Real-time monitoring method, system and storage medium for electroplating production process

Abstract

The present invention relates to the field of industrial automation and intelligent manufacturing technologies, and particularly to a real-time monitoring method, system and storage medium for the electroplating production process. By combining visual and traditional sensor data and using advanced feature engineering techniques to monitor the electroplating production process in real time, the dimension and quality of the monitoring data are improved. The multi-level machine learning models applied in the solution include deep learning, time series analysis and unsupervised learning, which effectively analyze complex production data and improve the comprehensiveness and depth of anomaly detection. By integrating the outputs of each model through decision fusion technology, a comprehensive anomaly detection report is generated, thereby improving the reliability and accuracy of diagnosis. In addition, the present invention adopts a dynamic feedback and self-adjustment mechanism to automatically adjust the production line parameters according to the anomaly detection results, optimize the process flow in real time, and significantly improve the production efficiency and product quality.

Description

Technical Field

[0001] The present invention relates to the technical fields of industrial automation and intelligent manufacturing, and particularly to a real-time monitoring method, system and storage medium for the electroplating production process. Background Art

[0002] In modern manufacturing, real-time monitoring of the electroplating production process is crucial because it directly affects product quality and production efficiency. Electroplating is a technique that deposits a thin layer of metal on the surface of an object through an electrochemical reaction and is widely used in industries such as automotive, electronics, and aviation. Since electroplating involves complex chemical reactions and precise operating conditions, such as temperature, current, and chemical composition, real-time monitoring has become the key to ensuring coating quality and uniformity.

[0003] The problems existing in electroplating production monitoring mainly include: parameter fluctuations, where small fluctuations in temperature and current during the electroplating process can affect the coating quality; uneven chemical composition, where uneven mixing of chemical components in the electroplating solution may result in uneven coatings; equipment aging, where the performance of the equipment degrades over long-term operation, affecting the stability of the entire electroplating process. These problems are usually caused by insufficient sensor accuracy, outdated data processing methods, or response delays in the monitoring system.

[0004] Although traditional electroplating monitoring systems (Chinese invention patent, publication number CN114757473A) can handle some basic monitoring tasks, they show obvious deficiencies when faced with complex and changing production conditions: traditional systems usually rely on simple threshold judgments or basic statistical methods and are difficult to handle complex situations with non-linearity or multiple parameters; they lack the ability to process real-time data and provide immediate feedback for adjustment, resulting in insufficient response to sudden problems and inability to effectively prevent the occurrence or expansion of problems; they mainly rely on traditional sensor data, such as temperature and current, and ignore data from other dimensions such as vision, limiting the accuracy of fault diagnosis and quality control. Summary of the Invention

[0005] In view of the many problems existing in the above-mentioned prior art, the present invention provides a real-time monitoring method, system and storage medium for the electroplating production process. The present invention comprehensively applies multi-modal data acquisition, advanced feature engineering, multi-level machine learning models, and decision fusion technology to achieve comprehensive monitoring and dynamic adjustment of the electroplating production process; by integrating data from visual sensors and traditional sensors, the present invention can capture production anomaly signals in more dimensions; multi-level machine learning models such as convolutional neural networks, long short-term memory networks, unsupervised learning algorithms, and autoencoders are used to analyze this data to improve the accuracy and depth of anomaly detection; the present invention significantly improves the accuracy of anomaly detection and the adaptability of the production process, reduces production defects, and improves production efficiency; the dynamic feedback mechanism allows real-time adjustment of process parameters, optimizes production operations, prevents the expansion of problems, and ensures production quality.

[0006] A real-time monitoring method for the electroplating production process includes the following steps:

[0007] Collect the original data of the electroplating production line, where the original data includes temperature, pressure, current, and video / images. Denoise and standardize the original data, and extract visual features, sensor statistics, and time series features to generate a comprehensive feature dataset;

[0008] Use a convolutional neural network to process visual features, a long short-term memory network to analyze time series features, and at the same time apply unsupervised learning models and autoencoder models for anomaly clustering and data reconstruction. Independently analyze through the comprehensive feature dataset and output the corresponding results;

[0009] Integrate the output results of each model using a decision fusion algorithm to generate an anomaly detection report, clarify the location and type of anomaly points, and identify all potential anomaly points;

[0010] Based on the anomaly detection report, dynamically adjust the process parameters of the production line, monitor the adjustment effect in real time, and continuously optimize the adjustment strategy according to the continuously collected original data.

[0011] Preferably, use image processing algorithms to extract and obtain visual features, calculate sensor statistics and time series features through sensor data, and generate a comprehensive feature dataset.

[0012] Preferably, identify abnormal patterns in the image through a convolutional neural network. The abnormal patterns include: unusual color changes, shape or texture anomalies, obtain visual anomaly detection results, and identify all visually abnormal points;

[0013] Analyze the patterns and trends in the time series features through a long short-term memory network, identify long-term patterns and trends in the data, predict future abnormal behaviors, obtain time series anomaly detection results, and be used to display any potential anomalies in the sensor data.

[0014] Preferably, an unsupervised learning algorithm is applied to cluster visual features and time series features, group data points according to the similarity of data, generate an abnormal clustering result, and the abnormal clustering result is used to identify abnormal clusters in the data;

[0015] The visual features and time series features are reconstructed through an autoencoder model, anomaly detection is performed based on the reconstruction error, and a reconstruction error result is generated. Among them, data points with high errors are marked as potential anomalies.

[0016] Preferably, the visual anomaly detection result, the time series anomaly detection result, the abnormal clustering result, and the reconstruction error result are used as inputs, and a decision fusion algorithm is applied for processing. The weights of the output results of each model are adjusted according to the corresponding historical performance, and an anomaly detection report is generated. The anomaly detection report includes: clearly identifying and explaining all detected anomaly points, and potential anomaly points obtained based on long short-term memory networks and data analysis.

[0017] Preferably, the anomaly detection result in the anomaly detection report is calculated through the following weighted sum:

[0018]

[0019] where ; represents the visual anomaly detection result; represents the time series anomaly detection result; represents the abnormal clustering result; represents the reconstruction error result; 、 、 、 respectively represent the corresponding weights.

[0020] Preferably, based on the anomaly detection report, the relevant process parameters of the electroplating production line are dynamically adjusted, and the adjusted parameters are fed back to the control system. After the operating state of the electroplating production line is updated in real time, the operating state of the electroplating production line is continuously monitored, the effect of the adjustment is evaluated, and according to the monitoring results and the continuously collected original data, the process parameters and adjustment strategies are continuously optimized.

