An automatic coating machine correction control system

Abstract

The present invention relates to an automated coating machine deviation correction control system. The system monitors and collects the position parameters and operating parameters of multiple position sensors of the coating machine in real time; obtains the position characteristic values ​​of the multiple position parameters to determine whether there is a deviation in the current position; when a position deviation exists, calculates the hash value of the operating parameter using a hash function and searches for the corresponding preset fault mode in a hash table; and employs a machine learning model to train the time mapping relationship between the position characteristic values, operating parameters, and fault modes. The system is used to predict the position deviation and fault mode for the next time period based on the current operating parameters and position characteristic values, and to perform deviation correction control. The present invention solves the problem of slow position deviation and fault response speed in existing technologies, while simultaneously improving the accuracy and efficiency of fault detection by utilizing a time series model and a random forest model.

Description

Technical Field

[0001] The invention belongs to the technical field of coating machine control and relates to an automatic coating machine deviation correction control system. Background Art

[0002] Amidst the booming development of modern industry, market competition is fierce across all sectors, consumer expectations for product quality are constantly rising, and industry standards are becoming increasingly stringent. Industries such as food packaging and electronics place extremely high demands on the precision and performance of coated products. Furthermore, with the trend towards large-scale production, coating machines must meet the demands of high-speed continuous production. However, high-speed operation can easily cause substrate deviation. Furthermore, with the rapid market changes and customers demanding shorter delivery cycles, coating machine deviation control must meet the dual demands of high quality and high efficiency.

[0003] Patent CN118419663A discloses a real-time tension control and correction system for coating machines. It includes a data acquisition unit, a simulation model building unit, a parameter adjustment unit, an operating correction unit, and a data replacement unit. The data acquisition unit is used to collect the coating machine's operating parameters and simultaneously obtain the material's position data on the coating machine's surface. By detecting whether the real-time operating parameters and position data match the fault simulation data, the system promptly stops operation and sends a signal for correction, reducing the likelihood of faults and improving system reliability and safety. The system can apply new tension correction parameters to the coating tension simulation model, updating all data to ensure that the simulation results are consistent with the actual situation. The system also simulates correction parameters based on the real-time operating parameters according to the coating tension simulation model, allowing for timely preparation of correction measures before a fault occurs.

[0004] However, although the system can detect whether real-time parameters match fault simulation data, faults occur quickly in actual production. There may be a certain delay from detecting deviations to stopping work and sending correction signals. For high-speed coating machines, this short delay may result in a large number of defective products. Summary of the Invention

[0005] This invention provides an automated coating machine deviation correction control system. By simplifying the logical judgment of position deviations, multiple position parameters are combined into a single, comprehensive characteristic value, reducing the complexity of multiple threshold judgments. A hash table is used for fast matching. The operating parameters under normal operating conditions are combined and stored using a hash function. Real-time deviation data is then searched in the table using the hash function, eliminating the time-consuming traditional linear search. A machine learning model learns the temporal mapping relationship between the coating machine's position characteristic values, operating parameters, and fault modes, enabling early fault prediction and addressing the slow fault response problem of existing technologies.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] The present application provides an automatic coating machine deviation correction control system, including a data acquisition module, a deviation judgment module, a fault identification module and a fault prediction module; the data acquisition module, the deviation judgment module, the fault identification module and the fault prediction module are communicatively connected, wherein:

[0008] The data acquisition module is used to monitor and collect position parameters and working parameters of multiple position sensors of the coating machine in real time;

[0009] The deviation judgment module is used to obtain position characteristic values ​​of multiple position parameters and judge whether there is a deviation in the current position;

[0010] The fault identification module is configured to calculate a hash value of the operating parameter by a hash function when there is a position deviation, and search the hash table for a corresponding preset fault mode;

[0011] The fault prediction module uses a machine learning model to train the time mapping relationship between position feature values, operating parameters and fault modes, and is used to predict the position deviation and fault mode of the next time period based on the current operating parameters and position feature values ​​and perform corrective control.

[0012] Furthermore, the working parameters include coating speed, coating tension, coating pressure, coating flow rate and coating viscosity.

