Intelligent cleaning control method, system and storage medium for reducing the incidence of silicon powder pollution

Abstract

The present invention discloses an intelligent cleaning control method, system and storage medium for reducing the incidence of silicon powder contamination, and relates to the field of silicon wafer cleaning technology. The method includes: presetting an image acquisition interval, using the acquisition interval as a constraint to control the visual sensor array to capture the image of the cleaning tank, and obtaining a water body surface and internal image sequence; analyzing the composition of the contaminants based on the surface image sequence to obtain contaminant time series data; grading the turbidity according to the internal image sequence to obtain turbidity time series data; pre-building a collaborative control model, synchronizing the above data to the model analysis, and outputting collaborative optimization parameters; updating the operating parameters of the cleaning control module with the optimization parameters as constraints. The method solves the technical problems of poor adaptability of the existing cleaning process and high incidence of silicon powder contamination when the silicon powder content is high. By combining the control analysis of the collaborative control model and dynamically adjusting the cleaning control parameters, the technical effect of reducing the incidence of silicon powder contamination and improving the efficiency of the cleaning process is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of silicon wafer cleaning, and in particular to an intelligent cleaning control method, system and storage medium for reducing the incidence of silicon powder contamination. Background Art

[0002] During silicon wafer manufacturing, the cleaning process is crucial for ensuring wafer surface quality, directly impacting the stability and yield of subsequent processes. However, existing cleaning methods primarily rely on a combination of chemicals and pure water, using fixed-parameter mechanical cleaning to remove surface contaminants. This approach exhibits significant limitations when faced with complex and changing contamination situations. When silicon wafers are laden with silicon powder or organic contaminants, the water in the cleaning tank gradually accumulates these pollutants over time, leading to increased turbidity in the cleaning tank water. Traditional methods are unable to dynamically adjust cleaning parameters, which can easily lead to problems such as decreased cleaning efficiency and increased incidence of contaminated wafers. Summary of the Invention

[0003] The present application provides an intelligent cleaning control method, system and storage medium for reducing the incidence of silicon powder contamination, which solves the technical problems of poor adaptability of existing cleaning processes and high incidence of silicon powder contamination when the silicon powder content is high.

[0004] In view of the above problems, the present application provides an intelligent cleaning control method, system and storage medium for reducing the incidence of silicon powder contamination.

[0005] In a first aspect of the present application, an intelligent cleaning control method for reducing the incidence of silicon powder contamination is provided, the method comprising:

[0006] An image acquisition interval is preset, and with the image acquisition interval as a constraint, the visual sensor array is controlled to capture images of the cleaning tank to obtain water image information, wherein the water image information includes a water surface image sequence and a water interior image sequence; dirt composition analysis is performed based on the water surface image sequence to obtain dirt time series data; turbidity classification of the cleaning tank is performed based on the water interior image sequence to obtain turbidity time series data; a collaborative control model is pre-constructed; control analysis is performed by synchronizing the dirt time series data and turbidity time series data to the collaborative control model, and collaborative optimization parameters are output; and operating parameters of the cleaning control module are updated with the collaborative optimization parameters as a constraint.

[0007] A second aspect of the present application provides an intelligent cleaning control system for reducing the incidence of silicon powder contamination, the system comprising:

[0008] Image capture component: presets an image acquisition interval, and uses the image acquisition interval as a constraint to control the visual sensor array to capture images of the cleaning tank and obtain water body image information, wherein the water body image information includes a water body surface image sequence and a water body internal image sequence; dirt composition analysis component: performs dirt composition analysis based on the water body surface image sequence to obtain dirt time series data; turbidity grading component: performs turbidity grading on the cleaning tank according to the water body internal image sequence to obtain turbidity time series data; model pre-construction component: pre-constructs a collaborative control model; control analysis component: performs control analysis by synchronizing the dirt time series data and turbidity time series data to the collaborative control model, and outputs collaborative optimization parameters; operation parameter update component: updates the operation parameters of the cleaning control module based on the collaborative optimization parameters.

[0009] The third aspect of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent cleaning control method for reducing the incidence of silicon powder contamination provided in the present application.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] First, a preset image acquisition interval is used as a constraint to control the visual sensor array to capture images of the cleaning tank, obtaining water image information. This water image information includes a sequence of water surface images and a sequence of water interior images. Next, the surface image sequence is used to analyze the contaminant composition, obtaining contaminant time-series data. The turbidity level of the cleaning tank is then graded based on the water interior image sequence, obtaining turbidity time-series data. Furthermore, a collaborative control model is pre-built. The contaminant time-series data and turbidity time-series data are synchronized with the collaborative control model for control analysis, outputting collaborative optimization parameters. Finally, the operating parameters of the cleaning control module are updated using the collaborative optimization parameters as a constraint. This approach addresses the technical issues of poor adaptability of existing cleaning processes and the high incidence of silicon dust contamination when silicon dust content is high. By combining control analysis with the collaborative control model, the cleaning control parameters are dynamically adjusted, achieving the technical effect of reducing the incidence of silicon dust contamination and improving cleaning process efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1A flow chart of an intelligent cleaning control method for reducing the incidence of silicon powder contamination provided in an embodiment of the present application;

[0014] Figure 2 Schematic diagram of the structure of an intelligent cleaning control system for reducing the incidence of silicon powder contamination provided in an embodiment of the present application.