[0021] A real-time monitoring system for the electroplating production process, comprising:

[0022] A data acquisition module configured to acquire the original data of the electroplating production line, where the original data includes temperature, pressure, current, and video / images;

[0023] A data processing module, configured to denoise, standardize the collected raw data, and extract visual features, sensor statistics, and time series features to generate a comprehensive feature dataset;

[0024] An analysis module, including: a visual processing sub-module that processes visual features using a convolutional neural network; a sequence analysis sub-module that analyzes time series features using a long short-term memory network; and an anomaly detection sub-module that applies unsupervised learning models and autoencoder models for anomaly clustering and data reconstruction;

[0025] A decision fusion module, configured to integrate the output results of each model of the analysis module using a decision fusion algorithm to generate an anomaly detection report, clarify the location and type of anomaly points, and identify all potential anomaly points;

[0026] An adjustment control module that dynamically adjusts the production line process parameters based on the anomaly detection report, monitors the adjustment effect in real time, and continuously optimizes the adjustment strategy according to the continuously collected raw data.

[0027] A storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps of real-time monitoring of the electroplating production process.

[0028] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0029] By combining high-resolution video / image data with traditional sensor data, the present invention realizes multi-dimensional anomaly signal capture, significantly enhances the monitoring ability of the system and the accuracy of fault diagnosis; advanced feature engineering technology can extract more valuable information from complex data, making the monitoring more comprehensive and detailed;

[0030] By applying various machine learning technologies including deep learning (CNN), time series analysis (LSTM), unsupervised learning (k-means, DBSCAN), and autoencoders, the present invention can perform complex anomaly detection; this multi-algorithm fusion makes anomaly detection more comprehensive and in-depth, improving the prediction accuracy and response speed of the monitoring system;

[0031] By adopting advanced decision fusion algorithms (such as weighted voting, stacking generalization), the present invention integrates the outputs of multiple models to provide a comprehensive anomaly detection report; this decision fusion technology improves the reliability of diagnosis, effectively reduces the possibility of misjudgment, and optimizes production decision support;

[0032] The present invention implements a dynamic feedback mechanism, automatically adjusts the production line parameters according to the anomaly detection results, and continuously optimizes these adjustments; this adaptive adjustment mechanism makes the production process more optimized, helps to solve problems immediately, prevents the deterioration of abnormal situations, and improves production efficiency and product quality at the same time. Brief Description of the Drawings

[0033] Figure 1 is a flowchart of the method of the present invention;

[0034] Figure 2 is a schematic diagram of the data processing flow of the present invention;

[0035] Figure 3 is a block diagram of the structure of the system of the present invention. Detailed implementation manners

[0036] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, many specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. However, it is obvious that one or more embodiments can be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts of the present disclosure.

[0037] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0038] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0039] In cases where expressions such as "at least one of A, B, and C, etc." are used, generally, it should be interpreted according to the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include, but not be limited to, a system having only A, only B, only C, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.). In cases where expressions such as "at least one of A, B, or C, etc." are used, generally, it should be interpreted according to the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include, but not be limited to, a system having only A, only B, only C, having A and B, having A and C, having B and C, and / or having A, B, and C, etc.).

[0040] Some block diagrams and / or flowcharts are shown in the accompanying drawings. It should be understood that some blocks in the block diagrams and / or flowcharts, or combinations thereof, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when executed by the processor, these instructions can create a device for implementing the functions / operations illustrated in these block diagrams and / or flowcharts. The technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). Additionally, the technology of the present disclosure can take the form of a computer program product on a computer-readable storage medium storing instructions, which can be used by or in conjunction with an instruction execution system.

[0041] As Figure 1 - Figure 2 shown, a real-time monitoring method for an electroplating production process includes the following steps:

[0042] Collect the original data of the electroplating production line, where the original data includes temperature, pressure, current, and video / images. Denoise and standardize the original data, and extract visual features, sensor statistics, and time series features to generate a comprehensive feature dataset;

[0043] During the electroplating production process, key parameters such as temperature, pressure, and current directly affect the quality and uniformity of the electroplating layer. In addition, video / image data can provide intuitive information about the surface state of the electroplating. For example, the stability of the current is directly related to the thickness and quality of the electroplating layer, while the uniformity of the temperature affects the rate and completion of the chemical reaction. Pressure changes can reflect the stability of the liquid flow in the production line, which is crucial for ensuring uniform coverage of the electroplating solution.

[0044] The environment of the electroplating production line is complex, and the original data often contains noise, such as electrical noise, mechanical vibration, etc. These noises will affect the quality of the data and the accuracy of subsequent analysis. Data denoising is achieved through filters (such as low-pass filters) or statistical methods (such as moving average method) to remove these non-productive fluctuations, thereby obtaining more accurate data readings. Standardization processing (such as z-score standardization or min-max normalization) ensures the comparability of data from different devices and sensors, eliminates the influence of dimensions and differences between devices, and provides a consistent baseline for data analysis.

[0045] Utilize image processing techniques, such as edge detection, texture analysis, color analysis, etc., to extract key information from the video / images. For example, using an edge detection algorithm (such as the Sobel algorithm) can identify cracks or defects on the surface of the electroplating layer, and texture analysis can evaluate the uniformity of the electroplating layer.

[0046] For sensor data, statistical measures (such as mean, variance, extreme values, etc.) and time series features (such as autocorrelation, trend component, periodic component, etc.) are extracted. These features help identify regular changes and potential abnormal patterns in the production process. For example, by analyzing the time series features of current data, it is possible to predict potential current fluctuations during the electroplating process, which are potential factors leading to a decline in the quality of the electroplated layer.

[0047] Through the above data collection and feature extraction steps, the electroplating production line can monitor key production parameters in real time, promptly detect abnormal situations during the production process, and thus make rapid adjustments or interventions. The comprehensive feature dataset provides a comprehensive view of the production process, enabling production managers to make more accurate decisions based on the comprehensive data.

[0048] For example, if time series analysis shows a sudden increase in current, combined with abnormalities detected on the surface of the electroplated layer by the vision system, the system can immediately prompt the production line to adjust relevant parameters or check for possible equipment failures, thereby preventing the spread of production defects. Through this real-time monitoring and rapid response mechanism, not only the efficiency and product quality of electroplating production are improved, but also resource waste and production costs are significantly reduced.

[0049] Preferably, image processing algorithms are used to extract and obtain visual features, and sensor statistics and time series features are obtained by calculating sensor data to generate a comprehensive feature dataset.

[0050] In an electroplating production line, image processing technology can be used to visually monitor and evaluate the quality of the electroplated layer. These algorithms mainly include:

[0051] Edge detection: Identifying the edges on the surface of the electroplated layer helps detect surface defects such as cracks or peeling. Commonly used edge detection algorithms include Canny, Sobel, and Prewitt, etc. These algorithms depict the contours of objects by strengthening the boundaries of brightness changes in the image.

