Cleanness real-time detection method and system based on high-precision clean space

Abstract

The invention relates to the technical field of air purification, in particular to a cleanliness real-time detection method and system based on a high-precision clean space, and the method comprises the following steps: 1, building a data set: arranging a plurality of sensor nodes in the clean space, parameter data of the particulate matter concentration p, the wind speed v, the temperature T, the humidity H and the pressure difference delta P in the clean space at the time node k are obtained; step 2, transmission of sensor data: transmitting the sensor data Sk = [pk, vk, Tk, Hk, delta Pk] T to a data processing server through a wireless communication network; establishing a wired network as a backup channel of a wireless communication network; and step 3: data preprocessing: removing noise data, error data and repeated data in the collected data. The method effectively solves the problems that the detection period is long, the timeliness is poor and the data accuracy is poor when traditional manual regular sampling is combined with laboratory analysis; and a single-function monitoring device has a limited monitoring range and is difficult to comprehensively reflect the clean condition of the whole clean space.

Description

Technical Field

[0001] The present invention relates to the technical field of air purification, and in particular to a real-time cleanliness detection method and system based on high-precision clean space. Background Art

[0002] In modern industrial production and scientific research and development fields, such as electronic chip manufacturing, biopharmaceutical R&D and production, and precision instrument manufacturing, there are extremely high requirements for the cleanliness of the production environment. Clean space is an important infrastructure in these fields, and its cleanliness is directly related to product quality, production efficiency, and the accuracy of scientific research experiments.

[0003] In some high-precision cleanrooms with extremely stringent requirements, minute fluctuations in particle concentration or deviations in temperature and humidity can lead to reduced product yields, distorted experimental data, or even production accidents. Therefore, real-time, accurate, and reliable monitoring of environmental parameters and warning of anomalies are necessary. Cleanliness testing technology, as an important means of ensuring environmental quality, provides fundamental support for cleanroom management.

[0004] However, existing high-precision clean space cleanliness testing technologies primarily rely on manual periodic sampling combined with laboratory analysis, or the deployment of single-function monitoring equipment. The former has a long testing cycle and poor timeliness, cannot promptly reflect the real-time status of the clean space, is easily interfered with by human factors, and has poor data accuracy. The latter has a limited monitoring range and can only provide local data, making it difficult to fully reflect the cleanliness status of the entire clean space. It also lacks the ability to effectively and comprehensively analyze changes in cleanliness under the influence of multiple factors.

[0005] Therefore, to address the above problems, a real-time cleanliness detection method and system based on high-precision clean space are proposed. Summary of the Invention

[0006] The object of the present invention is to provide a method and system for real-time detection of cleanliness of a clean space based on high precision, so as to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A real-time cleanliness detection method and system based on high-precision clean space includes the following steps:

[0009] Step 1: Dataset construction: Multiple sensor nodes are deployed in the clean space to obtain parameter data of particle concentration p, wind speed v, temperature T, humidity H, and pressure difference △P in the clean space at time node k;

[0010] Step 2: Transmission of sensor data: The sensor data S is transmitted via wireless communication network. k =[p k ,vk ,T k ,H k ,△P k ] T Transmit to the data processing server; establish a wired network as a backup channel for the wireless communication network;

[0011] Step 3: Data preprocessing: remove noise data, erroneous data, and duplicate data from the collected data;

[0012] Step 4: Data fusion processing: Fusion processing is performed on the data collected by different types of sensors, and multi-source data is integrated through the Kalman filter algorithm;

[0013] Step 5: Data analysis: Use deep learning models to analyze the pre-processed cleanliness-related data;

[0014] Step 6: Result display: The user interface displays the clean space parameter data, historical data curves, trend prediction results and fault diagnosis reports to the user in real time, and continuously monitors the cleanliness index. If the cleanliness index is monitored to exceed the preset safety threshold or a valid fault is confirmed, an intelligent alarm will be automatically triggered and an alarm message will be immediately sent to the preset relevant responsible personnel through various methods such as SMS, email and application push notification.

[0015] As a further optimized content of the present invention, the following steps are also included:

[0016] Step 7: Based on the trend prediction results and fault diagnosis information, generate and display decision support suggestions, including cleaning and maintenance plans or troubleshooting solutions.

