts the traceability result into a lightweight knowledge graph and injects it into the prediction model adjustment process.

[0042] The present application provides a dust-free workshop application method of an intelligent constant humidity control FFU integrated system, comprising:

[0043] deploying temperature and humidity sensors, production equipment heat load monitoring modules and meteorological data interfaces inside the dust-free workshop to collect environmental parameters inside and outside the workshop and equipment operating states in real time;

[0044] fusing historical production data and real-time environmental fluctuation characteristics, using a model with time series prediction ability generated by training to output the temperature and humidity change trend of each subzone of the workshop and the key adjustment time window in the future time period;

[0045] based on real-time production process requirements, environmental sensitivity thresholds and energy consumption constraints, dynamically selecting a target operation strategy from a process priority mode, an energy saving mode and an emergency mode;

[0046] using an optimization algorithm to adjust the wind field parameters of the FFU unit under the target operation strategy, including fan speed gradient configuration, outlet angle cooperative control and air circulation path optimization, while maintaining uniform distribution of temperature and humidity, minimizing the energy consumption of the intelligent constant humidity control FFU integrated system.

[0047] Based on the embodiments provided by the present application, the FFU unit and the temperature and humidity adjusting device are coordinated, a dynamic prediction model is constructed through multi-source data fusion (environmental parameters, production plan, meteorological data), the humidity control accuracy is improved, and the requirements for humidity fluctuations in key areas in the GMP specification of the pharmaceutical workshop are met. Using the parameter optimization module, the system energy consumption is reduced under the premise of ensuring uniformity of temperature and humidity. Based on the production process requirements and the environmental sensitivity threshold, the system can timely complete the operation strategy switching (such as switching from the process priority mode to the emergency mode), the response speed is significantly improved compared with manual operation, and the wafer oxidation defects caused by sudden humidity changes in the semiconductor workshop are effectively avoided. BRIEF DESCRIPTION OF DRAWINGS

[0048] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0049] Figure 1 An optional structure diagram of a smart constant humidity control FFU integrated system according to an embodiment of the application;

[0050] Figure 2 An optional flowchart of a closed-loop operation chain executed by a causal reinforcement intervention module according to an embodiment of the application;

[0051] Figure 3 An optional flowchart of a dust-free workshop application method of a smart constant humidity control FFU integrated system according to an embodiment of the application.

[0052] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0054] Optionally, as shown in the drawings, Figure 1 The application provides a smart constant humidity control FFU integrated system, which comprises:

[0055] The environment perception unit 101 is composed of a temperature and humidity sensor, a production equipment heat load monitoring module and a meteorological data interface, and is used for collecting the internal and external environmental parameters and the equipment running state in the workshop in real time.

[0056] The temperature and humidity sensor is multiple, and is uniformly distributed at different positions in the workshop, so as to realize comprehensive monitoring of the temperature and humidity in the workshop.

[0057] The prediction unit 102 is used for training a model with time series prediction ability by fusing historical production data and real-time environmental fluctuation characteristics, so as to output the temperature and humidity change trend and the key adjustment time window of each partition in the workshop in the future time period.

[0058] The decision unit 103 is used for dynamically selecting a target running strategy from a process priority mode, an energy saving mode and an emergency mode based on real-time production process requirements, environmental sensitivity thresholds and energy consumption constraints.

[0059] The parameter optimization unit 104 is configured to adopt an optimization algorithm to adjust the wind field parameters of the FFU unit under the target operation strategy, so as to minimize the system energy consumption while maintaining the uniform distribution of temperature and humidity; the adjustment includes fan speed gradient configuration, outlet angle collaborative control and airflow circulation path optimization.