[0013] Furthermore, the acquisition of position characteristic values ​​of multiple position parameters and the determination of whether there is a deviation in the current position are specifically implemented using a principal component analysis method, including the following steps:

[0014] S1. Collect position parameter data samples of multiple position sensors in historical periods;

[0015] S2. Reduce the dimensions of different position parameters through principal component analysis to obtain several principal components;

[0016] S3. Selecting a calculation formula for the first principal component to calculate positional eigenvalues ​​of multiple positional parameters;

[0017] S4. When the current position feature value exceeds a preset position feature value threshold, it is determined that there is a deviation in the current position.

[0018] Furthermore, the failure modes include guide roller wear or improper installation, loose or faulty transmission parts, coating head blockage or uneven discharge, excessive or unstable coating speed, abnormal coating tension, inappropriate coating viscosity and unstable coating flow.

[0019] Furthermore, the machine learning model includes the following construction steps:

[0020] T1. Collect a time series coating machine working data set with each complete coating process as a cycle in the historical period;

[0021] T2. Establish a time series model for synchronizing working parameters and position characteristic values ​​during each complete coating process.

[0022] Furthermore, the time series model includes the following construction steps:

[0023] T21. Data preprocessing: Remove noise and outliers from the collected time series coating machine working data, and standardize the data using the Z-score standardization method;

[0024] T22. Feature Engineering: Mining the temporal characteristics of time series data, combining working parameters with position feature values, and mining potential nonlinear relationships between features;

[0025] T23, ARMA model training and tuning: The preprocessed and feature-engineered data is divided into training and test sets in chronological order and at a specific ratio. The ARMA model is trained using the training set. By continuously adjusting the autoregressive order p and the moving average order q, the loss function is minimized, allowing the model to accurately learn the time series relationship between the working parameters and the position eigenvalues. The autoregressive order p and the moving average order q are optimized using cross-validation.

[0026] T24. Model evaluation and validation: Use the test set data to evaluate the trained ARMA model, use the mean square error and mean absolute error to measure the deviation between the predicted value and the actual value, and judge the model performance; if it does not meet expectations, return to adjust the feature engineering, autoregressive order p, and moving average order q.

[0027] Furthermore, the method of predicting the position deviation and failure mode of the next time period based on the current working parameters and position characteristic values ​​and performing correction control includes the following steps:

[0028] E1. Obtain the working parameters and position characteristic values ​​of the current time series, input them into the time series model, and predict the working parameters and position characteristic values ​​of the next time period;

[0029] E2. Based on the position feature value of the next time period, predict whether there will be position deviation in the next time period;

[0030] E3. When a position deviation is predicted for the next time period, the hash value of the working parameters for the next time period is calculated, and the failure mode for the next time period is predicted by searching the hash table;

[0031] E4. Adjust and correct the operating parameters and the position of the coating machine according to the predicted failure mode in the next time period.

[0032] Furthermore, in step E3, the failure mode of the next time period is predicted, and the working parameters and position characteristic values ​​are used as explanatory variables, and the correction plan corresponding to the failure mode is used as the response variable. A random forest model is trained to output the correction plan for the next time period based on the working parameters and position characteristic values ​​of the next time period.

[0033] Furthermore, the random forest model includes the following construction steps:

[0034] P1. Feature selection and processing: Select the working parameters and position characteristic values ​​related to the correction scheme labels in the historical period as features;

[0035] P2. Random forest model training: Establish multiple decision trees between working parameters and position feature values ​​and correction solutions. Each decision tree randomly extracts samples and features from the training data for training.

[0036] P3. Hyperparameter Tuning: Adjust the number of decision trees, maximum depth, and minimum number of samples required for node splitting, and optimize the random forest model through cross-validation.

[0037] P4. Model evaluation and application: Use an independent test set to evaluate the trained model, using accuracy, recall, and F1 score as evaluation metrics.

