Cleanroom environment monitoring method and system

Abstract

The application discloses a clean room environment monitoring method and system, and relates to the technical field of environment detection; multi-dimensional environment parameter data is obtained based on preset periodic synchronous acquisition of temperature, humidity, particulate matter concentration and microbial indexes in a target clean room; the multi-dimensional environment parameter data is transmitted to a data processing terminal; so that the data processing terminal carries out denoising processing on the multi-dimensional environment parameter data through a preset adaptive filtering algorithm to obtain effective multi-dimensional environment parameter data; the effective multi-dimensional environment parameter data is substituted into a clean room environment anomaly recognition model to obtain a judgment result of whether there is parameter anomaly, and alarm and control are carried out according to the judgment result. Through real-time dynamic adjustment of filtering parameters by the adaptive filtering algorithm and replacement of fixed threshold comparison by the clean room environment anomaly recognition model, invalid alarm caused by slight fluctuation is avoided, alarm redundancy and misjudgment are greatly reduced, and therefore the accuracy and monitoring efficiency of environment anomaly recognition are comprehensively improved.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring technology, specifically relating to cleanroom environmental monitoring methods and systems. Background Technology

[0002] Cleanrooms are core production sites for high-tech industries such as precision manufacturing, biomedicine, medical devices, and semiconductor processing. The stability and cleanliness of environmental parameters such as temperature, humidity, particulate matter concentration, and microbial content directly determine product quality, yield, and the safety of production operations. If abnormal cleanroom environmental parameters occur and cannot be detected and addressed in a timely manner, it can easily lead to problems such as batch product contamination, process failure, and production shutdown. Therefore, high-precision, high-reliability, and intelligent cleanroom environmental monitoring technology is the key to ensuring the stable operation of cleanrooms.

[0003] During the actual operation of a cleanroom, external interferences such as equipment operation, airflow circulation, and personnel operations can cause a large amount of noise and fluctuation in the raw data collected by sensors. Traditional fixed-parameter filtering algorithms cannot adapt to the monitoring needs of different cleanroom operating levels and different production conditions. The filtering parameters are fixed and lack adaptability. The noise removal effect varies for different types of environmental parameters, and it cannot effectively remove invalid interference data. As a result, the accuracy of the monitoring data used for analysis is low, which seriously affects the accuracy of subsequent environmental anomaly identification.

[0004] Furthermore, existing cleanroom environmental anomaly identification mechanisms are simplistic, mostly relying on fixed threshold comparisons. These thresholds cannot be dynamically adjusted based on real-time fluctuations in environmental parameters, ignoring the dynamic nature of cleanroom environments. Fixed threshold judgment modes are prone to missed or false alarms. On one hand, slight parameter fluctuations can easily trigger invalid alarms, resulting in alarm redundancy. On the other hand, latent or gradual anomalies caused by multi-parameter coupling cannot be effectively identified, ultimately leading to low efficiency in cleanroom environmental monitoring. Summary of the Invention

[0005] The purpose of this invention is to solve the problems of low accuracy and overall efficiency in cleanroom environmental monitoring caused by noise fluctuations in collected data due to on-site interference, and the inability of fixed threshold identification to adapt to dynamic environmental changes, which easily leads to false alarms and missed alarms. Therefore, this invention proposes a cleanroom environmental monitoring method and system.

[0006] In a first aspect of this invention, a method for monitoring a cleanroom environment is first proposed, the method comprising: Multidimensional environmental parameter data are obtained by synchronously collecting temperature, humidity, particulate matter concentration and microbial indicators in the target clean room based on a preset cycle; The multidimensional environmental parameter data is transmitted to a data processing terminal so that the data processing terminal can perform noise reduction processing on the multidimensional environmental parameter data using a preset adaptive filtering algorithm to obtain effective multidimensional environmental parameter data. The effective multidimensional environmental parameter data are substituted into the cleanroom environment anomaly identification model to obtain the judgment result of whether there is parameter anomaly, and alarms and controls are performed based on the judgment result.

[0007] Optionally, multidimensional environmental parameter data, including temperature, humidity, particulate matter concentration, and microbial indicators in the target cleanroom, can be obtained by synchronously collecting these data at preset intervals. In the target cleanroom, multiple multi-sensor collaborative acquisition nodes are deployed in a distributed manner according to the preset spatial grid layout rules. Each collaborative acquisition node integrates a temperature sensor, a humidity sensor, a particulate matter sensor, and a microbial sensor. Each collaborative acquisition node synchronizes its clock through a unified time synchronization module. During the sampling time within the same preset period, each sensor is triggered to collect data synchronously, and each set of collected data is marked with a timestamp, forming multidimensional environmental parameter data with time synchronization marks.