[0060] Based on the embodiments provided in the present application, the FFU unit and the temperature and humidity adjusting device are coordinated, a dynamic prediction model is constructed through multi-source data fusion (environmental parameters, production plan, meteorological data), the humidity control accuracy is improved, the requirements for humidity fluctuation in the key area in the GMP specification of the pharmaceutical workshop are met. The parameter optimization module is adopted to reduce the system energy consumption under the premise of ensuring the uniformity of temperature and humidity. Based on the production process requirements and the environmental sensitivity threshold, the system can timely complete the operation strategy switching (such as switching from the process priority mode to the emergency mode), the response speed is significantly improved compared with manual operation, and the wafer oxidation defects caused by sudden humidity change in the semiconductor workshop are effectively avoided.

[0061] Further, the intelligent constant humidity control FFU integrated system further comprises:

[0062] The linkage unit integrates the UWB personnel positioning system and the infrared heat source detection device, and is configured to trigger the FFU power ladder attenuation and the temperature and humidity adjusting device hibernation instruction of a target subzone when it is detected that there is no personnel activity in the subzone for 10 minutes continuously and the device heat radiation intensity is lower than a threshold value;

[0063] The threshold value refers to a standard value for judging whether the device is in a low heat load state. When the device heat radiation intensity is lower than the threshold value, combined with the condition that there is no personnel activity, the system considers that the heat load of the region is low, and energy-saving measures can be triggered.

[0064] For example, in an electronic manufacturing workshop, when the production equipment is in a standby or low-power operation state, the heat radiation intensity thereof can be low. Assuming that the device heat radiation intensity threshold value is set to 50W / m 2 When the infrared heat source detection device detects that the device heat radiation intensity in the target subzone is lower than 50W / m 2 for 10 minutes continuously, and at the same time, the UWB personnel positioning system detects that there is no personnel activity in the region, the system triggers the FFU power ladder attenuation and the temperature and humidity adjusting device hibernation instruction of the subzone to reduce the energy consumption.

[0065] Based on the embodiments provided in the present application, the UWB positioning and infrared heat source collaborative judgment mechanism avoids the false triggering of a single sensor (such as false judgment caused by device residual heat), and ensures the reliability of the hibernation instruction.

[0066] In some high-precision instrument manufacturing workshops, the temperature and humidity control of the environment is more stringent, and the device heat radiation intensity threshold value can be set to be lower, such as 30W / m2 , to ensure that the tiny heat changes of the equipment will not affect the environment. In some workshops with relatively loose requirements for temperature and humidity, this threshold may be appropriately increased, such as 70W / m 2 , to adapt to different production needs.

[0067] The feedback unit collects the environmental response data of each subzone and the execution state of the equipment, builds a deviation tracing model through a Bayesian network, and feeds back to the prediction unit and the parameter optimization unit for model iterative updating.

[0068] Further, the production equipment heat load monitoring module in the environment perception unit is connected to the PLC controller through an industrial bus, and the device start-stop signal, material processing capacity and heat load change curve are analyzed in real time; the environment perception unit comprises:

[0069] The redundant verification network adopts a hexagonal honeycomb topology structure to cover each subzone of the workshop, and each subzone is configured with three groups of cross-verified temperature and humidity sensor nodes.

[0070] The meteorological data cleaning module performs wavelet transform denoising and dynamic time alignment on the temperature and humidity, air pressure data of the external weather station.

[0071] The production equipment interface module collects the device running frequency, motor torque and cooling water flow data in the PLC controller in real time through the OPCUA protocol.

[0072] Based on the embodiments provided in the present application, the hexagonal honeycomb topology network suppresses the influence of local interference (such as air flow disturbance caused by personnel walking) on data collection through three groups of sensor cross verification.

[0073] Further, the prediction unit comprises:

[0074] The spatio-temporal hypergraph network module models each subzone of the workshop as a hypergraph node, connects subzones with environmental propagation relationships through hyperedges, and extracts cross-subzone temperature and humidity diffusion features using spatio-temporal convolution;

[0075] The time window predictor analyzes the time interval of historical regulation events and the Poisson distribution law of production activities based on a hidden Markov jump diffusion model, and predicts the trigger probability and duration of future key regulation time windows; wherein the time window predictor is event-driven;

[0076] The feature fusion module performs tensor fusion on the spatial correlation features output by the hypergraph network and the time window prediction results to generate a subzone temperature and humidity evolution map with an uncertainty interval.