[0015] Description of reference numerals: image capturing component 11 , dirt composition analyzing component 12 , dirt turbidity classification component 13 , model pre-building component 14 , control analyzing component 15 , operation parameter updating component 16 . DETAILED DESCRIPTION

[0016] The present application solves the technical problems of poor adaptability of existing cleaning processes and high incidence of silicon powder contamination when the silicon powder content is high by providing an intelligent cleaning control method, system and storage medium for reducing the incidence of silicon powder contamination.

[0017] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0018] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products or devices.

[0019] Example 1, as Figure 1 As shown, the present application provides an intelligent cleaning control method for reducing the incidence of silicon powder contamination, wherein the method includes:

[0020] An image acquisition interval is preset, and with the image acquisition interval as a constraint, the visual sensor array is controlled to capture images of the cleaning tank to obtain water body image information, wherein the water body image information includes a water body surface image sequence and a water body interior image sequence.

[0021] In one embodiment, the system terminal first sets a fixed image acquisition interval, which controls the frequency of image acquisition. This preset interval enables the visual sensor array in the cleaning system to continuously capture images of the cleaning tank at a specified interval. The captured image information is primarily divided into two parts: a sequence of images of the water surface, used to analyze the type, distribution, and quantity of surface pollutants; and a sequence of images of the water interior, used to monitor changes in water turbidity. This dual-level image information acquisition provides comprehensive data support for subsequent pollutant analysis and turbidity classification, facilitating precise control and dynamic adjustment of the cleaning process.

[0022] Pollutant composition analysis is performed based on the water body surface image sequence to obtain pollutant time series data.

[0023] In one embodiment, the system terminal uses a sequence of water surface images to perform a detailed analysis of contaminants on the water surface in the cleaning tank. Using a pre-built contaminant ratio recognition model, it extracts information about the contaminant composition, such as organic matter, silica powder clusters, and air bubbles, as well as their quantity and coverage area, from the images. This generates time-series data on the contaminants. This data provides a crucial basis for subsequent optimization and dynamic adjustment of cleaning parameters.

[0024] Furthermore, the method of performing a pollution composition analysis based on the water surface image sequence to obtain pollution time series data includes:

[0025] Interactively obtain multiple sample pollution image sets of multiple sample water body pollution; perform pollution area ratio direction annotation on the multiple sample pollution image sets to obtain multiple sample annotated image sets; use the multiple sample pollution image sets and the multiple sample annotated image sets as training data to perform direction training of a standard pollution recognition model to obtain multiple pollution proportion recognition branches; connect the multiple pollution proportion recognition branches in parallel to obtain a pollution proportion recognition model; synchronize the water body surface image sequence to the pollution proportion recognition model to perform pollution composition analysis to obtain pollution time series data.

[0026] Preferably, the system terminal first interactively acquires a set of images of multiple sample water pollutants. These images represent typical scenes of different water surface pollutants, for example, a combination of organic matter, silica powder clusters, and bubbles. These images are then annotated, with the specific annotation content including the distribution of the area ratio of each pollutant on the water surface. For example, a sample image may be annotated with organic matter accounting for 0.5%, silica powder clusters accounting for 2.5%, and bubbles accounting for 1%. During the labeling process, the total pollutant coverage area (represented by the sum of the coverage areas of organic matter, silica fume clumps, and other debris, S, T, and U, respectively) is also quantified into the following levels: Level 1: (S+T+U) ≤ 1% (total pollutant coverage area less than or equal to 1%); Level 2: (S+T+U) ≤ 2% (total pollutant coverage area less than or equal to 2%); Level 3: (S+T+U) ≤ 3% (total pollutant coverage area less than or equal to 3%); Level 4: (S+T+U) ≤ 4% (total pollutant coverage area less than or equal to 4%); and Level 5: (S+T+U) > 4% (total pollutant coverage area greater than 4%). After labeling, multiple sample pollutant image sets and multiple sample labeled image sets are used to train a standard pollutant recognition model. During the training process, a convolutional neural network (CNN) is used as the core algorithm, gradually extracting the spatial characteristics of pollutants through image processing layers. Before training begins, multiple sample image sets are normalized to standardize pixel values ​​to a uniform range, such as [0, 1], to improve model convergence speed and stability. These labeled sample image sets are then fed into multiple pollutant fraction identification branches, each responsible for identifying a pollutant category and its corresponding coverage area percentage. In the unstarved branches, multiple convolutional layers are used to extract the corresponding sample pollutant image sets. Pooling layers are then used to reduce the dimensionality of the features, gradually compressing the spatial information of the image data and enhancing the representation of important features. Fully connected layers are then used to perform classification and regression on the extracted features to predict the pollutant fraction. During the training of these branches, the mean squared error (MSE) is used as the optimization objective for the regression task to predict the specific percentage of pollutant coverage area. To improve the generalization ability of these branches, data augmentation techniques such as image rotation, flipping, and noise addition are introduced during training to enable these branches to adapt to different pollutant morphological variations. During training, the Adam optimizer is used to adjust branch parameters, and a learning rate decay strategy is set to ensure training stability and efficiency. After training, multiple contaminant proportion recognition branches are generated, corresponding to the analysis capabilities of pollutants such as organic matter, silica fume clumps, and bubbles. These recognition branches are then integrated in parallel to form a complete contaminant proportion recognition model. This model can receive a sequence of water surface images and analyze the type of pollutant and its surface area proportion in the images.For example, through internal analysis of multiple branches, it can be determined that at a certain moment, organic matter accounts for 0.8% (S = 0.8%), silica dust accounts for 1.2% (T = 1.2%), and other debris accounts for 1% (U = 1%) of the water surface pollutants. Using a preset total coverage area assessment strategy, this is quantified as Level 3. Ultimately, the analysis results are arranged in chronological order to generate pollution time series data, providing a basis for subsequent cleaning parameter optimization and dynamic adjustment.