[0052] Texture analysis: Evaluating the surface uniformity and roughness of the electroplated layer. By analyzing the texture features of the image, such as contrast, homogeneity, or texture directionality, intuitive information about whether the electroplated layer is evenly laid can be obtained.

[0053] Color analysis: By monitoring color changes, it is possible to detect changes in the composition of the electroplating solution or the oxidation of the electroplated layer. Color analysis can be achieved through color space transformation and statistical color distribution.

[0054] The extraction of these visual features can not only display the appearance status of the electroplated layer in real time, but also provide key data support for anomaly detection.

[0055] Sensor data analysis involves the following main steps:

[0056] (1)Statistic calculation: including mean, variance, maximum value, minimum value, etc. These statistical indicators help monitor the stability and variability of process parameters.

[0057] (2)Time series analysis: By calculating methods such as autocorrelation function and fast Fourier transform, the periodic changes and trends of parameters such as current and temperature over time can be analyzed. This is crucial for early identification of equipment performance degradation or process deviation.

[0058] For example, by analyzing the time series data of current, the current fluctuations caused by equipment aging or improper maintenance can be predicted and identified. Such fluctuations may lead to uneven thickness of the electroplated layer.

[0059] The present invention enables any deviation in the production process to be quickly detected and corrected through real-time image and data monitoring, thereby ensuring the uniformity and consistency of the electroplated layer; through continuous monitoring, the operating parameters of the production line can be adjusted according to real-time data, reducing material waste and improving the operating efficiency of the production line; time series analysis helps predict equipment failures and performance degradation, achieve preventive maintenance, and reduce unexpected downtime.

[0060] The generation of the comprehensive feature dataset is a bridge connecting data acquisition and decision execution. Through the comprehensive analysis of these data, managers can obtain a comprehensive production line status report, and then make more accurate operation decisions. This data-driven management mode is the core part of the automation and intelligence of modern electroplating production.

[0061] Using a convolutional neural network to process visual features, a long short-term memory network to analyze time series features, and at the same time applying unsupervised learning models and autoencoder models for anomaly clustering and data reconstruction, and respectively performing independent analysis on the comprehensive feature dataset and outputting the corresponding results;

[0062] A convolutional neural network (CNN) is a powerful image processing tool suitable for identifying patterns and features in images, such as defects, cracks, or uneven coverage of the electroplated layer. CNN automatically learns the low-level to high-level features of the image through multiple layers of filters without manual feature definition. These filters can capture the subtle differences on the surface of the electroplated layer. For example: identifying the discontinuity at the edge of the electroplated layer, which may indicate cracks or peeling; analyzing the texture consistency on the surface of the electroplated layer, and irregular texture may indicate uneven coverage or abnormal composition.

[0063] Long Short-Term Memory Network (LSTM) is a neural network particularly suitable for time series data analysis, capable of learning long-term dependencies in data. During the electroplating process, fluctuations in parameters such as current, temperature, and pressure over time can be subjected to pattern recognition and trend prediction by LSTM. For example, if the current suddenly changes beyond the normal fluctuation range, LSTM can identify this abnormal pattern, indicating potential equipment problems or process errors.

[0064] Unsupervised learning, such as K-means or DBSCAN clustering algorithms, is used to identify unlabeled groups or abnormal patterns in a dataset. During the electroplating process, these techniques can identify data clusters that do not conform to the normal production pattern. For example, clustering of abnormal operating conditions, clustering temperature and current data points, may reveal that specific clusters are associated with production defects.

[0065] Autoencoders are neural networks used for data dimensionality reduction and feature learning, learning a compressed representation of the data by attempting to reconstruct its input data. In electroplating production monitoring, autoencoders can be used to detect anomalies: data under normal operating conditions should be accurately reconstructed by the autoencoder. A high reconstruction error indicates a significant difference between the input data and normal data, possibly indicating an abnormal operation.

[0066] Through CNN and image analysis techniques, the present invention can detect and locate defects in the electroplating layer in real time, reducing the defective product rate; by using the analysis of time series data by LSTM, it can predict equipment failures and abnormal production parameters and make adjustments or maintenance in advance; potential anomalies and operation deviations identified through clustering and autoencoders can help optimize production parameters, improve production efficiency and product quality.

[0067] Preferably, an abnormal pattern in the image is identified through a convolutional neural network. The abnormal pattern includes unusual color changes, abnormal shapes, or textures, obtaining a visual anomaly detection result and marking all visually abnormal points;

[0068] By analyzing patterns and trends in time series features through a long short-term memory network, long-term patterns and trends in the data are identified, predicting future abnormal behaviors, obtaining a time series anomaly detection result for displaying any potential anomalies in the sensor data.

[0069] The application of convolutional neural networks (CNNs) in the field of image processing mainly utilizes their ability to automatically extract and learn the key features of images, which is crucial for real-time monitoring of visual anomalies in the electroplating production process. CNNs process image data through convolutional layers, pooling layers, and fully connected layers, extracting image features from simple to complex layer by layer. In electroplating monitoring, a CNN can be trained to identify specific anomaly patterns, such as color changes, irregular shapes, or abnormal textures. For example, in electroplating production, if unusual blue spots appear on the surface of a certain part of the product (possibly due to incomplete chemical reactions), the CNN can identify this anomaly by analyzing the color distribution in the image.

[0070] The detailed application of convolutional neural networks (CNNs) to electroplating anomaly monitoring includes:

[0071] The training of a CNN requires a large amount of labeled image data. In electroplating monitoring, these images should include surface images of electroplated layers in various normal and abnormal states, such as different degrees of oxidation, blistering, cracking, etc.

[0072] The image data needs to be labeled by experienced operators or through known fault records. The labeling process is to identify the anomaly patterns (such as color changes, abnormal shapes or textures) in the images and classify them.

[0073] The convolutional layer selects convolutional kernels of different sizes to capture features from fine textures to large-area shape changes. The ReLU (Rectified Linear Unit) activation function is usually adopted because of its high efficiency in non-linear mapping and its help in avoiding the vanishing gradient problem. The max-pooling layer is used to reduce the feature dimension while retaining important feature information.

[0074] The backpropagation algorithm and gradient descent (or its variants such as the Adam optimizer) are used for weight updates to minimize the loss function. Commonly used loss functions include cross-entropy loss, especially in classification tasks. To prevent overfitting, a dropout layer may be added or L2 regularization techniques may be used.