[0077] Based on the embodiments provided in the present application, the spatio-temporal hypergraph network explicitly models the environmental propagation path between subzones (such as the humidity influence of the clean area on the buffer area), which improves the cross-area prediction accuracy.

[0078] Further, the prediction unit further comprises:

[0079] a physical constraint network, which generates extreme working condition training data with physical rationality by fusing inter-plant device thermodynamic equations and fluid mechanics boundary conditions, including device sudden overload, air vent blockage and material moisture absorption mutation scenarios;

[0080] a search module, which reconstructs the supergraph network topology based on real-time collected partition layout change data, and the search strategy thereof includes: the partition layout change data includes device displacement and partition adjustment;

[0081] dynamically optimize the super-edge connection mode through the differentiable architecture search algorithm, so that the supergraph network adapts to the partition reorganization;

[0082] introduce topological persistence constraints to ensure smooth transition of network structure between adjacent production batches;

[0083] introduce a meta-knowledge distillation engine to encode the inter-plant environment response law implied in the historical prediction model into a lightweight rule set as the model initialization prior in the new scenario.

[0084] In the embodiments of the present application, the dynamic supergraph topology optimization equation is as follows:

[0085]

[0086] wherein W is a supergraph adjacency matrix, and the matrix element W kn represents the environmental propagation intensity (such as humidity diffusion coefficient, temperature gradient influence, etc.) between partition k and partition n; if there are 3 partitions (lithography area, etching area, packaging area) in the plant, then W can be: wherein W 12 = 0.8 represents that the environmental propagation intensity of the lithography area to the etching area is high; ⊙ represents Hadamard product, i.e. element-wise multiplication of matrices; the role of W {(t-1)} ⊙ M is to retain the key connections of the historical topology; through Hadamard product, the system will preferentially retain the historical effective connections when optimizing the new topology W (t) ; W (t) is the edge weight tensor of the supergraph network at time t after optimization; L task is the loss function (containing space-time convolution error) of the temperature and humidity prediction task; M {(t-1)} is the mask matrix (0 / 1 binary) of the historical topology structure at time t-1; p old and p new represent the probability distribution of the supergraph structure at different times; p old is the probability distribution representing the old supergraph structure, and p newthe probability distribution representing the new hypergraph structure; λ3KL(p old ||p new ) is the KL divergence between the old and new hypergraph structure distributions; λ1, λ2, λ3 are dynamic balance coefficients;

[0087] In practical applications, the values of λ1, λ2, and λ3 need to be adjusted according to specific application scenarios and optimization goals. For example, in a scenario where the accuracy of temperature and humidity prediction is required to be high, the value of λ1 can be set relatively large to focus more on the loss function of the prediction task. In a scenario where both the historical topological structure and the difference between the old and new structures need to be considered, the weights of λ2 and λ3 can be appropriately increased.

[0088] Suppose in a temperature and humidity control application in a dust-free workshop, we want to ensure the accuracy of temperature and humidity prediction while also considering the stability of the historical topological structure and the smooth transition between the old and new structures. We can set the values of λ1, λ2, and λ3 to 0.5, 0.3, and 0.2, respectively. This means that we give a higher weight to the loss function of the temperature and humidity prediction task, while also appropriately considering the mask matrix of the historical topological structure and the difference between the old and new hypergraph structure distributions. Such a setting helps to find a balance point in the optimization process, so that the edge weight tensor W(t) of the hypergraph network can meet the requirements of prediction accuracy while maintaining the stability and continuity of the structure.