[0027] The turbidity of the cleaning tank is graded according to the image sequence of the interior of the water body to obtain turbidity time series data.

[0028] In one embodiment, a water turbidity analysis model is used to analyze the turbidity of the water in the cleaning tank using a sequence of internal water images and to implement graded management. Specifically, the system terminal first extracts the grayscale value information of the internal water image. The grayscale value reflects the transparency and pollution level of the water. The turbidity of the water is then divided into five levels based on the grayscale value range (for example, grayscale values ​​of 240-248 are the highest level N, indicating clear water; grayscale values ​​of 0-128 are the lowest level R, indicating severe turbidity). The grayscale value of each image is mapped to the corresponding level after calculation using the model. Over time, the system terminal analyzes the water image at each time point and records the turbidity changes within different time periods to generate turbidity time series data. This time series data can reflect the dynamic changes in water pollution over time, providing key support for the subsequent optimization and adjustment of cleaning parameters.

[0029] Furthermore, the turbidity of the cleaning tank is graded according to the image sequence of the interior of the water body to obtain turbidity time series data, and the method includes:

[0030] A water body grade analysis model is pre-constructed, wherein the water body grade analysis model includes a cascaded grayscale analysis model and a water body grade judgment module; a first water body internal image corresponding to a first time series is extracted from the water body internal image sequence; the first water body internal image is synchronized to the grayscale analysis model of the water body grade analysis model for analysis to obtain a first water body turbidity; the first water body turbidity is synchronized to the water body grade judgment module of the water body grade analysis model for analysis to obtain a first water body grade; and so on, the turbidity of the cleaning tank is graded by synchronizing the water body internal image sequence to the water body grade analysis model to obtain the turbidity time series data.

[0031] Optionally, during the cleaning control process, the system terminal first pre-builds a water quality analysis model. This model consists of two core components: a grayscale analysis model and a water quality judgment module, which are cascaded together to perform joint analysis. The grayscale analysis model, trained based on sample data, extracts and analyzes grayscale values ​​from images of the water's interior. The water quality judgment module determines the specific water quality based on the grayscale values. After the water quality analysis model is determined, the system terminal extracts the image corresponding to the first time point from the water quality image sequence as the first water quality image. This image captures the turbidity of the water in the cleaning tank at that moment. The first water quality image is then input into the grayscale analysis model, which calculates the image's grayscale values ​​pixel by pixel. The average grayscale value reflects the water's transparency and pollution level. For example, a higher grayscale value indicates clear water, while a lower grayscale value indicates severe turbidity. The output of the analysis is the first water quality turbidity level. Afterwards, the turbidity of the first water body is synchronized to the water body grade judgment module, and the module grades the turbidity according to a preset gray value grade interval. For example, a grayscale value in the range of (240, 248] corresponds to water level N (clear). A grayscale value in the range of (128, 192] corresponds to water level Q (lightly polluted). A grayscale value in the range of [0, 128] corresponds to water level R (severely polluted). The module determines the first water level based on the grayscale value range and records the level information. The above process is repeated, analyzing each time-series image in the internal image sequence of the water body, extracting and grading its turbidity one by one, and arranging the grading results in chronological order to generate complete turbidity time-series data. Ultimately, the turbidity time-series data reflects the pollution change trend of the cleaning tank water throughout the entire cleaning cycle, providing a basis for subsequent adjustment of cleaning parameters (such as ultrasonic power and jet volume). Through this process, real-time monitoring and dynamic grading of water turbidity can be achieved, ensuring the stability and accuracy of the cleaning effect, while also providing data support for reducing energy consumption and improving cleaning efficiency.

[0032] Furthermore, a water body grade analysis model is pre-built, and the method includes:

[0033] Interactively obtain multiple sample water body turbidity images and multiple sample water body turbidity levels; use the multiple sample water body turbidity images and multiple sample water body turbidity levels as training data to complete the construction of the grayscale analysis model; interactively obtain multiple water body turbidity intervals of multiple water body grades; associate and store the multiple water body grades and multiple water body turbidity intervals based on a knowledge graph to complete the construction of the water body grade judgment module; and complete the construction of the water body grade analysis model by cascading the grayscale analysis model and the water body grade judgment module.

[0034] Optionally, in the process of building a water body grade analysis model, the system terminal interactively obtains multiple sample water body turbidity images and corresponding sample water body turbidity. These images represent different degrees of water body pollution, and the turbidity is calculated by grayscale value. For example, the grayscale value of an image is 200, reflecting a medium pollution state. Subsequently, the collected sample water body turbidity images and their corresponding sample turbidity are used as training data and input into the grayscale analysis model based on the convolutional neural network. The model learns the grayscale distribution characteristics of the image through the same training method as mentioned above, and gradually optimizes the parameters to achieve accurate extraction of the grayscale value of the input image and prediction of the turbidity. Afterwards, through data statistics or industry standards, multiple levels of water bodies and their corresponding grayscale value intervals are interactively defined, and these intervals are used as water body turbidity intervals. For example, the grayscale value of level N (clear) is ∈(240,248], the grayscale value of level O (slightly polluted) is ∈(224,240], the grayscale value of level P (moderately polluted) is ∈(192,224], the grayscale value of level Q (heavily polluted) is ∈(128,192], and the grayscale value of level R (severely polluted) is ∈[0,128]. Then, using the knowledge graph technology, the above-mentioned multiple water body levels are associated with the corresponding water body turbidity intervals and stored. For example, the knowledge graph node can contain the following information: Node 1: Level N → Water body turbidity interval (240,248], Node 2 :Level O→Water turbidity interval (224,240]. Through the association relationship of the knowledge graph, a water body level judgment module is constructed, which enables the corresponding water body level to be quickly judged according to the input turbidity value. Then, the grayscale analysis model and the water body level judgment module are cascaded and integrated to construct a water body level analysis model. The input water body turbidity image is first subjected to the grayscale analysis model in the water body level analysis model to extract the grayscale value and predict the turbidity, and then the predicted turbidity is synchronized to the water body level judgment module. The water body level is determined according to the water body turbidity interval, providing key support for cleaning control.