[0075] Long short-term memory networks (LSTMs) are specifically designed to process and predict key events in time series data. They control the inflow, retention, and outflow of information through gating mechanisms (input gate, forget gate, and output gate). LSTMs can learn long-term dependencies in time series, which makes them particularly suitable for identifying and predicting parameter trends and patterns in the electroplating process. It can learn from historical data what kind of behavior may lead to production anomalies. Example: In electroplating production, if the temperature and current data show an increasing periodic fluctuation, this may indicate an impending equipment failure. The LSTM can identify this pattern and predict possible abnormal behaviors in the future, thus enabling early intervention.

[0076] Construction of a prediction model for Long Short-Term Memory (LSTM):

[0077] Select time series data that reflect the key parameters of the electroplating process, such as current, temperature, pressure, etc. Determine the time window size of the input data, which will affect the model's ability to learn long-term and short-term patterns in the time series.

[0078] Use methods such as differencing and logarithmic transformation to handle non-stationarity, making the data more suitable for processing by the LSTM model. Design a multi-layer LSTM structure to capture more complex time series relationships and dynamic changes.

[0079] Use the first part of the time series for training and the latter part for testing and validation to ensure the performance of the model on unseen data. Find the optimal model settings by adjusting hyperparameters such as the learning rate, batch size, and number of iterations.

[0080] During the electroplating process, visual anomalies such as unusual color changes, shape or texture abnormalities are key indicators that the monitoring system must accurately identify, because these anomalies often indicate potential production problems or equipment failures. The following is a detailed description of these anomaly types, their potential impacts, and causes:

[0081] Unusual color changes, abnormal changes in the color of the electroplated layer may appear as abnormal dullness, bright spots, or color deviations. Color changes usually affect the appearance quality and corrosion resistance of the product, and may lead to the product being rejected or returned by customers.

[0082] Potential causes include:

[0083] (1) Chemical imbalance: Abnormal metal ion concentration or pH value deviation in the electroplating solution.

[0084] (2) Temperature fluctuations: Unstable temperature of the electroplating solution will affect the deposition rate of metal ions and the uniformity of the deposited layer.

[0085] (3) Impurity contamination: Accumulation of impurities in the electroplating solution, such as grease, dust, etc., affects the color and texture of the final product.

[0086] Shape abnormalities, abnormal shapes of the electroplated layer include uneven deposition, depressions, or blisters. Shape abnormalities will reduce the mechanical strength and durability of the product, and increase the difficulty of subsequent processing.

[0087] Potential causes include:

[0088] (1) Uneven current distribution: Uneven current density during electroplating can lead to excessive or insufficient local deposition.

[0089] (2) Equipment failure: For example, damage to internal components of the electroplating tank or incorrect electrode position.

[0090] Texture abnormality. The texture of the electroplated layer is usually abnormal, manifested as roughness, granularity or cracks. Texture abnormality not only affects the product appearance, but may also weaken its protective function and increase surface wear.

[0091] Potential causes include:

[0092] (1) Too fast or too slow electroplating speed: Abnormal changes in the deposition rate usually result in uneven texture.

[0093] (2) Imbalance in the electroplating solution composition: Lack or excess of specific chemical components.

[0094] Through high-precision cameras and image analysis software, the monitoring system can detect and report the above abnormal conditions in real time. Once an abnormality is detected, the relevant production parameter data is immediately analyzed, and the electroplating solution composition, temperature or current settings are adjusted to ensure that the problem is quickly resolved and the spread of quality problems is prevented.

[0095] Through the application of CNN in the present invention, real-time video monitoring on the electroplating production line can automatically identify and mark any visually abnormal points. This not only improves the speed and accuracy of abnormality detection, but also reduces the need for manual inspection. The production quality is thus significantly improved, while production delays and costs are reduced.

[0096] Using LSTM to analyze time series data can effectively predict and identify potential production abnormalities, such as impending equipment failures or abnormal changes in process parameters. This predictive ability makes the production process more reliable, reduces unexpected downtimes and production losses, and also optimizes maintenance and operation costs.

[0097] Generally speaking, the application of these advanced analysis techniques greatly enhances the monitoring ability and efficiency of the electroplating production process, ensuring the consistency of product quality and the smooth operation of the production line. Through real-time and accurate abnormality detection, production managers can better grasp the production situation, timely adjust production strategies, and maximize production efficiency and product quality.

[0098] Preferably, unsupervised learning algorithms are applied to cluster visual features and time series features, group data points according to the similarity of data, generate abnormal clustering results, and the abnormal clustering results are used to identify abnormal clusters in the data;

[0099] Visual features and time series features are reconstructed through an autoencoder model, and abnormality detection is performed based on the reconstruction error to generate a reconstruction error result, where data points with high errors are marked as potential abnormalities.

[0100] Unsupervised learning algorithms such as K-means or DBSCAN are used for the analysis of unlabeled data, automatically grouping them by measuring the similarity between data points. During the electroplating process, these algorithms can identify abnormal patterns that cluster together based on visual features and time series data. These patterns generally do not conform to the statistical distribution of normal production data.

[0101] It is usually based on Euclidean distance or cosine similarity. For example, when dealing with time series data, the distance between two time series may be calculated to determine whether their production behaviors are similar. These results help identify atypical behaviors in the production process, such as a set of data points showing temperature and current fluctuations different from normal operations.

[0102] Abnormal clustering enables the production monitoring system to automatically identify potential problems without the need to preset specific types of anomalies. For example, if a batch of products shows distinctive texture features, clustering analysis can quickly identify it as an abnormal batch that requires further inspection.

[0103] In electroplating production monitoring, clustering algorithms are used to identify potential anomalies and regular behaviors during operations, helping to optimize the production process and identify quality problems in advance. Different clustering algorithms are suitable for different types of data characteristics:

[0104] K-means clustering: Simple and computationally efficient, suitable for large-scale data processing. K-means forms clusters by assigning data points to the nearest centers and is applicable to datasets with an approximately spherical shape. Selecting the appropriate number of clusters usually requires prior knowledge or the use of methods such as the elbow method for determination.

[0105] DBSCAN (Density-Based Spatial Clustering of Applications with Noise): Does not require presetting the number of clusters, can identify clusters of any shape, and has good robustness to noise and outliers. The core parameters include the neighborhood size (ε) and the minimum number of points (MinPts), which need to be adjusted according to the density and distribution characteristics of the data.

[0106] Hierarchical clustering: Does not require pre-specifying the number of clusters and can visually display the similarity relationships between data points through a dendrogram. It mainly involves clustering merging strategies, such as methods of choosing the minimum distance, maximum distance, or average distance, etc.