[0089] In the embodiments of the present application,

[0090] H t = σ(W s × Conv3D(F {t-k:t} ) + W h × G(E {ij} ))

[0091] where H t ∈ R N×d represents the fusion feature matrix of each partition at time t (N is the number of partitions and d is the feature dimension); F {t-k:t} is the original temperature and humidity data of the past k time slices; i, j represent the node index in the hypergraph, which is used to define the node set connected by the hyperedge; E {ij} is the partition set connected by the hyperedge; G() is a hypergraph aggregation function that calculates the environmental propagation strength between partitions; W s , W h are learnable weight matrices (initialized through meta-knowledge distillation) for linear transformation of the original temperature and humidity data and the partition set connected by the hyperedge; W s has a size of dxd, where d is the feature dimension (e.g., 8), and can be initialized as a small random value, such as W s= [[0.1, 0.2,..., 0.8], [0.2, 0.3,..., 0.9],..., [0.8, 0.9,..., 0.5]] ; W h of size dxd, is also initialized with random decimal values, such as W h = [[0.3, 0.4,..., 0.6], [0.4, 0.5,..., 0.7],..., [0.6, 0.7,..., 0.4]] ;

[0092] σ is an activation function, for example, in the temperature and humidity control application of the dust-free workshop, ReLU can be selected as the activation function, which is defined as σ(x) = max(0, x), that is, when the input value is greater than 0, the value is output, otherwise 0 is output;

[0093] Based on the embodiments provided in the present application, the differentiable architecture search can realize dynamic adjustment of hypergraph topology, adapt to the reorganization demand of the workshop (such as adding an isolation area), avoid the retraining cost of the traditional fixed topology model, solve the problem that the traditional model cannot capture the propagation across the partition environment, for example, in the pharmaceutical workshop, the lagging influence of the humidity change in the filling area on the packaging area.

[0094] Further, the decision unit comprises:

[0095] The attention reinforcement learning framework inputs the hypergraph features output by the prediction unit and the time window probability as states, extracts environment sensitivity features of different spatial scales through a hierarchical attention mechanism; wherein the different spatial scales include the partition level, the device level and the workshop level.

[0096] The fuzzy differential game controller constructs a game model between the production department (pursuing maximum yield) and the device department (pursuing minimum energy consumption), and solves the Pareto optimal strategy by using a fuzzy differential equation;

[0097] The risk constraint type policy distillation module extracts the time window constraint rules from the event-driven model of the prediction unit, encodes them into the action mask of reinforcement learning, and limits the selection of high-risk strategies.

[0098] Based on the embodiments provided in the present application, the hierarchical attention mechanism focuses on the key partition (such as the sterile filling area), reduces the energy consumption of the edge area control while ensuring the accuracy of the core area.

[0099] Further, the decision unit further comprises a heterogeneous policy fusion module and a causal reinforcement intervention module:

[0100] The isomorphism strategy fusion module is configured to perform the following operations: extracting key propagation path features from the hypergraph network of the prediction unit, and constructing an environmental fluctuation propagation probability graph; using a random walk graph attention mechanism to simulate the propagation influence of different decision-making strategies in the probability graph, and generating a risk heat map; and dynamically adjusting the utility function of the game model by fusing the risk heat map and real-time production yield data through a double-flow gated network.

[0101] As shown in Figure 2 , the causal reinforcement intervention module is configured to perform the following closed-loop operation chain:

[0102] S201, when the feedback unit detects environmental deviation, using counterfactual causal reasoning algorithm to deduce the optimal intervention action set;

[0103] S202, constructing a virtual twin decision environment, and preforming the long-term impact of intervention actions in the digital twin;

[0104] S203, by strategy gradient remodeling strategy, migrating the optimized strategy in the virtual environment to the physical system.