[0008] Benefits: Each node is clock-synchronized via a unified time synchronization module, and synchronous sampling is triggered within the same preset period. Each data set is timestamped, forming multi-dimensional environmental parameter data with precise time synchronization markers, ensuring strict alignment of data from different locations and with different parameters on the time axis. This provides high-quality, spatiotemporally consistent input for subsequent adaptive filtering algorithms, enabling them to more accurately distinguish between global fluctuations and local noise, and strengthening their ability to analyze multi-parameter coupling relationships and gradual trends. Simultaneously, by incorporating this highly spatiotemporally aligned multi-dimensional data into the anomaly identification model, latent and gradual anomalies can be accurately identified from both spatial distribution and temporal evolution dimensions, fundamentally reducing missed and false positives caused by data asynchrony, and significantly improving the accuracy and efficiency of cleanroom environmental monitoring.

[0009] Optionally, the multidimensional environmental parameter data can be transmitted to a data processing terminal in real time via a wired or wireless communication link located in the cleanroom.

[0010] Optionally, the multidimensional environmental parameter data can be denoised using a preset adaptive filtering algorithm to obtain effective multidimensional environmental parameter data, including: Determine the current operating level of the target cleanroom, match an adaptive filtering algorithm according to the operating level, and determine the corresponding adaptive filtering parameters; Based on the adaptive filtering algorithm and adaptive filtering parameters, noise reduction processing is performed on temperature, humidity, particulate matter concentration, and microbial indicators in the multidimensional environmental parameter data one by one to obtain effective multidimensional environmental parameter data.

[0011] Benefits: Differentiated denoising is applied to temperature, humidity, particulate matter concentration, and microbial indicators individually. This allows for precise filtering based on the unique signal characteristics and noise patterns of each parameter, effectively eliminating invalid interference caused by equipment start-up and shutdown, airflow disturbances, and personnel operations. Simultaneously, it fully utilizes the high-quality spatiotemporally aligned data provided by preceding spatiotemporal synchronous acquisition, preserving the most effective signals reflecting real environmental changes. The resulting high-precision, effective, multi-dimensional environmental parameter data provides reliable input for subsequent anomaly identification models, significantly reducing missed and false positives caused by noise residue or signal distortion at the source of data quality, thus significantly improving the accuracy and efficiency of cleanroom environmental monitoring.

[0012] Optionally, substituting the effective multidimensional environmental parameter data into the cleanroom environment anomaly identification model to obtain a judgment result on whether parameter anomalies exist includes: The cleanroom environment anomaly identification model includes a dynamic threshold analysis sub-model and an anomaly pattern identification sub-model. Substitute the effective multidimensional environmental parameter data into the dynamic threshold analysis sub-model to obtain the target anomaly threshold; Substituting the effective multidimensional environmental parameter data into the anomaly pattern recognition sub-model yields the anomaly recognition result and the corresponding confidence level. The presence of parameter anomalies is determined based on the target anomaly threshold and confidence level, and the determination result is obtained.

[0013] Benefits: The dynamic threshold analysis sub-model adaptively generates target anomaly thresholds based on real-time valid data, ensuring that the judgment boundary adjusts dynamically with environmental fluctuations. This completely avoids the redundancy of minor fluctuations triggering invalid alarms under fixed thresholds and eliminates missed detections caused by rigid thresholds. Simultaneously, the anomaly pattern recognition sub-model mines the coupling relationships and evolutionary trends between temperature, humidity, particulate matter concentration, and microbial indicators from multi-dimensional data, accurately capturing latent and gradual anomalies that cannot be detected by fixed thresholds, and outputting confidence scores as a reliable basis for judgment. Finally, the judgment is made by combining dynamic thresholds and confidence scores, ensuring rapid response to obvious out-of-limit anomalies while using confidence scores to assess and filter false signals caused by residual noise. It identifies deep anomalies from a multi-parameter correlation perspective, building upon previously obtained high-precision valid data, making the entire anomaly identification process highly sensitive, highly resistant to interference, and with a low false alarm rate, thus improving the accuracy and efficiency of cleanroom environmental monitoring.

[0014] Optionally, substituting the effective multidimensional environmental parameter data into the dynamic threshold analysis sub-model to obtain the target anomaly threshold includes: For the aforementioned valid multidimensional environmental parameter data, calculate the mean and standard deviation of each type of environmental parameter data; Through formula Calculate the upper limit threshold; using the formula Calculate the lower threshold; in, Let be the mean of the environment parameters for class p. Let K be the standard deviation of the p-type environmental parameters, and K be the correction constant. The target anomaly threshold is determined based on the upper and lower thresholds corresponding to each environmental parameter type.