[0035] Pre-built collaborative control models.

[0036] In one embodiment, pre-building a collaborative control model is a key step in achieving precise dynamic adjustment during the cleaning process. Specifically, the collaborative control model integrates multiple data sources (such as dirt time-series data and turbidity time-series data) to achieve collaborative optimization and regulation of cleaning tank operating parameters (such as air flow and ultrasonic power). The construction process first requires the collection of a large amount of sample data, including multiple arrays of different pollutant proportions and corresponding air flow adjustment values, as well as data correlating multiple sample water levels with ultrasonic power parameters. After cleaning, classification, and standardization, this data serves as input for model training. Subsequently, using methods such as regression analysis, the air flow adjustment branch and the power adjustment branch are constructed. These two branches are integrated in parallel to ensure that the model can simultaneously address the optimization requirements of air flow and ultrasonic power. The model design adopts a hierarchical structure, with dirt time-series data input into the air flow adjustment branch to calculate the optimal air flow, while turbidity time-series data is input into the power adjustment branch to optimize ultrasonic power output. The model's collaborative control logic dynamically analyzes the interrelationships between these data to comprehensively evaluate the optimal cleaning parameter combination. Ultimately, the collaborative control model outputs a set of collaboratively optimized parameters (including air jet volume and ultrasonic power) to provide decision support for the cleaning control module. Through this pre-built process, the collaborative control model can effectively respond to different contamination conditions and achieve the optimal balance between cleaning efficiency and energy consumption.

[0037] Furthermore, the collaborative control model is pre-built, and the method includes:

[0038] Interactively obtain multiple sample dirt ratio arrays and multiple sample gas volume adjustment values; perform multi-source regression analysis on the multiple sample dirt ratio arrays and the multiple sample gas volume adjustment values ​​to obtain a gas volume adjustment function; construct a gas volume adjustment branch based on the gas volume adjustment function; interactively obtain multiple sample power parameters and multiple groups of sample water body levels; use a table structure to associate and store the multiple sample power parameters and the multiple groups of sample water body levels to complete the construction of the power adjustment branch; connect the gas volume adjustment branch and the power adjustment branch in parallel to complete the construction of the collaborative control model.

[0039] Optionally, during the collaborative control model construction process, the system terminal interactively obtains multiple sample contaminant percentage arrays and corresponding sample air volume adjustment values. This data records the distribution of contaminants under different cleaning conditions (the percentage array includes the area ratio of organic matter, silicon powder clusters, and other debris, such as S+T+U) and the required air volume to ensure cleaning effectiveness. Subsequently, a multi-source regression analysis method is used to model the relationship between the sample contaminant percentage array (for example, Level 1: S+T+U ≤ 1%) and the air volume adjustment value (for example, air volume 5%). Using the least squares method or other regression algorithm, an air volume adjustment function is generated that dynamically predicts the optimal air volume based on the input contaminant percentage. For example, when (S+T+U) ≤ 3%, the predicted air volume is 10%. The generated air volume adjustment function is then encapsulated to construct an output air volume adjustment branch. This branch receives real-time contaminant time series data and uses the air volume adjustment function to calculate the optimal air volume required for the current cleaning tank. For example, if the input contaminant percentage is 2.5%, the air volume adjustment branch outputs an air volume of 10%. Afterwards, multiple sample power parameters (such as ultrasonic power: 80%, 90%, 100%) and corresponding sample water body levels (such as N, O, P, Q, R) are obtained interactively. These data reflect the adaptability of different cleaning intensities to water bodies with different turbidity levels. For example, for samples with a water grayscale value of (224, 240], the recorded power is 90%. A table structure is then used to store the correspondence between power parameters and water body levels. For example, level N (clear): grayscale value (240, 248], power 100%, level Q (heavily polluted): grayscale value (128, 192], power 80%. Through this storage method, the power adjustment branch can quickly query and call the corresponding power parameters. Then, based on the stored association, a power adjustment branch is designed. This branch can receive turbidity time series data and dynamically adjust the ultrasonic power output according to the current water body level. For example, if the input grayscale value is 200, the power adjustment branch outputs power 90%. Finally, the air volume regulation branch and the power regulation branch are integrated in a parallel structure to form a collaborative control model. The air volume regulation branch optimizes the air jet volume based on the pollutant ratio array, while the power regulation branch optimizes the ultrasonic power based on the water level. The two operate independently but produce coordinated outputs. After the model is constructed, its accuracy and reliability are verified through test data, and the air volume regulation function and power parameter storage rules are further optimized to ensure the model's efficiency and accuracy in actual cleaning operations. Through this process, the collaborative control model can dynamically receive and analyze dirt time series data and turbidity time series data, and output the optimal air jet volume and power parameters, providing real-time control capabilities for the cleaning control module.