[0105]

[0106] where Δθ is the policy network parameter update amount, output by the strategy gradient remodeling; π virt is the virtual twin environment strategy, generated by the digital twin; Q fact is the actual action value function, representing the value of the actual executed action a in state s; Q cf is the counterfactual action value function, representing the value of the counterfactual action a cf in state s; s is the state, composed of the hypergraph features and the time window probability output by the prediction unit; a is the action, output by the fuzzy differential game controller, used to control the operation of the FFU unit; a cf is the counterfactual action, used for comparative analysis in the virtual environment; η is the migration learning rate, which is adaptively adjusted according to the stability of the workshop, and can be taken as 0.001; D is the deviation data set, provided by the causal discovery engine;

[0107] Based on the embodiments provided in this application, through the counterfactual causal reasoning algorithm, the system can deduce the optimal intervention action set, thereby improving the accuracy and reliability of decision-making. Compared with traditional experience-based decision-making methods, this method can more scientifically predict the consequences of different decisions, avoiding the risks brought by blind decision-making. The virtual twin decision-making environment is constructed, enabling the system to pre-echo the long-term impact of intervention actions in the digital twin. This enables the system to evaluate the effects of decisions in advance and adjust strategies in a timely manner, enhancing the adaptability to complex environmental changes. Through strategy gradient remodeling, the optimized strategy in the virtual environment is transferred to the physical system, achieving more efficient energy management and more accurate environmental control. The introduction of the causal reinforcement intervention module, combined with counterfactual causal reasoning and virtual twin technology, forms an innovative decision optimization mechanism. This mechanism has a certain pioneering nature in intelligent environmental control systems, providing new ideas and methods for decision optimization of similar systems. The migration learning rate η can be adaptively adjusted according to the stability of the workshop, ensuring the rationality and effectiveness of strategy updating. This adaptive mechanism enables the system to maintain good performance under different working conditions, further enhancing the intelligent level of the system.

[0108] Further, the linkage unit comprises:

[0109] The radar micro-motion sensing system distinguishes between short-term stays and long-term departures of personnel through Doppler feature analysis;

[0110] The thermal inertia prediction module constructs a thermal radiation decay model after the device is shut down based on the material thermal conduction equation;

[0111] The power mapping module calculates the FFU power decay curve according to the thermal inertia model to control the temperature and humidity fluctuations within the process threshold.

[0112] In a chip manufacturing workshop, accurate control of temperature and humidity is crucial. The chip manufacturing process has almost stringent requirements for the cleanliness and stability of the environment. Fluctuations in temperature and humidity not only affect the normal operation of production equipment, but also may cause irreversible damage to the quality and performance of chips.

[0113] The temperature control of the chip clean room is usually set within the range of 22℃±2℃, and excessively high or low temperature may cause performance degradation or even failure of the equipment. Stable temperature control helps to reduce the thermal stress of chips during the manufacturing process and reduce the product defect rate.

[0114] The humidity control of the chip clean room is usually set within the range of 45%±5%, and a high humidity environment helps to reduce the accumulation of static electricity and reduce the damage of electrostatic discharge to chips.

[0115] Based on the embodiments provided in the present application, the thermal inertia model predicts the residual heat decay curve of the equipment, realizes the gradual adjustment of the FFU power, and avoids the influence of sudden changes in temperature and humidity on product quality.

[0116] Further, the deviation traceability analysis module of the feedback unit comprises:

[0117] The causal discovery engine identifies the key causal link of the environmental control deviation by using the PC algorithm based on conditional independence.

[0118] The incremental knowledge distillation component converts the traceability result into a lightweight knowledge graph and injects it into the prediction model adjustment process.

[0119] Based on the embodiments provided in the present application, the incremental knowledge distillation retains the core features of the historical model (such as the environmental response difference between the early shift and the late shift), preventing knowledge forgetting in the model iteration.