[0015] Optionally, determining the target anomaly threshold based on the upper and lower thresholds corresponding to each environmental parameter type includes: Through formula Obtain the absolute deviation; in, This represents the current measured value of the p-type environmental parameter. This represents the absolute deviation of the p-type environment parameter; The target anomaly threshold is obtained by mapping the absolute deviation of each environmental parameter type using the Sigmoid function and then performing weighted fusion.

[0016] Optionally, substituting the effective multidimensional environmental parameter data into the anomaly pattern recognition sub-model to obtain the anomaly recognition result and corresponding confidence level includes: Feature vectors are extracted from the effective multidimensional environmental parameter data; the feature vectors include the current values ​​of temperature, humidity, particulate matter concentration, and microbial concentration, as well as their respective first-order difference rates of change. After standardizing the feature vector, it is input into a pre-trained autoencoder network, and the reconstructed feature vector is obtained after processing by the encoder and decoder of the autoencoder. The reconstruction error value is obtained by calculating the mean square error between the feature vector and the reconstructed feature vector. The anomaly identification result and corresponding confidence level are obtained by searching a preset database based on the reconstruction error value; the preset database stores the reconstruction error value, alarm information and processing solution corresponding to different parameter anomalies.

[0017] Optionally, based on the target anomaly threshold and confidence level, it is determined whether there are parameter anomalies, and the determination result includes: If the target anomaly threshold is less than or equal to the preset safety threshold, it is determined that there is no parameter anomaly. If the target anomaly threshold is greater than the preset safety threshold, then it is determined whether the confidence level is greater than or equal to the target anomaly threshold. If the confidence level is greater than or equal to the target anomaly threshold, the judgment result is that an anomaly has occurred, and the corresponding parameter anomaly and alarm information are determined. If the confidence level is less than the target anomaly threshold, the result is that an anomaly has occurred, and an unknown fault alarm is generated.

[0018] In a second aspect of the invention, a cleanroom environment monitoring system is provided, comprising: The data acquisition module is used to synchronously collect multi-dimensional environmental parameter data of the target clean room temperature, humidity, particulate matter concentration and microbial indicators based on a preset cycle; A data denoising module is used to transmit the multidimensional environmental parameter data to a data processing terminal, so that the data processing terminal can perform denoising processing on the multidimensional environmental parameter data using a preset adaptive filtering algorithm to obtain effective multidimensional environmental parameter data. The anomaly analysis module is used to input the effective multidimensional environmental parameter data into the cleanroom environment anomaly identification model to obtain a judgment result on whether there is parameter anomaly, and to issue alarms and control based on the judgment result.

[0019] The beneficial effects of this plan are: By dynamically adjusting filtering parameters in real time using an adaptive filtering algorithm, noise and fluctuation interference under different operating conditions can be effectively removed, significantly improving the accuracy of multi-dimensional data such as temperature, humidity, particulate matter concentration, and microbial indicators. At the same time, by replacing fixed threshold comparison with a cleanroom environment anomaly identification model, the real-time fluctuation characteristics of parameters can be captured, accurately identifying hidden anomalies and gradual anomalies coupled with multiple parameters, avoiding invalid alarms caused by slight fluctuations, greatly reducing alarm redundancy and false judgments, thereby comprehensively improving the accuracy and monitoring efficiency of environmental anomaly identification, and providing a reliable basis for alarm and linkage control. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 A flowchart illustrating a cleanroom environment monitoring method provided in an embodiment of the present invention; Figure 2 Provided for embodiments of the present invention Figure 2 A method for judging abnormal environmental parameters in cleanrooms is provided. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0023] It should be noted that all formula calculations in the scheme are purely numerical calculations.

[0024] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention provides a method for monitoring the cleanroom environment. See also... Figure 1 The method includes the following steps: S101, based on a preset cycle, synchronously collects temperature, humidity, particulate matter concentration and microbial indicators in the target clean room to obtain multidimensional environmental parameter data; S102, transmit the multidimensional environmental parameter data to the data processing terminal; so that the data processing terminal can perform noise reduction processing on the multidimensional environmental parameter data through a preset adaptive filtering algorithm to obtain effective multidimensional environmental parameter data; S103: Substitute effective multidimensional environmental parameter data into the cleanroom environment anomaly identification model to obtain the judgment result of whether there is parameter anomaly, and issue alarms and controls based on the judgment result.

[0026] In one implementation, the preset period is determined by technical personnel; multiple multi-sensor collaborative acquisition nodes are distributed and deployed in the target clean room according to the spatial grid layout rules, and each node integrates temperature, humidity, particulate matter and microbial sensors; each node synchronizes its clock through a unified time synchronization module, triggers all sensors to collect data synchronously at the same sampling time within the preset period, and marks each set of data with a precise timestamp to form multi-dimensional environmental parameter data with time synchronization marks.