[0040] The pollution time series data and the turbidity time series data are synchronized to the collaborative control model to perform control analysis and output collaborative optimization parameters.

[0041] In one embodiment, during the cleaning control process, the system terminal synchronously inputs real-time dirt time-series data and turbidity time-series data into the collaborative control model. The collaborative control model comprehensively analyzes these two types of data to identify the correlation between pollutants and water turbidity. Based on built-in adjustment logic and optimization rules, it calculates the optimal cleaning parameters for the current cleaning state as collaborative optimization parameters, including the optimal air jet volume and ultrasonic power. These parameters are directly transmitted to the cleaning control module, which is used to dynamically adjust the operating state of the cleaning equipment, thereby achieving efficient pollutant removal and energy consumption optimization. Through this process, it is possible to respond to water changes in real time under complex cleaning conditions, maintaining the stability and accuracy of the cleaning effect.

[0042] Furthermore, by synchronizing the pollution time series data and the turbidity time series data to the collaborative control model for control analysis and outputting collaborative optimization parameters, the method includes:

[0043] The pollution time series data is synchronized to the air volume regulation branch of the collaborative control model to solve the jet volume and obtain multiple alternative optimization parameters; the multiple alternative optimization parameters are serialized, and the maximum value is extracted according to the sorting result to obtain the jet volume optimization parameter; the turbidity time series data is serialized, and the maximum value is extracted according to the sorting result to obtain the turbidity extreme value; the turbidity extreme value is synchronized to the power regulation branch of the collaborative control model to solve the power optimization and output the ultrasonic optimization power; the jet volume optimization parameter and the ultrasonic optimization power constitute the collaborative optimization parameter.

[0044] Optionally, the system terminal first synchronizes the pollutant time series data to the air volume regulation branch of the collaborative control model. The air volume regulation branch uses the air volume regulation function to calculate the jet volume at different time points based on the type of pollutants (such as bubbles, organic matter, silicon powder clumps, etc.), coverage area (such as S+T+US+T+U), and dynamic changes in pollutants, and generates multiple alternative optimization parameters. These alternative parameters are then serialized, and the maximum value is extracted based on the sorting results to obtain the jet volume optimization parameters to ensure the rapid removal of floating pollutants on the water surface. At the same time, the turbidity time series data is serialized and analyzed. For example, the grayscale value at time t1 is 200, corresponding to the turbidity level P; the grayscale value at time t2 is 180, corresponding to the turbidity level Q. The maximum grayscale is extracted from these time series data as the turbidity extreme value. Subsequently, the extracted turbidity extreme value is synchronized to the power regulation branch of the collaborative control model. The power regulation branch calculates the optimal ultrasonic power based on the association rules between turbidity level and ultrasonic power (such as P level corresponds to 90% power, Q level corresponds to 80% power). The adjusted ultrasonic power can effectively decompose the silicon powder particles and other stubborn dirt deposited at the bottom of the water body, thereby improving the cleanliness of the water body. The jet volume optimization parameters and ultrasonic optimization power together constitute the collaborative optimization parameters, which act on the jet control unit and ultrasonic power control unit of the cleaning system respectively. For example, 10.8% of the jet volume is used to quickly remove floating dirt on the surface of the water body, and 90% of the ultrasonic power is used to deal with particles deposited at the bottom of the water body, ensuring the efficiency and accuracy of the entire cleaning process. Through this collaborative optimization process, the system terminal can comprehensively consider the pollution conditions on the surface and inside of the water body, dynamically adjust the cleaning parameters, and achieve comprehensive removal of dirt on the surface of the water body and fine particles on the bottom, while optimizing energy consumption and cleaning effects.

[0045] The operating parameters of the cleaning control module are updated with the collaborative optimization parameters as constraints.

[0046] In one embodiment, the collaborative optimization parameters generated by the collaborative control model (including the jet volume optimization parameter and the ultrasonic power optimization parameter) are used as constraints to dynamically adjust the operating parameters of the cleaning control module to achieve precise and efficient cleaning operations. During the cleaning process, the jet control unit and the ultrasonic power control unit in the cleaning control module work together, adjusting the operating parameters in real time to ensure that the entire cleaning system can adapt to different pollution conditions, achieving efficient removal of floating dirt on the water surface and sediment particles at the bottom, while maximizing energy consumption and cleaning effect.

[0047] Furthermore, the operating parameters of the cleaning control module are updated with the collaborative optimization parameters as constraints, and the method includes:

[0048] The jet volume optimization parameters are sent to the jet control unit of the cleaning control module; the ultrasonic optimization power is sent to the power control unit of the cleaning control module; when the jet volume of the jet control unit is dynamically adjusted based on the jet volume optimization parameters, the ultrasonic power of the power control unit is dynamically adjusted based on the ultrasonic optimization power.