[0107] The application of abnormal clustering results in electroplating production is multi-faceted and directly affects operation decisions and production process adjustments. The following are some specific application examples:

[0108] Adjust the electroplating solution composition: If cluster analysis reveals that a certain set of data is associated with changes in some chemical components in the electroplating solution, this may indicate that the chemical balance of the electroplating solution has been disrupted. Based on this finding, chemical engineers can decide to increase or decrease the proportion of specific chemical components. For example, increasing the copper ion concentration to improve the deposition efficiency. The adjusted electroplating solution formula is input into the production line through an automatic control system, and the monitoring system continues to track the adjusted production data to ensure that the adjustment achieves the expected effect.

[0109] Optimize process parameters: From the clustering results, if it is found that a specific combination of temperature and current frequently leads to production problems, engineers can decide to adjust these parameters to a more optimal level. For example, reducing the current density to reduce the occurrence of surface defects. The adjusted process parameters are transmitted to the control system in real time, and the impact of the adjustment is monitored. The improvement of production quality is confirmed through data comparison.

[0110] An autoencoder is a neural network model used for data dimensionality reduction and feature extraction, which reconstructs its own input through training data. In electroplating monitoring, the autoencoder learns the representation of visual and time series features under normal operating conditions.

[0111] The autoencoder is trained by minimizing the reconstruction error between the input and output. Under normal conditions, the model can accurately reconstruct the input data. In the detection stage, new production data is input into the trained model. If the data comes from a normal production process, the expected reconstruction error will be very low. A high reconstruction error indicates that the input data contains features that the model has not learned, which is usually related to production anomalies.

[0112] The application of the autoencoder improves the ability to detect subtle anomalies. For example, if the current data during a certain period shows a slight difference from the pattern during training due to equipment failure, even if this difference is very small, the reconstruction error will increase, thus triggering an anomaly alarm.

[0113] The autoencoder is an effective tool for data dimensionality reduction and feature learning, and is widely used in anomaly detection. By designing different types of autoencoders, the ability to capture the features of electroplating monitoring data can be improved:

[0114] Sparse autoencoder: By introducing a sparsity constraint in the hidden layer, it can learn more useful data representations, which helps to capture more important features. Add a sparsity penalty term (such as KL divergence) to the loss function to force the activation values of the hidden layer to be close to zero in most cases.

[0115] Denoising autoencoder: By training the model to reconstruct the input data corrupted by noise, the denoising autoencoder improves the generalization ability of the model for unseen anomalies. Add noise (such as Gaussian noise) to the input data, and train the model to learn to ignore the noise and reconstruct the original undisturbed data.

[0116] Network architecture and parameter optimization: Select an appropriate network depth and width based on the complexity of the data. Increasing the number of layers and nodes can enhance the model complexity but may also lead to overfitting. Commonly used activation functions are ReLU or Sigmoid functions, and the most suitable activation function is selected according to specific data characteristics. Use a suitable optimization algorithm (such as Adam) for parameter optimization to ensure the convergence speed and model performance.

[0117] The reconstruction error is a measure of the difference between the output of the autoencoder and the input data. A significant reconstruction error usually indicates the presence of features or patterns in the input data that the model fails to predict, which may be due to: the data features are significantly different from the normal patterns learned by the model during training, the device performance degrades, or unexpected changes occur during the production process.

[0118] The key to setting the threshold lies in balancing the sensitivity and specificity of detection: a too low threshold may lead to a high false alarm rate, frequently marking normal variations as anomalies; a too high threshold may lead to missed detections, failing to capture real anomalies.

[0119] Threshold determination method: Based on the reconstruction error distribution on the training dataset, select a quantile with statistical significance (such as 95%) as the threshold; use cross - validation techniques to test different threshold settings on an independent validation dataset to find the optimal balance point.

[0120] Combining these two methods, the monitoring system of the electroplating production line can more comprehensively monitor and analyze production data. Imagine a scenario where the system identifies a group of production batches with similar abnormal temperature readings through cluster analysis. At the same time, the autoencoder detects a high reconstruction error associated with these batches. This double verification not only quickly and accurately points out the specific location of the problem but also increases the accuracy of anomaly detection, allowing operators to timely adjust production parameters or perform equipment maintenance to prevent production quality problems.

[0121] Integrate the output results of each model using a decision fusion algorithm to generate an anomaly detection report, clarify the location and type of anomaly points, and identify all potential anomaly points;

[0122] In the real - time monitoring system of electroplating production, the decision fusion algorithm is a key component used to integrate the output results from multiple detection models to generate a comprehensive anomaly detection report.

[0123] Each detection model (such as a vision detection system, a time series analysis model, an autoencoder, etc.) independently identifies potential anomalies in the data. The decision fusion algorithm first collects the outputs of these models, and each model provides evidence of the existence of anomalies. For example, the vision system may identify a color anomaly on the coating surface, while the time series model may simultaneously indicate abnormal fluctuations in temperature or chemical composition at the corresponding time point.

[0124] The decision fusion algorithm evaluates the consistency and reliability of the outputs of each model, and uses methods such as weighted voting, model stacking, or Bayesian methods to integrate the insights of each model and determine the final anomaly determination. For example, if multiple models point to similar problems in the same production batch, this information will be weighted and combined to enhance the credibility of the anomaly report.

[0125] The integrated data is used to generate a detailed anomaly detection report. This report not only indicates the location and type of the anomaly, but also predicts the potential development of the anomaly, provides possible causes, and recommended response measures. For example, the report may indicate that the current in a certain electroplating tank has increased abnormally, speculate that it is due to an imbalance in the electrolyte composition, and recommend checking the relevant feeding system.

[0126] By integrating the outputs of multiple models, the system can provide more comprehensive and accurate anomaly detection results than a single model. Decision fusion reduces the risk of false positives and false negatives by analyzing data from different sources and types.

[0127] The comprehensive anomaly detection report enables production managers to obtain key information in a timely manner, make adjustments or take preventive measures quickly, optimize the production process, and reduce potential losses.

[0128] The present invention provides a data-driven decision support framework to help management understand the complex interactions in the production process, thereby better managing production quality and process efficiency.

[0129] The application of the decision fusion algorithm in the electroplating production monitoring system greatly improves the comprehensiveness and accuracy of anomaly detection. By effectively combining the advantages of multiple detection models, this system can provide more reliable monitoring results, support more efficient production management and decision-making. In addition, through real-time data analysis and feedback, the production process becomes more controllable and optimized, ensuring product quality and production efficiency.

[0130] Preferably, the vision anomaly detection results, time series anomaly detection results, anomaly clustering results, and reconstruction error results are used as inputs, and the decision fusion algorithm is applied for processing. The weights of the output results of each model are adjusted according to the corresponding historical performance to generate an anomaly detection report. The anomaly detection report includes: clearly identifying and explaining all detected anomaly points, and potential anomaly points obtained based on long short-term memory networks and data analysis.