[0027] In one implementation, obtaining multidimensional environmental parameter data based on the synchronous collection of temperature, humidity, particulate matter concentration, and microbial indicators in the target clean room according to a preset period includes: distributing multiple multi-sensor collaborative acquisition nodes in the target clean room according to a preset spatial grid layout rule; each collaborative acquisition node integrates a temperature sensor, a humidity sensor, a particulate matter sensor, and a microbial sensor; each collaborative acquisition node synchronizes its clock through a unified time synchronization module, triggering each sensor to synchronously collect data at the sampling time within the same preset period, and marking each set of collected data with a timestamp to form multidimensional environmental parameter data with time synchronization markers.

[0028] In one implementation, the preset spatial grid layout rule is as follows: in the target cleanroom, multiple multi-sensor collaborative acquisition nodes are distributed and deployed according to the spatial grid layout rules required by standards such as ISO14644 or GMP. Each node integrates a temperature sensor (PT100), a humidity sensor (capacitive), a laser particulate counter (0.1-5μm), and a microbial sampling sensor (a real-time bioaerosol detection module based on fluorescence method). All nodes are synchronized at the nanosecond level through the IEEE1588 precise time protocol or a wireless synchronization module.

[0029] In one implementation, in newly built cleanrooms or cleanrooms with extremely strict requirements for electromagnetic interference, industrial Ethernet (Profinet) or fiber optic wired links are deployed to ensure low latency and high interference immunity for data transmission; in cleanrooms that are already built or have a lot of mobile equipment, Wi-Fi 6 or 5G private networks are used and redundant APs are configured. After each acquisition cycle, the data is pushed to the data processing terminal located outside the cleanroom or inside the computer room within 100ms.

[0030] In one implementation, multidimensional environmental parameter data is transmitted to a data processing terminal in real time via a wired or wireless communication link arranged in the cleanroom. The wired communication link is at least one of RS-485 bus, industrial Ethernet, or fiber optic cable, and the related cables are made of low-release, easy-to-wipe sheath material and connected to the sensor node through a dustproof sealed connector. The wireless communication link is based on at least one of ZigBee, Wi-Fi, or LoRa protocol. The wireless communication module and antenna are encapsulated in a sealed housing that meets the dust-free requirements of the cleanroom, and the antenna adopts a built-in design with no protruding antenna components on the outside of the housing.

[0031] In one embodiment, obtaining effective multidimensional environmental parameter data by denoising the multidimensional environmental parameter data using a preset adaptive filtering algorithm includes: Determine the current operating level of the target cleanroom, match an adaptive filtering algorithm based on the operating level, and determine the corresponding adaptive filtering parameters; Based on the adaptive filtering algorithm and adaptive filtering parameters, noise reduction processing is performed on temperature, humidity, particulate matter concentration and microbial indicators in the multidimensional environmental parameter data to obtain effective multidimensional environmental parameter data.

[0032] In one implementation, the operating level includes Level 1 (high cleanliness): suitable for scenarios such as semiconductor lithography and core filling areas in biopharmaceuticals, with cleanliness requirements reaching ISO 3~5 (≥0.5μm particulate matter concentration not exceeding 3520 particles / m³). 3 In such environments, the noise amplitude is extremely low, with occasional pulse interference caused by minor equipment vibrations or minimal personnel entry. Noise reduction should preserve these subtle fluctuations and avoid excessive smoothing. Level 2 (Medium Cleanliness): Suitable for scenarios such as medical device assembly and general pharmaceutical preparation areas, with cleanliness requirements of ISO 6-7 (≥0.5μm particulate matter concentration of 35200-352000 particles / m³). 3 (Between); the environment contains moderate-amplitude noise caused by periodic airflow switching and personnel movement, and noise reduction needs to balance noise suppression and signal fidelity; Level 3 (low cleanliness): suitable for general electronic assembly, packaging workshops, etc., with cleanliness requirements of ISO 8~9 (≥0.5μm particulate matter concentration can reach 3,520,000 particles / m³). 3(Above); noise intensity is high and frequency is high, which may even drown out the real signal. The main goal of noise reduction is to extract macroscopic change trends.

[0033] In one implementation, for Level 1 (high cleanliness level), where temperature and humidity signals change slowly, an adaptive Kalman filter is used. The process noise covariance Q is set to 0.001 (temperature) and 0.002 (humidity), and the initial measurement noise covariance R is set to 0.1 and 0.15, respectively. Since particulate matter concentration is prone to single-point spikes, an adaptive median filter is used, with an initial window size of 3 and a maximum window size of 5. The spike detection threshold is a deviation from the median within the window exceeding 3 times the local standard deviation. For microbial indicator data, which change slowly and outliers are mostly sampling noise, a Hampel filter is used, with a sliding window length of 5 and a removal threshold of 3 times the absolute deviation of the median.