[0017] As a further optimization of the present invention, in step 3, for a single parameter, such as the particle concentration p k , perform Z-score anomaly detection to remove erroneous data from the collected data, specifically:

[0018]

[0019] If |z k |>3, it is considered an outlier and removed;

[0020] Where μ p is the mean value of the particle concentration in a window of size N, calculated as σ p is the standard deviation of the particle concentration in a window of size N, calculated as

[0021] As a further optimization of the present invention, in step 3, a time series smoothing method is used to reduce the influence of random noise on the data, and the wind speed v k For example, specifically:

[0022]

[0023] Where, is the wind speed after preprocessing, and M is the size of the sliding window.

[0024] As a further optimization of the present invention, in step 3, two n-dimensional data points are measured by the Euclidean distance formula. and The similarities between them are removed, and duplicate data is removed, specifically:

[0025]

[0026] If d(x i , x j )<ε, then x i and x j If it is duplicate data, one of the data points will be retained and the other will be removed;

[0027] Where, d(x i , x j ) is x i and x j The straight-line distance in n-dimensional space, ε is the set distance threshold.

[0028] As a further optimization of the present invention, in step 4, the data collected by different types of sensors are fused and processed, and the multi-source data is integrated by using a Kalman filter algorithm, etc., including the following steps:

[0029] S41: Integrate multi-source sensor data and estimate the optimal state: Assume that the state vector x k Consistent with the sensor data, the dynamic model is a constant speed model:

[0030]

[0031] Where x k is the state vector at the kth moment, including all sensor parameters;

[0032] S42: Prediction step: Based on the optimal estimate and dynamic model of the previous moment, the estimated state value of the current moment is calculated in advance:

[0033]

[0034] P k|k-1 =FP k-1|k-1 FT +Q

[0035] Where, is the prior state estimate at the kth moment, is the posterior state estimate at the k-1th moment, F is the state transfer matrix, P k|k-1 is the prior estimated covariance matrix, P k-1|k-1 is the estimated covariance matrix of the previous moment, F T is the transposed matrix of the state transfer matrix F after the rows and columns are swapped, Q is the process noise covariance matrix;

[0036] S43: Update step: Correct the predicted value by the current measurement value to obtain the optimal state estimate:

[0037]

[0038] If R approaches 0, the measurement value Z is trusted first. k , if P k|k-1 When it approaches 0, the predicted value is trusted first.

[0039] Where Z k is the measurement vector at the kth moment, H is the observation matrix, R is the measurement noise covariance matrix, P k|k-1 H T (HP k|k-1 H T +R) -1 is the implicit calculation of the Kalman gain.

[0040] As a further optimization of the present invention, in step 5, the deep learning model MLP is used to analyze the preprocessed and fused data and output the cleanliness index or prediction result:

[0041]

[0042] Where, f k is the feature vector at the kth moment output after processing in step 4, is the estimated value of the fused parameter;

[0043] Multilayer Perceptron (MLP) model formula:

[0044] h k =φ(W1f k +b1)

[0045] c k =W2h k +b2

[0046] Where h kis the hidden layer output vector, φ(·) is the activation function, W1, W2 are weight matrices, b1, b2 are bias vectors, c k It is the cleanliness index of the output.

[0047] As a further optimization of the present invention, in step 6, the cleanliness sequence at the next m time points is predicted by the trend prediction formula:

[0048] C future =(c k+1 ,c k+2 ,...,c k+m )

[0049] Intelligent alarm triggering conditions:

[0050]

[0051] Where c min is the lower limit of the safety threshold, c max is the upper limit of the safety threshold, d k is the failure probability, d th is the fault threshold.

[0052] As further optimized content of the present invention, including:

[0053] Dataset construction module: used to arrange multiple sensor nodes in the clean space, obtain parameter data of particle concentration, wind speed, temperature, humidity and pressure difference in the clean space at time nodes, and construct a data set;

[0054] Data transmission module: used to transmit sensor data to the data processing server;

[0055] Data preprocessing module: used to remove noise data, erroneous data and duplicate data from the collected data;

[0056] Data fusion processing module: used to fuse data collected by different types of sensors and integrate multi-source data through Kalman filtering algorithm;

[0057] Data analysis module: used to analyze the pre-processed cleanliness-related data using a deep learning model;

[0058] Result display module: used to display clean space parameter data, historical data curves, trend prediction results and fault diagnosis reports in real time, and trigger an alarm when the cleanliness index is abnormal to notify relevant personnel;

[0059] Decision support module: used to generate and display decision support recommendations.