[0120] As Figure 3 shown, optionally, the present application provides a dust-free workshop application method of the intelligent constant humidity control FFU integrated system, comprising:

[0121] S301, deploying temperature and humidity sensors, production equipment heat load monitoring modules and meteorological data interfaces inside the dust-free workshop, and collecting real-time environmental parameters and equipment operating states inside and outside the workshop;

[0122] S302, fusing historical production data and real-time environmental fluctuation characteristics, using the model with time series prediction ability generated by training to output the temperature and humidity change trend of each partition of the workshop in the future time period and the key adjustment time window;

[0123] S303, based on real-time production process requirements, environmental sensitivity thresholds and energy consumption constraints, dynamically selecting a target operation strategy from the process priority mode, energy saving mode and emergency mode;

[0124] S304, using an optimization algorithm to adjust the FFU unit under the target operation strategy, including fan speed gradient configuration, air outlet angle cooperative control and air circulation path optimization, while maintaining uniform distribution of temperature and humidity, minimizing the energy consumption of the intelligent constant humidity control FFU integrated system.

[0125] The intelligent constant humidity control FFU integrated system provided in the present application can be applied to the following engineering projects: dust-free workshop, constant temperature and humidity workshop, high-precision constant temperature and humidity laboratory, medical device dust-free workshop, food dust-free workshop, pharmaceutical dust-free workshop, health product dust-free workshop, CV dust-free workshop, coating dust-free workshop, 100,000-level dust-free workshop, 10,000-level dust-free workshop and 100-level dust-free workshop.

[0126] In the embodiments of the present application, the system and method are applied in a dust-free workshop. The workshop is installed with high-efficiency air ports, FFUs, air shower rooms, cargo shower rooms, weighing rooms, laminar flow hoods, transfer windows, air shower transfer windows and other equipment. The temperature and humidity in the workshop are accurately controlled by the intelligent constant humidity control FFU integrated system. When people and materials enter and exit the workshop, the air shower room and the cargo shower room can effectively remove dust and microorganisms attached to the surface, and the weighing room provides a stable environment for accurate weighing. The transfer window and the air shower transfer window avoid the influence of external air on the temperature, humidity and cleanliness in the workshop during the transfer of goods.

[0127] Based on the embodiments provided in the present application, the intelligent constant humidity control FFU integrated system and the dust-free workshop application method thereof can effectively solve the problems existing in the prior art in actual application, have significant advantages and wide application prospect.

[0128] It should be noted that in the present application, the embodiments implemented on the side of the intelligent constant humidity control FFU integrated system can be mutually referenced with the embodiments implemented on the side of the dust-free workshop application method of the intelligent constant humidity control FFU integrated system, and the present application will not be repeated here.