[0034] In one implementation, for Level 2 (medium cleanliness), temperature and humidity are weighted moving averages combined with adaptive thresholds, with window sizes of 10 and 12 respectively. The weight coefficients decay linearly over time (the most recent data has the highest weight). When the rate of change of three consecutive sampling points exceeds 0.1℃ / min (temperature) or 0.2%RH / min (humidity), the window is temporarily reduced to 5 to quickly respond to real changes. Particulate matter concentration still uses adaptive median filtering, but the window is expanded to an initial 5 and a maximum of 9, and the peak judgment threshold is relaxed to deviate from the median by more than 2.5σ. Microbial indicators use second-order low-pass Butterworth filtering with a cutoff frequency of 0.1Hz (corresponding to a period of 10 seconds) and zero-phase filtering is used to avoid phase delay.

[0035] In one implementation, for Level 3 (low cleanliness level), temperature and humidity are directly calculated using exponentially weighted moving average (EWMA), with smoothing factors α set to 0.1 and 0.15 respectively, resulting in an attenuation rate of over 90% for instantaneous spikes. Particulate matter concentration is first subjected to large-window adaptive median filtering (initial window 11, maximum window 21, spike determination threshold is deviation from the median exceeding 2σ), and then the median filtering result is smoothed a second time using EWMA (α=0.2). Microbial indicators are first calculated using a moving average with a window length of 7, and then the rate of change of adjacent sampling points is subjected to amplitude limiting filtering to limit the rate of change of adjacent points to no more than 50%.

[0036] In one implementation, after matching the algorithm and parameters, the data processing terminal filters the time series data of each sensor channel. For continuously changing parameters such as temperature and humidity, the matching filtering algorithm is substituted point by point in time order, and the output value is calculated in real time. For particulate matter concentration and microbial indicators, a sliding window method is used for processing, and the data points in the window are updated in real time. After processing the data of one sampling period, a set of corresponding effective multidimensional environmental parameter data is output.

[0037] In one embodiment, see Figure 2 , Figure 2 A method for judging abnormal cleanroom environmental parameters is provided, specifically including: S1031, The cleanroom environment anomaly identification model includes a dynamic threshold analysis sub-model and an anomaly pattern identification sub-model; S1032, Substitute effective multidimensional environmental parameter data into the dynamic threshold analysis sub-model to obtain the target anomaly threshold; S1033, Substitute effective multidimensional environmental parameter data into the anomaly pattern recognition sub-model to obtain anomaly recognition results and corresponding confidence levels; S1034, determine whether there is parameter anomaly based on the target anomaly threshold and confidence level, and obtain the judgment result.

[0038] In one implementation, the target anomaly threshold is obtained by substituting effective multidimensional environmental parameter data into a dynamic threshold analysis sub-model, including: For valid multidimensional environmental parameter data, calculate the mean and standard deviation of each type of environmental parameter data; Through formula Calculate the upper limit threshold; using the formula Calculate the lower threshold; in, Let be the mean of the environment parameters for class p. Let K be the standard deviation of the p-type environmental parameters, and K be the correction constant. The target anomaly threshold is determined based on the upper and lower thresholds corresponding to each environmental parameter type.

[0039] In one implementation, determining the target anomaly threshold based on the upper and lower thresholds corresponding to each environmental parameter type includes: Through formula Obtain the absolute deviation; in, This represents the current measured value of the p-type environmental parameter. This represents the absolute deviation of the p-type environment parameter; The target anomaly threshold is obtained by mapping the absolute deviation of each environmental parameter type using the Sigmoid function and then performing weighted fusion.

[0040] In one implementation, for the p-th type of environmental parameter, the system maintains a sliding time window, which is specifically determined by the technicians; the mean and standard deviation of the parameter data within the window are calculated, with a correction constant K; Level 1 (high cleanliness): K=3; Level 2 (medium cleanliness): K=4; Level 3 (low cleanliness): K=5.

[0041] In one implementation, traditional fixed thresholds, once set, cannot adapt to changes in environmental baseline drift and fluctuations. When the environment is stable, the fixed threshold may be too lenient, missing subtle but persistent trend anomalies. When environmental fluctuations increase due to production activities, the fixed threshold may be too tight, frequently triggering invalid alarms. By using a sliding window to calculate the mean and standard deviation in real time and dynamically adjusting the upper and lower limits, the threshold always fits the current statistical characteristics. Fixed thresholds may misjudge low-temperature anomalies, while dynamic thresholds automatically follow the decreasing mean, only detecting anomalies when the deviation exceeds K*σ. p The alarm was triggered only in time, thus avoiding a large number of false alarms.