[0131] In the electroplating production monitoring system, the fusion processing of multi-source data is a key step to ensure that the advantages of various monitoring technologies are fully utilized. The core of the decision fusion algorithm lies in integrating the detection results from different sources and natures to form a comprehensive and reliable anomaly detection output.

[0132] The multi-source inputs include visual anomaly detection results, time series anomaly detection results, anomaly clustering results, and reconstruction error results of autoencoders. Decision fusion algorithms such as the weighted average method, majority voting method, or more complex machine learning models (such as random forests or neural networks) are used to integrate these data sources. Based on the performance of each model in historical data, weight assignments are made to their output results. This dynamic weight adjustment ensures that the models are appropriately considered according to their reliability and accuracy when evaluating new data.

[0133] The generated anomaly detection report not only lists all detected anomaly points but also explains the possible causes and natures of these anomalies, while pointing out potential anomaly points predicted based on pattern recognition and data analysis. Suppose a camera on the electroplating line captures a change in the color of a certain part of the product, and at the same time, the time series data analyzed by the LSTM model also shows an abnormal fluctuation in current at the same time point. These two anomalies may point to different manifestations of the same problem. Through the decision fusion algorithm, these two signals are comprehensively analyzed, strengthening the certainty of the problem cause in the report.

[0134] By integrating the results of multiple detection technologies, the system can reduce the occurrence of false positives and false negatives and improve the accuracy of anomaly detection. For example, relying solely on visual detection may misjudge the shadow of the equipment as a product defect, while the synchronous performance of time series analysis may help confirm that this is not a real product problem.

[0135] Combined with time series analysis and historical performance data, the system can not only detect current anomalies but also predict potential problems in the future, providing an earlier opportunity for intervention for the operation team. For example, continuously monitored slight temperature increases may indicate an impending more serious equipment failure.

[0136] In the decision fusion process, it is crucial to correctly quantify the historical performance of each model and adjust their weights in the fusion decision accordingly.

[0137] The metrics used include:

[0138] Accuracy: Measures the proportion of correct predictions in the model output and is applicable to evaluating the overall effectiveness.

[0139] Recall: Measures the proportion of relevant instances identified by the model among all relevant instances. The key lies in identifying all possible outliers.

[0140] Precision: Measures the proportion of instances identified as anomalies that are actually anomalies. It is important for reducing false positives.

[0141] F1 Score: The harmonic mean of precision and recall, which is a balance between the two and is particularly suitable for scenarios with imbalanced class data.

[0142] Weight Calculation: Dynamically adjusts the weight of each model in decision fusion based on the metrics (such as F1 score) it shows in historical data. For example, a model with better performance (higher F1 score) will be given a higher weight in the fusion decision. A time decay factor can be introduced to adjust the weight to reflect that the model's recent performance is more important than its early performance.

[0143] The anomaly detection report provides comprehensive analysis results, including the specific location, type of anomalies, and recommended response measures, enabling production managers to make quick and effective decisions. For example, the report may recommend adjusting the chemical ratio of a certain electroplating bath or suspending production for necessary equipment maintenance.

[0144] Long Short-Term Memory Networks (LSTM) and data analysis tools play a crucial role in predicting potential outliers. These tools not only help identify current anomalies but also predict future possible risks.

[0145] LSTM can handle and predict anomalies based on time series data, such as sudden changes in current or chemical concentration. LSTM predicts future trends and potential anomalies by learning the long-term dependencies of past data. Integrate the prediction results of LSTM with other data analysis results (such as predictions from statistical models or other machine learning techniques) to obtain a comprehensive view of potential outliers.

[0146] Using the predicted potential outliers, production managers can conduct risk assessments, adjust production plans, or pre-adjust machine settings to avoid potential production interruptions. The prediction results are integrated into a decision support system, providing real-time data support and recommendations to help operators make quick and accurate response decisions.

[0147] The application of decision fusion algorithms has greatly improved the monitoring effect in the electroplating production process. By integrating the advantages of multiple detection technologies, it has achieved more accurate and comprehensive anomaly identification and prediction. This advanced data analysis method provides strong support for the production process, ensuring product quality while reducing production risks.

[0148] Through the application of prediction tools and methods, the monitoring system in the electroplating production process can not only respond to current abnormal situations but also prevent potential problems in the future. This forward-looking monitoring strategy greatly enhances the stability and efficiency of the production process, ultimately improving product quality and reducing production costs.

[0149] Preferably, the abnormal detection result in the abnormal detection report is calculated by the following weighted sum:

[0150]

[0151] where ; represents the visual abnormal detection result; represents the time series abnormal detection result; represents the abnormal clustering result; represents the reconstruction error result; 、 、 、 respectively represent the corresponding weights.

[0152] The setting of the weights is based on the past performance of each detection method, including their accuracy, recall rate, F1 score, etc. Methods with higher weights have a greater influence on the final abnormal detection score. The weights can be adjusted regularly according to new verification results to ensure that the weights of each method reflect their latest performance and reliability.

[0153] Through the calculation method of the weighted sum, the advantages of various detection technologies can be integrated to make up for the possible deficiencies of a single technology. For example, visual detection may be sensitive to certain surface defects, while time series analysis may be more capable of capturing abnormalities caused by internal parameter changes. Adjusting the weights according to real-time data enables the abnormal detection system to dynamically adapt to changes in the production environment and promptly identify new or uncommon abnormal patterns. The system can adjust the weights of each model according to the characteristics of a specific production line or product type, making the abnormal detection more targeted and effective.

[0154] Suppose that in an electroplating production line, a potential quality problem is discovered through the real-time monitoring system:

[0155] The visual system detects unusual color changes on the product surface;

[0156] Time series analysis shows that at the same time, the current and temperature parameters fluctuate abnormally;

[0157] Abnormal clustering and reconstruction error analysis also mark this time point as a potential abnormality.

[0158] In this case, the system will use the weighted summation method to synthesize various detection results and calculate a total anomaly score based on the set weights. If this score is higher than the preset threshold, the system will generate a detailed anomaly report, identifying the specific location and type of the anomaly, as well as providing possible causes and recommended response measures.

[0159] The weighted sum method provides an effective way to integrate multiple data sources in the real-time monitoring of the electroplating production process, improving the accuracy of anomaly detection and the responsiveness of operations. Through this method, the production process can be more stable, effectively reducing production defects and enhancing product quality.

[0160] Based on the anomaly detection report, dynamically adjust the production line process parameters, monitor the adjustment effect in real time, and continuously optimize the adjustment strategy according to the continuously collected raw data.