[0049] Optionally, during the cleaning control process, the collaborative control model calculates the optimized parameter of the jet volume (for example, the jet volume is 10.8%), which is first sent to the jet control unit of the cleaning control module. The jet control unit dynamically adjusts the nozzle opening and the outlet pressure according to the parameter to keep the jet volume at the optimized value of 10.8% to quickly remove floating pollutants (such as bubbles, organic matter, and silicon powder clumps) on the surface of the water body. At the same time, the ultrasonic optimization power (for example, the power is 90%) generated by the collaborative control model is sent to the power control unit. After receiving the parameter, the power control unit adjusts the output power of the ultrasonic generator so that the ultrasonic emission intensity reaches the optimized value of 90%. This adjustment is aimed at fine particle pollutants (such as silicon powder particles) deposited at the bottom of the cleaning tank, and achieves efficient decomposition and removal of particles through optimized ultrasonic energy. During operation, the jet control unit and the power control unit maintain synchronous dynamic adjustment with the collaborative optimization parameters as constraints. When the jet control unit performs jet volume adjustment (such as fine-tuning the jet volume to between 10.5% and 11% based on real-time feedback), the power control unit will simultaneously respond to the ultrasonic power (such as adjusting the power output to between 89% and 91%) to ensure that the two work together to avoid the impact of changes in a single parameter on the cleaning effect. The cleaning system monitors the jet effect and ultrasonic cleaning effect in real time through sensors, and feeds the monitoring data back to the collaborative control model. The model further optimizes the parameter settings of the jet volume and ultrasonic power based on the latest dirt time series data and turbidity time series data to ensure that the entire cleaning process is always efficient and stable. Through this process, the cleaning control module can quickly remove floating dirt on the surface of the water body and thoroughly remove bottom sediment particles under the synergistic effect of the jet volume optimization parameters and ultrasonic optimization power, achieving comprehensive and efficient cleaning operations while effectively controlling energy consumption.

[0050] Further, the method comprises:

[0051] A cleaning control update window is preset; when the duration of the image capture of the cleaning tank by the visual sensor array constrained by the image acquisition interval reaches the cleaning control update window, the water body image information is updated; and the collaborative control model cyclically updates the operating parameters of the cleaning control module based on the cleaning control update window.

[0052] Preferably, the system terminal pre-sets a cleaning control update window based on the contamination characteristics of the cleaning tank and the performance requirements of the cleaning system. For example, setting the update window to 10 minutes means that a comprehensive cleaning parameter update is performed every 10 minutes. This window serves as a periodic constraint for the cleaning control module to adjust operating parameters. During the cleaning operation, the visual sensor array monitors the cleaning tank in real time according to a preset image acquisition interval (for example, capturing an image every 30 seconds). When the cumulative image acquisition time reaches the cleaning control update window (e.g., 10 minutes), a comprehensive update of the water body image information is triggered, including image sequences of the water body surface and interior. This updated image information is used to recalculate the dirt time series data and turbidity time series data, ensuring that the analysis results reflect the latest cleaning status in real time. After receiving the updated dirt time series data and turbidity time series data, the collaborative control model performs a new round of parameter optimization calculations, using the cleaning control update window as a periodic trigger condition. The model dynamically adjusts the output of the gas volume control branch and the power control branch based on the latest data. For example, if the dirt coverage area decreases from 3% to 1.5%, the optimized jet volume parameter output by the air volume control branch is adjusted from 10.8% to 7%. If the water grayscale value increases from 200 to 230, the optimized ultrasonic power output by the power control branch is adjusted from 90% to 100%. The new optimized parameters calculated by the collaborative control model are sent to the jet control unit and power control unit of the cleaning control module. These units dynamically adjust the operating status of the cleaning tank based on the new optimized parameters to adapt to the latest cleaning requirements. The cleaning control update window cycles continuously at a preset period. When the next update window begins, the system terminal collects a new round of water image information and optimizes the operating parameters again using the collaborative control model. This cyclical cycle ensures that the cleaning system can maintain efficient cleaning results and energy utilization according to the dynamic changes in actual pollution conditions. This process implements intelligent operating parameter updates based on time constraints, periodically and dynamically optimizing the cleaning operation, enabling it to adapt to complex and changing pollution conditions and maintain efficient operation.

[0053] In summary, the embodiments of the present application have at least the following technical effects:

[0054] First, a preset image acquisition interval is used as a constraint to control the visual sensor array to capture images of the cleaning tank, obtaining water image information. This water image information includes a sequence of water surface images and a sequence of water interior images. Next, the surface image sequence is used to analyze the contaminant composition, obtaining contaminant time-series data. The turbidity level of the cleaning tank is then graded based on the water interior image sequence, obtaining turbidity time-series data. Furthermore, a collaborative control model is pre-built. The contaminant time-series data and turbidity time-series data are synchronized with the collaborative control model for control analysis, outputting collaborative optimization parameters. Finally, the operating parameters of the cleaning control module are updated using the collaborative optimization parameters as a constraint. This approach addresses the technical issues of poor adaptability of existing cleaning processes and the high incidence of silicon dust contamination when silicon dust content is high. By combining control analysis with the collaborative control model, the cleaning control parameters are dynamically adjusted, achieving the technical effect of reducing the incidence of silicon dust contamination and improving cleaning process efficiency.

[0055] Example 2, based on the same inventive concept as the intelligent cleaning control method for reducing the incidence of silicon powder pollution in the previous embodiment, Figure 2 As shown, the present application provides an intelligent cleaning control system for reducing the incidence of silicon powder pollution, wherein the system includes:

[0056] Image capture component 11: presets an image acquisition interval, and uses the image acquisition interval as a constraint to control the visual sensor array to capture images of the cleaning tank and obtain water body image information, wherein the water body image information includes a water body surface image sequence and a water body internal image sequence; dirt composition analysis component 12: performs dirt composition analysis based on the water body surface image sequence to obtain dirt time series data; turbidity grading component 13: performs turbidity grading on the cleaning tank according to the water body internal image sequence to obtain turbidity time series data; model pre-construction component 14: pre-constructs a collaborative control model; control analysis component 15: performs control analysis by synchronizing the dirt time series data and turbidity time series data to the collaborative control model, and outputs collaborative optimization parameters; operation parameter update component 16: uses the collaborative optimization parameters as a constraint to update the operation parameters of the cleaning control module.