[0042] In one implementation, the original absolute deviation will produce a non-zero value even within the normal fluctuation range. If the weighted sum is directly applied, it may lead to frequent weak alarms. By using the Sigmoid transformation, false positives caused by noise and transient interference are reduced, while retaining the effective response to real anomalies.

[0043] In one embodiment, substituting effective multidimensional environmental parameter data into the anomaly pattern recognition sub-model to obtain anomaly recognition results and corresponding confidence levels includes: Feature vectors are extracted from effective multidimensional environmental parameter data; the feature vectors contain the current values ​​of temperature, humidity, particulate matter concentration, and microbial concentration, as well as their respective first-order difference rates of change. After standardizing the feature vector, it is input into a pre-trained autoencoder network, and the reconstructed feature vector is obtained after processing by the encoder and decoder of the autoencoder. The reconstruction error value is obtained by calculating the mean square error between the eigenvector and the reconstructed eigenvector. The anomaly identification results and corresponding confidence levels are obtained by searching the preset database based on the reconstruction error value; the preset database is also searched to save the corresponding reconstruction error values, alarm information and handling solutions under different parameter anomalies.

[0044] In one implementation, because different features have different dimensions and orders of magnitude, directly inputting them into the autoencoder can lead to feature-dominated errors with large numerical values. Therefore, Z-score normalization is performed on each feature dimension. An autoencoder is an unsupervised neural network consisting of an encoder and a decoder. This model employs an undercomplete autoencoder (the bottleneck layer dimension is smaller than the input layer), forcing the network to learn the main latent features of the data. The autoencoder includes an input layer, encoder layer 1, encoder layer 2, decoder layer 1, and an output layer. The input layer has no activation function, the encoder layer 1, encoder layer 2, and decoder layer 1 all use ReLU activation functions, and the output layer uses Linear activation function.

[0045] In one implementation, a large amount of historical data collected under normal operating conditions of the cleanroom is used to form a training set. The mean square error between the input vector and the output reconstructed vector is minimized. The optimization algorithm adopts Adam with a learning rate of 0.001 and training continues until the loss converges.

[0046] In one implementation, the system pre-builds a preset database, which stores the reconstruction error range, alarm information and handling solutions corresponding to various anomaly types identified through offline experiments or historical fault cases; the calculated reconstruction error value is matched with the range in the database to locate the corresponding anomaly type.

[0047] In one implementation, a fixed threshold is used to independently judge each parameter, which fails to detect coupling anomalies such as simultaneous slow increases in temperature and humidity without exceeding individual limits. However, the autoencoder learns the cooperative relationship between parameters under normal conditions during training. When a pattern that violates this cooperative relationship occurs, the autoencoder cannot reconstruct well, resulting in a large reconstruction error. Traditional thresholding methods often fail for slow-drift anomalies because each individual point does not exceed the fixed threshold. The feature vector of this model contains the first-order differential rate of change, making the autoencoder sensitive to trend changes. Slow increases will continuously show small positive values ​​on the ΔPM feature, deviating from the normal stable state, causing the reconstruction error to gradually accumulate and exceed the threshold. Therefore, the model can identify abnormal trends before the absolute value of the parameters exceeds the limit, achieving early warning.

[0048] In one implementation, the known anomaly types in the database are limited, while new types of faults may occur in actual production. The fixed threshold method can only output normal or abnormal, and cannot distinguish whether they are known. This model pre-sets unknown composite anomaly branches in the database to trigger unknown fault alarms, prompting engineers to intervene manually, and adds new data to the training set to achieve incremental model updates, continuously improving the recognition capability.

[0049] In one embodiment, the determination of whether parameter anomalies exist is based on a target anomaly threshold and a confidence level, and the determination result includes: If the target anomaly threshold is less than or equal to the preset safety threshold, it is determined that there is no parameter anomaly. If the target anomaly threshold is greater than the preset safety threshold, then determine whether the confidence level is greater than or equal to the target anomaly threshold. If the confidence level is greater than or equal to the target anomaly threshold, the judgment result is that an anomaly has occurred, and the corresponding parameter anomaly and alarm information are determined. If the confidence level is less than the target anomaly threshold, the result is that an anomaly has occurred, and an unknown fault alarm is generated.

[0050] In one implementation, the system is configured with a preset safety threshold value of (0,1) during initialization. The preset safety threshold is specifically determined by technical personnel and is determined according to the cleanroom operation level. The preset safety threshold for high cleanliness level (Level 1) is less than the preset safety threshold for medium cleanliness level (Level 2), and the preset safety threshold for medium cleanliness level (Level 2) is less than the preset safety threshold for low cleanliness level (Level 3).