[0129] The above is only the preferred embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent flow transformation made by using the content of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. Intelligent constant humidity control FFU integrated system, characterized by: include: The environmental sensing unit, consisting of temperature and humidity sensors, a production equipment heat load monitoring module, and a meteorological data interface, is used to collect real-time environmental parameters inside and outside the workshop and equipment operating status; The prediction unit is used to train and generate a model with time series prediction capabilities by integrating historical production data with real-time environmental fluctuation characteristics. This model outputs the temperature and humidity change trends and key adjustment time windows for each workshop partition in the future time period. A decision-making unit, which dynamically selects a target operation strategy from process priority mode, energy-saving mode, and emergency mode based on real-time production process requirements, environmental sensitivity thresholds, and energy consumption constraints; A parameter optimization unit is used to adjust the wind field parameters of the FFU unit under the target operation strategy using an optimization algorithm to minimize system energy consumption while maintaining uniform temperature and humidity distribution; the adjustment includes fan speed gradient configuration, coordinated control of air outlet angles, and optimization of airflow circulation paths; The linkage unit integrates the UWB personnel positioning system and the infrared heat source detection device. When it detects that there is no personnel activity in the target zone for 10 consecutive minutes and the thermal radiation intensity of the equipment is lower than the threshold, it triggers the FFU power step-by-step attenuation and the temperature and humidity adjustment equipment sleep command in the zone; The feedback unit collects environmental response data and equipment execution status of each partition, builds a deviation tracing model through a Bayesian network, and feeds it back to the prediction unit and the parameter optimization unit for iterative model updates; The prediction unit includes: a spatiotemporal hypergraph network module, which models each workshop partition as a hypergraph node, connects partitions with environmental communication relationships through hyperedges, and uses spatiotemporal convolution to extract temperature and humidity diffusion characteristics across partitions; a time window predictor, which analyzes the time intervals of historical regulation events and the Poisson distribution law of production activities based on the hidden Markov jump diffusion model, and predicts the triggering probability and duration of future key regulation time windows; a feature fusion module, which performs tensor fusion on the spatial correlation features output by the hypergraph network and the time window prediction results to generate a partition temperature and humidity evolution map with uncertainty intervals; Among them, the prediction unit also includes: a physical constraint network, which generates physically reasonable extreme working condition training data by integrating the thermodynamic equations of workshop equipment and fluid mechanics boundary conditions, including scenarios of sudden equipment overload, vent blockage and material moisture absorption mutation; a search module, which reconstructs the hypergraph network topology structure based on real-time collected partition layout change data. Its search strategy includes: dynamically optimizing the hyperedge connection method through a differentiable architecture search algorithm to make the hypergraph network adapt to partition reorganization; introducing topological persistence constraints to ensure a smooth transition of the network structure between adjacent production batches; introducing a meta-knowledge distillation engine to encode the workshop environment response laws implicit in the historical prediction model into a lightweight rule set as a model initialization prior in the new scenario.

2. The intelligent constant humidity control FFU integrated system according to claim 1 is characterized in that: The production equipment heat load monitoring module in the environmental perception unit is connected to the PLC controller via an industrial bus to analyze the equipment start and stop signals, material processing volume and heat load change curve in real time; the environmental perception unit includes: The redundant verification network uses a hexagonal cellular topology to cover all zones in the workshop, with each zone configured with three sets of cross-verified temperature and humidity sensor nodes; Meteorological data cleaning module, which performs wavelet transform denoising and dynamic time alignment on the temperature, humidity, and air pressure data from external weather stations; The production equipment interface module collects the equipment operating frequency, motor torque and cooling water flow data in the PLC controller in real time through the OPC UA protocol.

3. The intelligent constant humidity control FFU integrated system according to claim 1 is characterized in that: The decision-making unit includes: The attention reinforcement learning framework uses the hypergraph features and time window probabilities output by the prediction unit as state inputs, and extracts environmental sensitivity features at different spatial scales through a hierarchical attention mechanism; Fuzzy differential game controller, builds a game model between the production department and the equipment department, and uses fuzzy differential equations to solve the Pareto optimal strategy; The risk-constrained strategy distillation module extracts time window constraint rules from the event-driven model of the prediction unit and encodes them into action masks for reinforcement learning to limit the selection of high-risk strategies.

4. The intelligent constant humidity control FFU integrated system according to claim 3 is characterized in that: The decision-making unit also includes: A heterogeneous strategy fusion module is configured to: extract key propagation path features from the hypergraph network of the prediction unit to construct a probability map for environmental fluctuation propagation; employ a random walk graph attention mechanism to simulate the propagation impact of different decision strategies in the probability map to generate a risk heat map; and dynamically adjust the game model utility function by fusing the risk heat map with real-time production yield data through a dual-stream gating network. The causal reinforcement intervention module executes a closed-loop operation chain: when the feedback unit detects an environmental deviation, it uses a counterfactual causal reasoning algorithm to deduce the optimal set of intervention actions; constructs a virtual twin decision-making environment to preview the long-term impact of the intervention actions in the digital twin; and reshapes the strategy through policy gradients to migrate the optimization strategy in the virtual environment to the physical system.