[0057] Furthermore, the dirt composition analysis component 12 is used to perform the following method:

[0058] Interactively obtain multiple sample pollution image sets of multiple sample water body pollution; perform pollution area ratio direction annotation on the multiple sample pollution image sets to obtain multiple sample annotated image sets; use the multiple sample pollution image sets and the multiple sample annotated image sets as training data to perform direction training of a standard pollution recognition model to obtain multiple pollution proportion recognition branches; connect the multiple pollution proportion recognition branches in parallel to obtain a pollution proportion recognition model; synchronize the water body surface image sequence to the pollution proportion recognition model to perform pollution composition analysis to obtain pollution time series data.

[0059] Furthermore, the turbidity classification component 13 is used to perform the following method:

[0060] A water body grade analysis model is pre-constructed, wherein the water body grade analysis model includes a cascaded grayscale analysis model and a water body grade judgment module; a first water body internal image corresponding to a first time series is extracted from the water body internal image sequence; the first water body internal image is synchronized to the grayscale analysis model of the water body grade analysis model for analysis to obtain a first water body turbidity; the first water body turbidity is synchronized to the water body grade judgment module of the water body grade analysis model for analysis to obtain a first water body grade; and so on, the turbidity of the cleaning tank is graded by synchronizing the water body internal image sequence to the water body grade analysis model to obtain the turbidity time series data.

[0061] Furthermore, the turbidity classification component 13 is used to perform the following method:

[0062] Interactively obtain multiple sample water body turbidity images and multiple sample water body turbidity levels; use the multiple sample water body turbidity images and multiple sample water body turbidity levels as training data to complete the construction of the grayscale analysis model; interactively obtain multiple water body turbidity intervals of multiple water body grades; associate and store the multiple water body grades and multiple water body turbidity intervals based on a knowledge graph to complete the construction of the water body grade judgment module; and complete the construction of the water body grade analysis model by cascading the grayscale analysis model and the water body grade judgment module.

[0063] Furthermore, the model pre-construction component 14 is used to perform the following method:

[0064] Interactively obtain multiple sample dirt ratio arrays and multiple sample gas volume adjustment values; perform multi-source regression analysis on the multiple sample dirt ratio arrays and the multiple sample gas volume adjustment values ​​to obtain a gas volume adjustment function; construct a gas volume adjustment branch based on the gas volume adjustment function; interactively obtain multiple sample power parameters and multiple groups of sample water body levels; use a table structure to associate and store the multiple sample power parameters and the multiple groups of sample water body levels to complete the construction of the power adjustment branch; connect the gas volume adjustment branch and the power adjustment branch in parallel to complete the construction of the collaborative control model.

[0065] Furthermore, the control and analysis component 15 is used to perform the following method:

[0066] The pollution time series data is synchronized to the air volume regulation branch of the collaborative control model to solve the jet volume and obtain multiple alternative optimization parameters; the multiple alternative optimization parameters are serialized, and the maximum value is extracted according to the sorting result to obtain the jet volume optimization parameter; the turbidity time series data is serialized, and the maximum value is extracted according to the sorting result to obtain the turbidity extreme value; the turbidity extreme value is synchronized to the power regulation branch of the collaborative control model to solve the power optimization and output the ultrasonic optimization power; the jet volume optimization parameter and the ultrasonic optimization power constitute the collaborative optimization parameter.

[0067] Furthermore, the operating parameter updating component 16 is used to perform the following method:

[0068] The jet volume optimization parameters are sent to the jet control unit of the cleaning control module; the ultrasonic optimization power is sent to the power control unit of the cleaning control module; when the jet volume of the jet control unit is dynamically adjusted based on the jet volume optimization parameters, the ultrasonic power of the power control unit is dynamically adjusted based on the ultrasonic optimization power.

[0069] Furthermore, the operating parameter updating component 16 is used to perform the following method:

[0070] A cleaning control update window is preset; when the duration of the image capture of the cleaning tank by the visual sensor array constrained by the image acquisition interval reaches the cleaning control update window, the water body image information is updated; and the collaborative control model cyclically updates the operating parameters of the cleaning control module based on the cleaning control update window.

[0071] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

[0073] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. An intelligent cleaning control method for reducing the incidence of silicon powder pollution, characterized in that: The method comprises: Preset an image acquisition interval, and use the image acquisition interval as a constraint to control the visual sensor array to capture images of the cleaning tank to obtain water image information, wherein the water image information includes a water surface image sequence and a water interior image sequence; Performing a pollution composition analysis based on the water body surface image sequence to obtain pollution time series data; Classifying the turbidity of the cleaning tank according to the image sequence of the interior of the water body to obtain turbidity time series data; Pre-built collaborative control model; Synchronizing the pollution time series data and the turbidity time series data to the collaborative control model for control analysis, and outputting collaborative optimization parameters; Taking the collaborative optimization parameters as constraints, updating the operating parameters of the cleaning control module; The pollutant composition analysis is performed based on the water surface image sequence to obtain pollutant time series data, including: Interactively obtaining multiple sample pollution image sets of multiple sample water pollution; Performing directional marking of the dirt area ratio on the multiple sample dirt image sets to obtain multiple sample marked image sets; Using the multiple sample dirt image sets and the multiple sample annotated image sets as training data, performing directional training on a standard dirt recognition model to obtain multiple dirt proportion recognition branches; Connecting the plurality of dirt proportion identification branches in parallel to obtain a dirt proportion identification model; The water body surface image sequence is synchronized with the dirt proportion identification model to perform dirt composition analysis and obtain dirt time series data.