[0051] In one implementation, for the current sampling time, the system executes the following two sub-models in parallel or sequentially: the target anomaly threshold at the current time is obtained from the dynamic threshold analysis sub-model, and the anomaly identification result and corresponding confidence level at the current time are obtained from the anomaly pattern recognition sub-model. When the target anomaly threshold is very small, it means that all environmental parameters are closely aligned with their respective dynamic mean values ​​without significant deviation. At this time, even if the confidence level of the anomaly pattern recognition sub-model is very high (which may be a misjudgment), the system will prioritize trusting the dynamic threshold model and will not trigger any alarms, thus avoiding false alarms caused by the accidental high reconstruction error of the autoencoder when the environment is absolutely normal.

[0052] In one implementation, when the target anomaly threshold exceeds a preset safety threshold, it indicates at least a certain degree of anomaly risk. In this case, the results of the anomaly pattern recognition sub-model need to be combined to distinguish between known, explainable anomalies and unknown, novel anomalies. The output of the dynamic threshold sub-model is essentially the degree to which multiple parameters deviate from the normal dynamic range, which can be regarded as the probability or severity of anomaly existence. The confidence level output by the anomaly pattern recognition sub-model is the degree of matching between the current anomaly pattern and known anomaly types in the database. When the confidence level is greater than or equal to the target anomaly threshold, it indicates that not only does an anomaly exist, but the pattern of this anomaly also has a high-confidence correspondence in the known database, thus providing a specific anomaly type and corresponding countermeasures. When the confidence level is less than the target anomaly threshold, it indicates that although there is a significant parameter deviation, the pattern corresponding to the autoencoder reconstruction error does not match any known anomaly type, suggesting that it may be a novel fault or a complex multi-parameter coupled anomaly, requiring manual intervention.

[0053] In one implementation, the dynamic threshold sub-model may output slightly higher than the safety threshold for small, short-term normal fluctuations. However, relying solely on this threshold would result in false alarms. By introducing a second-level judgment that requires the confidence level to also reach the threshold, false positives that are statistically off but whose patterns do not match any known anomalies can be filtered out. This is because the reconstruction error of the autoencoder for normal fluctuations is usually very small, corresponding to a low confidence level. Conversely, the abnormal pattern recognition sub-model may occasionally produce a high reconstruction error due to sensor noise (i.e., a high-confidence false alarm). However, in this case, the output of the dynamic threshold sub-model is often very low, and the first-level judgment will intercept it without triggering an alarm.

[0054] In one implementation, a two-level judgment logic combining dynamic threshold priority screening and autoencoder confidence secondary confirmation combines the advantages of statistical anomaly detection and pattern recognition anomaly detection: the former ensures sensitivity and interpretability to parameter deviations, while the latter provides the ability to match complex anomaly patterns. The joint decision-making of the two avoids misjudgment by a single model and can distinguish between known anomalies and unknown faults, greatly improving the accuracy and practical value of cleanroom environment anomaly identification.

[0055] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A method for monitoring the environment of a cleanroom, characterized in that, The method includes: Multidimensional environmental parameter data are obtained by synchronously collecting temperature, humidity, particulate matter concentration and microbial indicators in the target clean room based on a preset cycle; The multidimensional environmental parameter data is transmitted to a data processing terminal, so that the data processing terminal can perform noise reduction processing on the multidimensional environmental parameter data using a preset adaptive filtering algorithm to obtain effective multidimensional environmental parameter data. The effective multidimensional environmental parameter data are substituted into the cleanroom environment anomaly identification model to obtain the judgment result of whether there is parameter anomaly, and alarms and controls are performed based on the judgment result.

2. The cleanroom environment monitoring method according to claim 1, characterized in that, Multidimensional environmental parameter data are obtained by synchronously collecting temperature, humidity, particulate matter concentration, and microbial indicators in the target cleanroom at preset intervals, including: In the target cleanroom, multiple multi-sensor collaborative acquisition nodes are deployed in a distributed manner according to the preset spatial grid layout rules. Each collaborative acquisition node integrates a temperature sensor, a humidity sensor, a particulate matter sensor, and a microbial sensor. Each collaborative acquisition node synchronizes its clock through a unified time synchronization module. During the sampling time within the same preset period, each sensor is triggered to collect data synchronously, and each set of collected data is marked with a timestamp, forming multidimensional environmental parameter data with time synchronization marks.

3. The cleanroom environment monitoring method according to claim 1, characterized in that, The multidimensional environmental parameter data is transmitted to the data processing terminal in real time via wired or wireless communication links deployed in the cleanroom.