[0060] As a further optimized content of the present invention, a computer program is stored thereon, and when the computer program is executed by a processor, the steps in the method as described in any one of claims 1 to 8 are executed.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. In the present invention, multiple sensor nodes are arranged in the clean space to obtain parameter data such as particulate matter concentration, wind speed, temperature, humidity and pressure difference in real time, and the data is transmitted to the data processing server through a wireless communication network and a wired backup channel. This effectively solves the problems of traditional manual periodic sampling combined with laboratory analysis, which has a long detection cycle, poor timeliness and poor data accuracy, as well as the disadvantages of single-function monitoring equipment with a limited monitoring range and difficulty in fully reflecting the cleanliness status of the entire clean space. The system removes noise, errors and duplicate data in the data preprocessing stage, significantly improving data accuracy and reliability. The Kalman filter algorithm is used to fuse multi-source data to overcome the limitations and errors of single sensors. Achieve a comprehensive and accurate estimation of the environmental status of the entire clean space; use deep learning models to analyze the fused data, accurately evaluate cleanliness and predict future trends, generate decision support suggestions such as cleaning maintenance plans or troubleshooting solutions based on trend prediction results and fault diagnosis information, optimize operation and maintenance decision-making efficiency, ensure stable operation of clean spaces, improve production efficiency and product quality, and reduce operating costs; at the same time, once the system detects an abnormality in the cleanliness index, it immediately triggers an intelligent alarm and notifies relevant personnel through various means to achieve real-time, accurate and reliable monitoring of the cleanliness of the clean space and abnormality warning, avoiding risks such as product quality degradation, experimental data distortion or production accidents due to delayed monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a flow chart of the real-time cleanliness detection method based on high-precision clean space of the present invention;

[0064] Figure 2 This is a system block diagram of a real-time cleanliness detection system based on a high-precision clean space according to the present invention;

[0065] Figure 3 This is a flow chart for determining the triggering conditions of the intelligent alarm of the present invention. DETAILED DESCRIPTION

[0066] See also Figure 1-Figure 3 , the present invention provides a technical solution:

[0067] A real-time cleanliness detection method and system based on high-precision clean space includes the following steps: Step 1: Dataset construction: multiple sensor nodes are arranged in the clean space to obtain parameter data of particle concentration p, wind speed v, temperature T, humidity H and pressure difference ΔP in the clean space at time node k;

[0068] Step 2: Transmission of sensor data: The sensor data S is transmitted via wireless communication network. k =[p k , v k , T k , H k , ΔP k ] T Transmit to the data processing server; establish a wired network as a backup channel for the wireless communication network, effectively preventing data transmission interruption caused by wireless communication network failure, significantly improving the reliability of data transmission and ensuring monitoring continuity;

[0069] Step 3: Data preprocessing: Remove noise, errors, and duplicates from the collected data to improve its accuracy and reliability. Through a systematic data preprocessing process, outliers, noise, and redundant information are effectively eliminated, significantly improving the quality and credibility of the data relied upon for subsequent analysis.

[0070] Step 4: Data fusion processing: Data collected by different types of sensors are fused and processed. Multi-source data is integrated through Kalman filtering algorithms, overcoming the limitations and errors of single sensors and achieving a more comprehensive and accurate real-time estimation of environmental conditions.

[0071] Step 5: Data analysis: Using a deep learning model to analyze the pre-processed cleanliness-related data can more effectively capture the nonlinear relationship between complex environmental parameters and cleanliness indicators, achieving more accurate cleanliness assessment or prediction.

[0072] Step 6: Result display: The user interface displays the clean space parameter data, historical data curves, trend prediction results and fault diagnosis reports to the user in real time, and continuously monitors the cleanliness index. If the cleanliness index is detected to exceed the preset safety threshold or a valid fault is confirmed, an intelligent alarm will be automatically triggered, and the alarm information will be immediately sent to the preset relevant responsible personnel through various means such as SMS, email and application push notifications. This greatly facilitates users to monitor and manage the status of the clean space, ensures that they can be notified and responded to in the first time when the cleanliness is abnormal or the equipment fails, and effectively guarantees the safe operation of the clean space.

[0073] As a technical solution for further implementation of this solution, the following steps are also included:

[0074] Step 7: Based on the trend prediction results and fault diagnosis information, decision support suggestions are generated and displayed. The suggestions include cleaning and maintenance plans or troubleshooting solutions. On the basis of providing monitoring, prediction and diagnosis information, specific decision suggestions are further generated to directly assist managers in making operation and maintenance decisions, thereby improving the practicality and intelligence level of the system and optimizing operation and maintenance efficiency.