[0161] Implementing a dynamic adjustment strategy based on the anomaly detection report on an electroplating production line is a complex but crucial process. This strategy relies on the capabilities of a real-time monitoring system to not only respond promptly to identified anomalies but also predict potential production deviations, thereby optimizing the entire production process.

[0162] By analyzing the data in the anomaly detection report, the system can identify the specific process parameters that cause the anomaly, such as current intensity, electroplating solution temperature, or chemical composition ratio. Based on this information, the automatic control system adjusts the corresponding parameters to correct or optimize the production process. For example: If the detection report indicates that product defects are caused by excessive current, the system will automatically reduce the current intensity in order to restore product quality.

[0163] By analyzing the data in the anomaly detection report, the system can identify the specific process parameters that cause the anomaly, such as current intensity, electroplating solution temperature, or chemical composition ratio. Based on this information, the automatic control system adjusts the corresponding parameters to correct or optimize the production process. For example: If the detection report indicates that product defects are caused by excessive current, the system will automatically reduce the current intensity in order to restore product quality.

[0164] The system continuously collects and analyzes new production data and uses this information to further fine-tune and optimize the process parameters. Through machine learning and data analysis tools, the system can learn which adjustment measures are most effective, thereby continuously improving the production process. For example: The system may find that under certain specific conditions, fine-tuning the pH value of the electroplating solution is more effective than changing the temperature, so it adjusts its algorithm to preferentially adjust the pH value in future similar situations.

[0165] Through the dynamic adjustment strategy, production interruptions and scrap rates can be reduced, resource utilization can be optimized, and overall production efficiency can be improved. At the same time, maintain product quality at the best level and meet strict quality standards.

[0166] Real-time monitoring and dynamic adjustment reduce the need for manual intervention and lower production costs. Quick response to problems in production reduces potential losses and costs.

[0167] Predicting potential anomalies and making adjustments in advance can serve as a preventive maintenance strategy, reducing the risk of severe failures and enhancing the stability and safety of the production line.

[0168] Implementing a dynamic adjustment strategy based on anomaly detection reports in electroplating production can not only solve production problems in real time but also predict and prevent potential problems from developing, thus optimizing the entire production process. This strategy, by comprehensively using the latest monitoring data, machine learning techniques, and automated control systems, greatly improves production efficiency, reduces operating costs, ensures product quality, and demonstrates the potential and value of intelligent manufacturing in modern manufacturing.

[0169] Preferably, based on the anomaly detection report, dynamically adjust the relevant process parameters of the electroplating production line, feedback the adjusted parameters to the control system, continuously monitor the operating status of the electroplating production line after updating the operating status in real time, evaluate the effect of the adjustment, and continuously optimize the process parameters and adjustment strategy according to the monitoring results and continuously collected original data.

[0170] By collecting the operating data of the electroplating production line in real time (such as temperature, current, chemical composition concentration, etc.), use this data to guide the adjustment of process parameters. This method relies on high-quality sensors and data acquisition systems to ensure the accuracy and timeliness of the data.

[0171] Based on the suggestions in the anomaly detection report or the deviations automatically detected, the control system immediately adjusts the process parameters, such as adjusting the current intensity or chemical mixing ratio, to correct or optimize the production process. For example: If the monitoring system finds that the coating thickness of a certain batch of products is uneven, the system will automatically adjust the current setting to ensure the uniformity of the coating and immediately feedback these adjustments to the control system of the production line.

[0172] After adjustment, the system continuously monitors the operating status of the production line and the new output data, and uses algorithms to evaluate the effect of the adjustment. This includes the immediate evaluation of production quality and the long-term performance trend analysis. For example: After adjusting the chemical composition, the system will monitor the quality parameters of the products, such as adhesion and corrosion resistance, in the following hours or even days to ensure that the adjustment has achieved the expected improvement effect.

[0173] Dynamic adjustment ensures rapid response to various changes during the production process, minimizing the generation of defective products, thus enhancing overall production efficiency and reducing costs; by meticulously controlling process parameters, such as precisely adjusting chemical formulations and electroplating conditions, material waste can be reduced, and the use of energy and raw materials can be optimized; precisely controlling the use of chemicals not only reduces costs but also decreases potential environmental pollution, helping enterprises comply with increasingly stringent environmental regulations.

[0174] The dynamic adjustment of process parameters is not only an application of real-time monitoring technology but also a key strategy for improving automation levels and optimizing production efficiency in modern electroplating production processes. By integrating high-precision monitoring devices and advanced data analysis techniques, electroplating enterprises can ensure that each batch of products meets high-standard quality requirements while enhancing production flexibility and economic benefits. The successful implementation of this method relies on continuous technological innovation and process improvement, making the production process more intelligent and controllable.

[0175] As Figure 3 shown, a real-time monitoring system for an electroplating production process includes:

[0176] A data acquisition module configured to collect raw data from the electroplating production line, where the raw data includes temperature, pressure, current, and video / images; this module is equipped with a variety of sensors, including temperature, pressure, and current sensors, as well as high-resolution video / image acquisition devices. These devices continuously collect real-time data on the electroplating production line, providing the raw input for subsequent data analysis. During the electroplating process, the stability of temperature and current directly affects the uniformity and quality of the coating. Capturing the fluctuations of these parameters in real-time allows for immediate adjustment to avoid production defects.

[0177] A data processing module configured to denoise, standardize the collected raw data, and extract visual features, sensor statistics, and time series features to generate a comprehensive feature dataset; this module performs denoising and standardization processing on the collected raw data and then extracts key visual features, sensor statistics, and time series features through advanced algorithms to generate a comprehensive feature dataset. Visual data can be used to detect defects on the surface of electroplated parts, such as scratches or uneven coatings, while time series features help identify periodic problems or trends in the production process.

[0178] Analysis module, including: a visual processing sub-module that processes visual features using a convolutional neural network; a sequence analysis sub-module that analyzes time series features using a long short-term memory network; an anomaly detection sub-module that applies unsupervised learning models and autoencoder models for anomaly clustering and data reconstruction; including a visual processing sub-module and a sequence analysis sub-module, which respectively use a convolutional neural network (CNN) and a long short-term memory network (LSTM) to process and analyze the extracted features. In addition, the anomaly detection sub-module uses unsupervised learning and autoencoder models to perform anomaly clustering and reconstruction on the data, thereby identifying potential anomalies. CNN can identify minor changes in visual features, such as abnormal changes in the color of the plating; LSTM predicts possible future anomaly trends by analyzing time series data, such as the instability of current.