5. The intelligent constant humidity control FFU integrated system according to claim 1 is characterized in that: The linkage unit comprises: The radar micro-motion sensing system uses Doppler feature analysis to distinguish between short-term stays and long-term departures of personnel; Thermal inertia prediction module, which builds a thermal radiation attenuation model after equipment shutdown based on the material heat conduction equation; The power mapping module calculates the FFU power attenuation curve based on the thermal inertia model to control temperature and humidity fluctuations to not exceed the process threshold.

6. The intelligent constant humidity control FFU integrated system according to claim 1 is characterized in that: The deviation tracing and analysis module of the feedback unit includes: Causal discovery engine, which uses a PC algorithm based on conditional independence to identify the key causal links of environmental control deviations; The incremental knowledge distillation component converts the traceability results into a lightweight knowledge graph and injects it into the prediction model adjustment process.

7. The dust-free workshop application method of the intelligent constant humidity control FFU integrated system is characterized by: include: Deploy temperature and humidity sensors, production equipment heat load monitoring modules, and meteorological data interfaces in the clean room to collect real-time environmental parameters inside and outside the workshop and equipment operating status; By integrating historical production data with real-time environmental fluctuation characteristics and using a trained model with time series prediction capabilities, the system outputs the temperature and humidity change trends and key adjustment time windows for each workshop partition in the future time period. Dynamically select the target operation strategy from process priority mode, energy-saving mode and emergency mode based on real-time production process requirements, environmental sensitivity thresholds and energy consumption constraints; An optimization algorithm is used to adjust the wind field parameters of the FFU unit under the target operation strategy, including fan speed gradient configuration, coordinated control of air outlet angles, and airflow circulation path optimization, to minimize the energy consumption of the intelligent constant humidity control FFU integrated system while maintaining uniform temperature and humidity distribution; Using the UWB personnel positioning system and infrared heat source detection device, when it is detected that there is no human activity in the target zone for 10 consecutive minutes and the equipment heat radiation intensity is lower than the threshold, the FFU power of the zone is triggered to step down and the temperature and humidity adjustment equipment is put into sleep mode; Collect environmental response data and device execution status from each partition, build a deviation tracing model using a Bayesian network, and iteratively update the model; Among them, the fusion of historical production data and real-time environmental fluctuation characteristics uses a trained model with time series prediction capabilities to output the temperature and humidity change trends and key adjustment time windows of each partition of the workshop in the future time period, including: modeling each partition of the workshop as a hypergraph node, connecting partitions with environmental propagation relationships through hyperedges, and using spatiotemporal convolution to extract cross-partition temperature and humidity diffusion characteristics; based on the hidden Markov jump diffusion model, analyzing the time intervals of historical adjustment events and the Poisson distribution law of production activities, and predicting the triggering probability and duration of future key adjustment time windows; tensor fusion of the spatial correlation features output by the hypergraph network and the time window prediction results to generate a partition temperature and humidity evolution map with uncertainty intervals; Among them, the fusion of historical production data and real-time environmental fluctuation characteristics uses a model with time series prediction capabilities generated through training to output the temperature and humidity change trends and key adjustment time windows of each partition of the workshop in the future time period. It also includes: by fusing the thermodynamic equations of workshop equipment and fluid mechanics boundary conditions, generating physically reasonable extreme working condition training data, including equipment sudden overload, vent blockage and material moisture absorption mutation scenarios; reconstructing the hypergraph network topology structure based on real-time collected partition layout change data, and its search strategy includes: dynamically optimizing the hyperedge connection method through a differentiable architecture search algorithm to make the hypergraph network adapt to partition reorganization; introducing topological persistence constraints to ensure a smooth transition of the network structure between adjacent production batches; introducing a meta-knowledge distillation engine to encode the workshop environment response laws implicit in the historical prediction model into a lightweight rule set as a model initialization prior in the new scenario.

Images