[0075] As a further technical solution for implementing this solution, in step 3, for a single parameter, such as the particle concentration p k , perform Z-score anomaly detection to remove erroneous data from the collected data, specifically:

[0076]

[0077] If |z k If |>3, it is considered an outlier and removed;

[0078] Where μ p is the mean value of the particle concentration in a window of size N, calculated as σ p is the standard deviation of the particle concentration in a window of size N, calculated as The use of the Z-score method based on statistical principles for outlier detection can objectively and efficiently identify and eliminate obvious erroneous data caused by transient sensor failures or interference, significantly improving the reliability of the parameter data.

[0079] As a technical solution for further implementation of this scheme, in step 3, the time series smoothing method is used to reduce the influence of random noise on the data. k For example, specifically:

[0080]

[0081] Where, is the preprocessed wind speed, M is the size of the sliding window, and the high-frequency random noise in the sensor measurement is effectively filtered out by time series smoothing methods such as sliding average, making the data curves of parameters such as wind speed smoother and more stable, and more realistically reflecting the slow-changing trend of environmental parameters.

[0082] As a technical solution for further implementation of this solution, in step 3, the Euclidean distance formula is used to measure the two n-dimensional data points. and The similarities between them are removed, and duplicate data is removed, specifically:

[0083]

[0084] If d(x i ,x j)<ε, then x i and x j If it is duplicate data, one of the data points will be retained and the other will be removed;

[0085] Where, d(x i , x j ) is x i and x j The straight-line distance in n-dimensional space, ε is the set distance threshold, and the Euclidean distance is used to accurately judge the similarity between multidimensional data points, effectively identifying and eliminating duplicate data caused by redundant transmission or collection, reducing data redundancy and improving the efficiency of data storage and analysis.

[0086] As a technical solution for further implementing this solution, in step 4, the data collected by different types of sensors are fused and processed, and the integration of multi-source data through the Kalman filter algorithm includes the following steps:

[0087] S41: Integrate multi-source sensor data and estimate the optimal state: Assume that the state vector x k Consistent with the sensor data, the dynamic model is a constant speed model:

[0088]

[0089] Where x k is the state vector at the kth moment, including all sensor parameters;

[0090] S42: Prediction step: Based on the optimal estimate and dynamic model of the previous moment, the estimated state value of the current moment is calculated in advance:

[0091]

[0092] P k|k-1 =FP k-1|k-1 F T +Q

[0093] Where, is the prior state estimate at the kth moment, is the posterior state estimate at the k-1th moment, F is the state transfer matrix, P k|k-1 is the prior estimated covariance matrix, P k-1|k-1 is the estimated covariance matrix of the previous moment, F T is the transposed matrix of the state transfer matrix F after the rows and columns are swapped, Q is the process noise covariance matrix;

[0094] S43: Update step: Correct the predicted value by the current measurement value to obtain the optimal state estimate:

[0095]

[0096] If R approaches 0, the measurement value Z is trusted first. k , if P k|k-1 When it approaches 0, the predicted value is trusted first.

[0097] Where Z k is the measurement vector at the kth moment, H is the observation matrix, R is the measurement noise covariance matrix, P k|k-1 H T (HP k|k-1 G T +R) -1 is the implicit calculation of the Kalman gain;

[0098] The Kalman filter algorithm is applied in detail, combined with the system dynamic model and real-time measurement values, to adaptively estimate the optimal fusion state; the trust weights of the predicted values ​​and measured values ​​are dynamically adjusted through the covariance matrix, achieving the optimal fusion of multi-source sensor data in time and space, significantly improving the accuracy and stability of environmental state estimation, and is particularly suitable for dynamically changing clean environments.

[0099] As a technical solution for further implementation of this solution, in step 5, the deep learning model MLP is used to analyze the preprocessed and fused data and output the cleanliness index or prediction result:

[0100]

[0101] Where, f k is the feature vector at the kth moment output after processing in step 4, is the estimated value of the fused parameter;

[0102] Multilayer Perceptron (MLP) model formula:

[0103] h k =φ(W1f k +b1)

[0104] c k =W2h k +b2

[0105] Where h k is the hidden layer output vector, φ(·) is the activation function, W1, W2 are weight matrices, b1, b2 are bias vectors, c k To output the cleanliness index, a multi-layer perceptron (MLP) deep learning model is used to process the fused high-quality feature vectors. This model can effectively learn and model the complex nonlinear mapping relationship between multiple environmental parameters and cleanliness indicators, overcoming the limitations of traditional linear or simple statistical models, thereby providing more accurate real-time cleanliness evaluation or prediction results.