2. The intelligent cleaning control method for reducing the incidence of silicon powder pollution according to claim 1, characterized in that: The turbidity of the cleaning tank is graded according to the image sequence of the interior of the water body to obtain turbidity time series data, and the method includes: Pre-constructing a water body grade analysis model, wherein the water body grade analysis model includes a cascaded grayscale analysis model and a water body grade judgment module; extracting a first water body interior image corresponding to a first time sequence from the water body interior image sequence; Synchronizing the first water body internal image to the grayscale analysis model of the water body grade analysis model for analysis to obtain the first water body turbidity; Synchronizing the first water body turbidity to the water body grade judgment module of the water body grade analysis model for analysis to obtain a first water body grade; The turbidity time series data of the cleaning tank is obtained by synchronizing the image sequence of the interior of the water body to the water body grade analysis model to perform turbidity classification of the cleaning tank.

3. The intelligent cleaning control method for reducing the incidence of silicon powder pollution according to claim 2, characterized in that: A water body grade analysis model is pre-built, and the method includes: Interactively obtain multiple sample water pollution images and multiple sample water turbidity levels; Using the plurality of sample water pollution images and the plurality of sample water turbidity levels as training data, completing the construction of the grayscale analysis model; Interactively obtain multiple water turbidity intervals of multiple water body levels; The plurality of water body grades and the plurality of water body turbidity intervals are stored in association with each other based on the knowledge graph, thereby completing the construction of the water body grade judgment module; The construction of the water body grade analysis model is completed by cascading the grayscale analysis model and the water body grade judgment module.

4. The intelligent cleaning control method for reducing the incidence of silicon powder pollution according to claim 3, characterized in that: Pre-building a collaborative control model, the method comprising: Interactively obtain multiple sample dirt percentage arrays and multiple sample gas volume adjustment values; Performing multi-source regression analysis on the plurality of sample dirt percentage arrays and the plurality of sample air volume adjustment values ​​to obtain an air volume adjustment function; Constructing an air volume regulation branch based on the air volume regulation function; Interactively obtain multiple sample power parameters and multiple groups of sample water body grades; A table structure is used to associate and store the multiple sample power parameters and the multiple groups of sample water body levels to complete the construction of the power regulation branch; The gas volume regulating branch and the power regulating branch are connected in parallel to complete the construction of the collaborative control model.

5. The intelligent cleaning control method for reducing the incidence of silicon powder pollution according to claim 4, characterized in that: The method includes: synchronizing the pollution time series data and the turbidity time series data to the collaborative control model for control analysis and outputting collaborative optimization parameters. The air injection volume is solved by synchronizing the waste time series data to the air volume adjustment branch of the collaborative control model to obtain multiple candidate optimization parameters; Serializing the plurality of candidate optimization parameters, and extracting maximum values ​​according to the sorting results to obtain an optimized parameter for the jet flow rate; Serializing the pollution time series data, and extracting the maximum value according to the sorting result to obtain the extreme value of pollution; Synchronizing the extreme value of the pollution to the power regulation branch of the collaborative control model to perform power optimization and output ultrasonic optimized power; The jet volume optimization parameter and the ultrasonic wave optimization power constitute the collaborative optimization parameter.

6. The intelligent cleaning control method for reducing the incidence of silicon powder pollution according to claim 5, characterized in that: Taking the collaborative optimization parameters as constraints, updating the operating parameters of the cleaning control module, the method includes: Sending the jet quantity optimization parameter to the jet control unit of the cleaning control module; sending the ultrasonic optimized power to the power control unit of the cleaning control module; When the jet volume of the jet control unit is dynamically adjusted with the jet volume optimization parameter as a constraint, the ultrasonic power of the power control unit is dynamically adjusted with the ultrasonic optimization power as a constraint.

7. The intelligent cleaning control method for reducing the incidence of silicon powder pollution according to claim 1, characterized in that: The method comprises: Preset cleaning control update window; When the duration of the image capture of the cleaning tank by the visual sensor array with the image acquisition interval as a constraint reaches the cleaning control update window, updating the water body image information; The collaborative control model performs cyclic updates on the operating parameters of the cleaning control module based on the cleaning control update window as a constraint.

8. An intelligent cleaning control system that reduces the incidence of silicon powder pollution is characterized by: The intelligent cleaning control method for reducing the incidence of silicon powder contamination according to any one of claims 1 to 7 is implemented, wherein the system comprises: Image capture component: presetting an image acquisition interval and using the image acquisition interval as a constraint to control the visual sensor array to capture images of the cleaning tank to obtain water body image information, wherein the water body image information includes a water body surface image sequence and a water body interior image sequence; Pollutant composition analysis component: performs pollutant composition analysis based on the water surface image sequence to obtain pollutant time series data; Turbidity classification component: classifies the turbidity of the cleaning tank according to the image sequence of the interior of the water body to obtain turbidity time series data; Model pre-built components: pre-built collaborative control model; Control analysis component: synchronizes the pollution time series data and the turbidity time series data to the collaborative control model to perform control analysis and output collaborative optimization parameters; Operation parameter updating component: updates the operation parameters of the cleaning control module based on the collaborative optimization parameters as constraints.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the intelligent cleaning control method for reducing the occurrence rate of silicon powder pollution as described in any one of claims 1 to 7 is implemented.

Images