[0038] Beneficial effects of the present invention:

[0039] This invention monitors and collects the position parameters and operating parameters of multiple position sensors on a coating machine in real time; obtains the position characteristic values ​​of multiple position parameters to determine whether there is a deviation in the current position; when a position deviation exists, calculates the hash value of the operating parameter using a hash function and searches the hash table for the corresponding preset fault mode; and uses a machine learning model to train the time mapping relationship between the position characteristic values, operating parameters, and fault modes. This is used to predict the position deviation and fault mode in the next time period based on the current operating parameters and position characteristic values, and to perform corrective control. This invention solves the problem of slow response speed in detecting position deviations and faults in existing technologies, while simultaneously improving the accuracy and efficiency of fault detection by utilizing a time series model and a random forest model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0041] Figure 1 This is a structural diagram of an automatic coating machine correction control system in the present invention.

[0042] Figure 2 This is a flowchart of determining whether there is a deviation in the current position by using position feature values ​​in one embodiment of the present invention.

[0043] Figure 3 This is a flow chart for predicting position deviation and failure mode in the next time period and performing correction control based on current operating parameters and position characteristic values ​​in one embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0045] See also Figure 1-Figure 3 The present application provides an automatic coating machine deviation correction control system, including a data acquisition module, a deviation judgment module, a fault identification module and a fault prediction module; the data acquisition module, the deviation judgment module, the fault identification module and the fault prediction module are communicatively connected, wherein:

[0046] The data acquisition module is used to monitor and collect position parameters and operating parameters of multiple position sensors of the coating machine in real time, wherein the operating parameters include coating speed, coating tension, coating pressure, coating flow rate and coating viscosity;

[0047] In this embodiment, the core function of the data acquisition module is to monitor and collect various key data during the operation of the coating machine in real time, including position parameters fed back by multiple position sensors and a series of working parameters. In terms of position parameter collection, multiple position sensors are arranged at various key nodes of the coating machine, such as the discharge end, the starting and ending positions of the coating area, and specific positions where there may be risks of material deviation. These sensors can accurately sense the position information of the coating material at each moment through photoelectric sensing, laser ranging or ultrasonic detection, and provide the system with intuitive data on whether the material deviates from the preset path. The working parameters include parameters such as coating speed, coating tension, coating pressure, coating flow rate and coating viscosity, which reflect the operating status of the coating machine and the key elements of the coating process from different dimensions. Among them, the coating speed directly affects the distance the coating material moves per unit time. If the speed is too fast, the coating material is affected by inertia and is likely to deviate from the predetermined position when turning or passing through complex structures, resulting in inaccurate coating position and uneven coating. If the speed is too slow, it will not only affect production efficiency, but may also cause the material to shift in position under tension due to the long-term force acting in a single direction.

[0048] Coating tension is a key factor in maintaining stable coating material flow. Excessive tension can cause excessive stretching of the material, leading to positional deviations due to uneven localized force when passing through guide rollers or coating heads. Too little tension can cause material slack, leading to wrinkles and snaking during operation, potentially causing uncontrolled coating position.

[0049] Coating pressure primarily affects the transfer and distribution of coatings onto the substrate. Uneven coating pressure can lead to variations in the coating's distribution across the substrate surface, with more coating on one side and less on the other. This uneven coating can cause different forces on each side of the material, leading to positional shifts in the coating during subsequent runs.

[0050] Paint flow determines the amount of paint applied to the substrate per unit time. If the paint flow is unstable, the coating thickness will vary at different locations in the coating area. Thicker coating areas may cause localized deformation of the material due to factors such as gravity and drying shrinkage, affecting the accuracy of the coating position.

[0051] Paint viscosity reflects its flow characteristics. If the viscosity is too high, the paint will have poor fluidity, making it difficult to spread evenly on the substrate and prone to localized accumulation. This can cause the coating material to shift position during application due to uneven force. If the viscosity is too low, the paint will flow too quickly on the substrate, making it difficult to control the coating range and affecting the accuracy of the coating position.