[0179] Decision fusion module, configured to integrate the output results of each model of the analysis module using a decision fusion algorithm to generate an anomaly detection report, clarify the location and type of the anomaly point, and identify all potential anomaly points; this module integrates the outputs from different analysis modules, uses a decision fusion algorithm (such as weighted voting or stacking generalization) to synthesize this data, and generates a detailed anomaly detection report that clarifies the specific location and type of the anomaly point. If both video analysis and sensor data analysis indicate a problem in a certain area, decision fusion integrates this information, improving the confidence and accuracy of the diagnosis.

[0180] Adjustment control module, based on the anomaly detection report, dynamically adjusts the process parameters of the production line, monitors the adjustment effect in real time, and continuously optimizes the adjustment strategy according to the continuously collected raw data. Based on the anomaly detection report, this module dynamically adjusts the process parameters on the production line, such as temperature, current, etc., and monitors the effect of these adjustments in real time, and optimizes the adjustment strategy according to the continuously collected data. If an abnormal increase in current is detected, the system automatically adjusts the power output to avoid product quality problems caused by overcurrent.

[0181] The system of the present invention provides a highly integrated solution to ensure the continuity and quality of electroplating production through real-time data monitoring and intelligent analysis. Through real-time response and adjustment, the system can minimize production interruptions, reduce the output of defective products, thereby reducing costs and increasing production efficiency. The real-time feedback and dynamic adjustment mechanism of the system makes the production process more flexible, able to quickly adapt to changes in production conditions, and maintain a high level of operating performance and product quality.

[0182] A storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps of real-time monitoring of the electroplating production process.

[0183] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0184] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can also be implemented. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0185] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0186] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0187] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0188] The memory includes non-permanent memory in the computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0189] A computer-readable medium includes permanent and non-permanent, removable and non-removable media and can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transitory medium that can be used to store information accessible by a computing device. As defined herein, a computer-readable medium does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0190] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising the element.

[0191] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A real-time monitoring method for an electroplating production process, characterized in that: The following steps are involved: Collecting raw data from the electroplating production line, including temperature, pressure, current, and video / image, denoising and standardizing the raw data, and extracting visual features, sensor statistics, and time series features to generate a comprehensive feature data set; Convolutional neural networks are used to process visual features, and long short-term memory networks are used to analyze time series features. Unsupervised learning models and autoencoder models are used to perform anomaly clustering and data reconstruction. The comprehensive feature data sets are used to perform independent analysis and output corresponding results. Convolutional neural networks are used to identify abnormal patterns in images. Abnormal patterns include unusual color changes, shape or texture abnormalities, obtain visual anomaly detection results, and identify all visual abnormal points; long short-term memory networks are used to analyze patterns and trends in time series features, identify long-term patterns and trends in data, predict abnormal behaviors in the future, and obtain time series anomaly detection results to display any potential anomalies in sensor data; Unsupervised learning algorithms are used to cluster visual features and time series features, and data points are grouped according to their similarities to generate abnormal clustering results, which are used to identify abnormal clusters in the data. Visual features and time series features are reconstructed through an autoencoder model, and anomaly detection is performed based on reconstruction errors to generate reconstruction error results, in which data points with high errors are marked as potential anomalies. The output results of each model are integrated using a decision fusion algorithm to generate an anomaly detection report, clarify the location and type of anomalies, and identify all potential anomalies; Based on the abnormal detection report, the production line process parameters are dynamically adjusted, the adjustment effect is monitored in real time, and the adjustment strategy is continuously optimized based on the continuously collected raw data.

2. The real-time monitoring method of the electroplating production process according to claim 1, characterized in that: Image processing algorithms are used to extract and obtain visual features, and sensor statistics and time series features are obtained by calculating sensor data to generate a comprehensive feature data set.

3. The real-time monitoring method of the electroplating production process according to claim 1, characterized in that: The visual anomaly detection results, time series anomaly detection results, anomaly clustering results and reconstruction error results are used as input and processed by the decision fusion algorithm. The weight of each model output result is adjusted according to the corresponding historical performance to generate an anomaly detection report. The anomaly detection report includes: clearly identifying and explaining all detected anomalies, as well as potential anomalies based on long short-term memory networks and data analysis.

4. The real-time monitoring method of the electroplating production process according to claim 1, characterized in that: The anomaly detection results in the anomaly detection report It is calculated by the following weighted sum: in, ; represents the visual anomaly detection result; Represents the time series anomaly detection results; Indicates abnormal clustering results; Represents the reconstruction error result; , , , Represent the corresponding weights respectively.

5. The real-time monitoring method of the electroplating production process according to claim 1, characterized in that: Based on the abnormal detection report, dynamically adjust the relevant process parameters of the electroplating production line, feed the adjusted parameters back to the control system, update the operating status of the electroplating production line in real time, continuously monitor the operating status of the electroplating production line, evaluate the effect of the adjustment, and continuously optimize the process parameters and adjustment strategies based on the monitoring results and continuously collected raw data.

6. A real-time monitoring system for electroplating production process, characterized in that: include: A data acquisition module, configured to collect raw data of the electroplating production line, wherein the raw data includes temperature, pressure, current and video / image; A data processing module configured to denoise and standardize the collected raw data, and extract visual features, sensor statistics, and time series features to generate a comprehensive feature data set; The analysis module includes: a visual processing submodule that uses a convolutional neural network to process visual features; a sequence analysis submodule that uses a long short-term memory network to analyze time series features; an anomaly detection submodule that applies an unsupervised learning model and an autoencoder model to perform anomaly clustering and data reconstruction, wherein the convolutional neural network is used to identify abnormal patterns in the image, and the abnormal patterns include: unusual color changes, shape or texture abnormalities, obtain visual anomaly detection results, and identify all visual abnormal points; the long short-term memory network is used to analyze patterns and trends in time series features, identify long-term patterns and trends in the data, predict abnormal behaviors in the future, and obtain time series anomaly detection results for displaying any potential anomalies in the sensor data; Unsupervised learning algorithms are used to cluster visual features and time series features, and data points are grouped according to their similarities to generate abnormal clustering results, which are used to identify abnormal clusters in the data. Visual features and time series features are reconstructed through an autoencoder model, and anomaly detection is performed based on reconstruction errors to generate reconstruction error results, in which data points with high errors are marked as potential anomalies. A decision fusion module is configured to integrate the output results of each model of the analysis module using a decision fusion algorithm to generate an anomaly detection report, clarify the location and type of anomalies, and identify all potential anomalies; The adjustment control module dynamically adjusts the production line process parameters based on the anomaly detection report, monitors the adjustment effect in real time, and continuously optimizes the adjustment strategy based on the continuously collected raw data.

7. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the real-time monitoring method of the electroplating production process described in any one of claims 1 to 5 are implemented.

Images