[0106] As a technical solution for further implementation of this scheme, in step 6, the cleanliness sequence at the next m time points is predicted using the trend prediction formula:

[0107] C future =(c k+1 , c k+2 ,...,c k+m )

[0108] Intelligent alarm triggering conditions:

[0109]

[0110] Where c min is the lower limit of the safety threshold, c max is the upper limit of the safety threshold, d k is the failure probability, d th is the fault threshold;

[0111] It provides the ability to predict future cleanliness changes, facilitating advance planning and intervention; it sets clear dual intelligent alarm trigger conditions based on cleanliness thresholds and failure probabilities to ensure the accuracy and timeliness of alarms, effectively avoiding missed alarms (abnormal cleanliness) and false alarms (equipment failures), making alarm information more targeted and action-guiding.

[0112] The technical solutions for further implementation of this plan include:

[0113] Dataset construction module: used to arrange multiple sensor nodes in the clean space, obtain parameter data of particle concentration, wind speed, temperature, humidity and pressure difference in the clean space at time nodes, and construct a data set;

[0114] Data transmission module: used to transmit sensor data to the data processing server;

[0115] Data preprocessing module: used to remove noise data, erroneous data and duplicate data from the collected data to improve the accuracy and reliability of the data;

[0116] Data fusion processing module: used to fuse data collected by different types of sensors and integrate multi-source data through Kalman filtering algorithm;

[0117] Data analysis module: used to analyze the pre-processed cleanliness-related data using a deep learning model;

[0118] Result display module: used to display clean space parameter data, historical data curves, trend prediction results and fault diagnosis reports in real time, and trigger an alarm when the cleanliness index is abnormal to notify relevant personnel;

[0119] Decision support module: used to generate and display decision support recommendations.

[0120] As a technical solution for further implementing this solution, a computer program is stored thereon, characterized in that when the computer program is executed by a processor, the steps of the method as described in any one of claims 1 to 8 are executed, so that the described innovative detection method can be automatically and efficiently run on the corresponding hardware platform, and the detection method is converted into a product or function that can be actually deployed and applied.

[0121] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method of the present invention and its core ideas. The above is only a preferred implementation method of the present invention. It should be pointed out that due to the limitations of textual expression, there are objectively infinite specific structures. For ordinary technicians in this technical field, without departing from the principles of the present invention, they can make several improvements, modifications or changes, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes or combinations, or the direct application of the inventive concept and technical solution to other occasions without improvement, should be regarded as the scope of protection of the present invention.

Claims

1. A real-time cleanliness detection method based on high-precision clean space, characterized in that: The following steps are involved: Step 1: Dataset construction: Multiple sensor nodes are deployed in the clean space to obtain parameter data of particle concentration p, wind speed v, temperature T, humidity H, and pressure difference △P in the clean space at time node k; Step 2: Transmission of sensor data: The sensor data S is transmitted via wireless communication network. k =[p k ,v k ,T k ,H k ,△P k ] T Transmit to the data processing server; establish a wired network as a backup channel for the wireless communication network; Step 3: Data preprocessing: remove noise data, erroneous data, and duplicate data from the collected data; Step 4: Data fusion processing: Fusion processing is performed on the data collected by different types of sensors, and multi-source data is integrated through the Kalman filter algorithm; Step 5: Data analysis: Use deep learning models to analyze the pre-processed cleanliness-related data; Step 6: Result display: The user interface displays the clean space parameter data, historical data curves, trend prediction results and fault diagnosis reports to the user in real time, and continuously monitors the cleanliness index. If the cleanliness index is monitored to exceed the preset safety threshold or a valid fault is confirmed, an intelligent alarm will be automatically triggered and an alarm message will be immediately sent to the preset relevant responsible personnel through various methods such as SMS, email and application push notification.

2. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: The following steps are also included: Step 7: Based on the trend prediction results and fault diagnosis information, generate and display decision support suggestions, including cleaning and maintenance plans or troubleshooting solutions.

3. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: In step 3, for a single parameter, such as particle concentration p k , perform Z-score anomaly detection to remove erroneous data from the collected data, specifically: If |z k |>3, it is considered an outlier and removed; Where μ p is the mean value of the particle concentration in a window of size N, calculated as σ p is the standard deviation of the particle concentration in a window of size N, calculated as 4. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: In step 3, the time series smoothing method is used to reduce the influence of random noise on the data. k For example, specifically: Where, is the wind speed after preprocessing, and M is the size of the sliding window.

5. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: In step 3, the Euclidean distance formula is used to measure the distance between two n-dimensional data points. and The similarities between them are removed, and duplicate data is removed, specifically: If d(x i , x j )<ε, then x i and x j If it is duplicate data, one of the data points will be retained and the other will be removed; Where, d(x i , x j ) is x i and x j The straight-line distance in n-dimensional space, ε is the set distance threshold.

6. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: In step 4, the data collected by different types of sensors are fused and processed. The integration of multi-source data through the Kalman filter algorithm includes the following steps: S41: Integrate multi-source sensor data and estimate the optimal state: Assume that the state vector x k Consistent with the sensor data, the dynamic model is a constant speed model: Where x k is the state vector at the kth moment, including all sensor parameters; S42: Prediction step: Based on the optimal estimate and dynamic model of the previous moment, the estimated state value of the current moment is calculated in advance: Where, is the prior state estimate at the kth moment, is the posterior state estimate at the k-1th moment, F is the state transfer matrix, P k|k-1 is the prior estimated covariance matrix, P k-1|k-1 is the estimated covariance matrix of the previous moment, F T is the transposed matrix of the state transfer matrix F after the rows and columns are swapped, Q is the process noise covariance matrix; S43: Update step: Correct the predicted value by the current measurement value to obtain the optimal state estimate: If R approaches 0, the measurement value Z is trusted first. k , if P k|k-1 When it approaches 0, the predicted value is trusted first. Where Z k is the measurement vector at the kth moment, H is the observation matrix, R is the measurement noise covariance matrix, P k|k-1 H T (HP k|k- 1H T +R) -1 is the implicit calculation of the Kalman gain.

7. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: In step 5, the deep learning model MLP is used to analyze the preprocessed and fused data and output the cleanliness index or prediction result: Where, f k is the feature vector at the kth moment output after processing in step 4, is the estimated value of the fused parameter; Multilayer Perceptron (MLP) model formula: h k =φ(W1f k +b1) <h2 style=";text-align:left;direction:ltr">c<h2 style=";text-align:left;direction:ltr"> k <h2 style=";text-align:left;direction:ltr"> =W2h<h2 style=";text-align:left;direction:ltr"> k <h2 style=";text-align:left;direction:ltr"> +b2 Where h k is the hidden layer output vector, φ(·) is the activation function, W1, W2 are weight matrices, b1, b2 are bias vectors, c k It is the cleanliness index of the output.

8. The method for real-time detection of cleanliness of a high-precision clean space according to claim 1 is characterized in that: In step 6, the cleanliness sequence at the next m time points is predicted using the trend prediction formula: C future =(c k+1 ,c k+2 ,...,c k+m ) Intelligent alarm triggering conditions: Where c min is the lower limit of the safety threshold, c max is the upper limit of the safety threshold, d k is the failure probability, d th is the fault threshold.

9. A detection system based on the high-precision real-time detection method for clean space cleanliness according to any one of claims 1 to 8, characterized in that: include: Dataset construction module: used to arrange multiple sensor nodes in the clean space, obtain parameter data of particle concentration, wind speed, temperature, humidity and pressure difference in the clean space at time nodes, and construct a data set; Data transmission module: used to transmit sensor data to the data processing server; Data preprocessing module: used to remove noise data, erroneous data and duplicate data from the collected data; Data fusion processing module: used to fuse data collected by different types of sensors and integrate multi-source data through Kalman filtering algorithm; Data analysis module: used to analyze the pre-processed cleanliness-related data using a deep learning model; Result display module: used to display clean space parameter data, historical data curves, trend prediction results and fault diagnosis reports in real time, and trigger an alarm when the cleanliness index is abnormal to notify relevant personnel; Decision support module: used to generate and display decision support recommendations.

10. The detection system based on the high-precision real-time detection method for clean space cleanliness according to claim 9, wherein a computer program is stored thereon, characterized in that When the computer program is executed by a processor, the steps in the method according to any one of claims 1 to 8 are executed.

Images