4. The cleanroom environment monitoring method according to claim 1, characterized in that, The effective multidimensional environmental parameter data obtained by denoising the multidimensional environmental parameter data through a preset adaptive filtering algorithm includes: Determine the current operating level of the target cleanroom, match an adaptive filtering algorithm according to the operating level, and determine the corresponding adaptive filtering parameters; Based on the adaptive filtering algorithm and adaptive filtering parameters, noise reduction processing is performed on temperature, humidity, particulate matter concentration, and microbial indicators in the multidimensional environmental parameter data to obtain effective multidimensional environmental parameter data.

5. A cleanroom environment monitoring method according to claim 1, characterized in that, Substituting the effective multidimensional environmental parameter data into the cleanroom environment anomaly identification model yields the following judgment results regarding the presence or absence of parameter anomalies: The cleanroom environment anomaly identification model includes a dynamic threshold analysis sub-model and an anomaly pattern identification sub-model. Substitute the effective multidimensional environmental parameter data into the dynamic threshold analysis sub-model to obtain the target anomaly threshold; Substituting the effective multidimensional environmental parameter data into the anomaly pattern recognition sub-model yields the anomaly recognition result and the corresponding confidence level. The presence of parameter anomalies is determined based on the target anomaly threshold and confidence level, and the determination result is obtained.

6. A cleanroom environment monitoring method according to claim 5, characterized in that, Substituting the effective multidimensional environmental parameter data into the dynamic threshold analysis sub-model yields the target anomaly threshold, including: For the aforementioned valid multidimensional environmental parameter data, calculate the mean and standard deviation of each type of environmental parameter data; Through formula Calculate the upper limit threshold; using the formula Calculate the lower threshold; in, Let be the mean of the environment parameters for class p. Let K be the standard deviation of the p-type environmental parameters, and K be the correction constant. The target anomaly threshold is determined based on the upper and lower thresholds corresponding to each environmental parameter type.

7. A cleanroom environment monitoring method according to claim 6, characterized in that, The target anomaly threshold is determined based on the upper and lower thresholds corresponding to each environmental parameter type, including: Through formula Obtain the absolute deviation; in, This represents the current measured value of the p-type environmental parameter. This represents the absolute deviation of the p-type environment parameter; The target anomaly threshold is obtained by mapping the absolute deviation of each environmental parameter type using the Sigmoid function and then performing weighted fusion.

8. A cleanroom environment monitoring method according to claim 5, characterized in that, Substituting the effective multidimensional environmental parameter data into the anomaly pattern recognition sub-model yields the anomaly recognition results and corresponding confidence levels, including: Feature vectors are extracted from the effective multidimensional environmental parameter data; the feature vectors include the current values ​​of temperature, humidity, particulate matter concentration, and microbial concentration, as well as their respective first-order difference rates of change. After standardizing the feature vector, it is input into a pre-trained autoencoder network, and the reconstructed feature vector is obtained after processing by the encoder and decoder of the autoencoder. The reconstruction error value is obtained by calculating the mean square error between the feature vector and the reconstructed feature vector. The anomaly identification result and corresponding confidence level are obtained by searching a preset database based on the reconstruction error value; the preset database stores the reconstruction error value, alarm information and processing solution corresponding to different parameter anomalies.

9. A cleanroom environment monitoring method according to claim 8, characterized in that, Based on the target anomaly threshold and confidence level, it is determined whether there are parameter anomalies, and the determination results include: If the target anomaly threshold is less than or equal to the preset safety threshold, it is determined that there is no parameter anomaly. If the target anomaly threshold is greater than the preset safety threshold, then it is determined whether the confidence level is greater than or equal to the target anomaly threshold. If the confidence level is greater than or equal to the target anomaly threshold, the judgment result is that an anomaly has occurred, and the corresponding parameter anomaly and alarm information are determined. If the confidence level is less than the target anomaly threshold, the result is that an anomaly has occurred, and an unknown fault alarm is generated.

10. A cleanroom environment monitoring system, characterized in that, The system includes: The data acquisition module is used to synchronously collect multi-dimensional environmental parameter data of the target clean room temperature, humidity, particulate matter concentration and microbial indicators based on a preset cycle; A data denoising module is used to transmit the multidimensional environmental parameter data to a data processing terminal, so that the data processing terminal can perform denoising processing on the multidimensional environmental parameter data using a preset adaptive filtering algorithm to obtain effective multidimensional environmental parameter data. The anomaly analysis module is used to input the effective multidimensional environmental parameter data into the cleanroom environment anomaly identification model to obtain a judgment result on whether there is parameter anomaly, and to issue alarms and control based on the judgment result.

Images