[0052] The deviation judgment module is used to obtain position characteristic values ​​of multiple position parameters and judge whether there is a deviation in the current position;

[0053] Traditional fault diagnosis algorithms may involve a large number of complex conditional judgments and logical nesting. For example, when determining whether a coating machine's material has deviated, it may be necessary to sequentially check the data from multiple sensors at different locations, and the data from each sensor must undergo multiple threshold judgments. These judgments also involve complex logical relationships, such as "and" and "or" relationships, making the entire judgment process cumbersome and time-consuming. These complex logics are sorted out and simplified, and multiple related conditional judgments are combined into a comprehensive judgment condition. For example, by analyzing historical data, the correlation between sensor data at different locations is found, and mathematical models are used to integrate multiple sensor data into a single characteristic value. By simply performing a single threshold judgment on this characteristic value, it is possible to quickly determine whether the material has deviated, thereby reducing a large number of multiple threshold judgment steps and improving judgment speed.

[0054] Furthermore, the acquisition of position characteristic values ​​of multiple position parameters and the determination of whether there is a deviation in the current position are specifically implemented using a principal component analysis method, including the following steps:

[0055] S1. Collect location parameter data samples from multiple location sensors over a historical period:

[0056] During coater operation, sensors at various locations continuously generate position parameter data. This data contains information about the coater at different times and locations, reflecting the positional status of the material within the coater. Collecting historical data samples allows for a comprehensive understanding of the coater's performance under various operating conditions, including normal operation, varying coating speeds, material types, and environmental conditions. The quantity and diversity of these data samples are crucial for subsequent analysis, as they form the foundational dataset for understanding and analyzing the coater's positional status. For example, data from the past several months or even years can be collected, covering different shifts, seasons, and material batches. This data might be stored in a database as a time series, recording the positional parameters of various position sensors (such as those at the feed port, discharge port, near the coating head, and on the material conveyor track) at different times. Analyzing this historical data can reveal underlying patterns and characteristics, providing sufficient data support for subsequent principal component analysis.

[0057] S2. Reduce the dimensions of different position parameters through principal component analysis to obtain several principal components:

[0058] Principal component analysis (PCA) is a commonly used statistical method that transforms multiple correlated variables (here, multiple positional parameters) into a new set of uncorrelated variables, known as principal components. In the context of a coating machine, complex correlations may exist between the multiple positional parameters. For example, changes in the feed port position may affect the discharge port position, and the positional parameters at different locations are not completely independent. The dimensionality reduction process of PCA linearly combines these positional parameters, transforming the original high-dimensional positional parameter space into a low-dimensional principal component space. In this process, each principal component is a linear combination of the original positional parameters and is sorted by the amount of explained variance. The first principal component explains the greatest variance and contains the most information in the original data, while subsequent principal components explain progressively less variance. This method simplifies the original multiple positional parameters into a small number of principal components while preserving most of the data information. For example, if there are 10 positional parameters, PCA processing may yield three principal components. These three principal components capture most of the information from the original 10 positional parameters with fewer dimensions, making subsequent analysis and processing more concise and efficient. Such dimensionality reduction processing helps to remove redundant information in the data, highlight the main features of the data, and reduce the amount of calculation and the complexity of analysis.

[0059] S3. Select the calculation formula of the first principal component to calculate the position eigenvalues ​​of multiple position parameters:

[0060] Among the several principal components obtained, the first principal component holds a special place. This is because it explains the largest proportion of the variance in the original data and is the comprehensive variable that best represents the overall trend of change in the original position parameters. Selecting the first principal component as the position eigenvalue is a reasonable choice because it integrates the most important information from the original position parameters. It can be considered an indicator that represents the overall position status of the coater. For example, a large value for the first principal component may indicate a significant overall shift in the material's position within the coater; a value close to zero may indicate that the material's position is relatively stable, close to its normal position. This position eigenvalue concisely summarizes the information reflected by multiple position parameters, providing a single quantitative indicator for subsequent deviation judgment.

[0061] S4. When the current position feature value exceeds the preset position feature value threshold, it is determined that there is a deviation in the current position:

[0062] To determine whether the current position has deviated, a reference standard is required: a preset position characteristic value threshold. This threshold is set based on analysis of historical data and experience and understanding of the coater's normal operation. During normal operation, the first principal component (position characteristic value) should fluctuate within a relatively stable range, with the threshold defining the boundary of this range. When the first principal component (i.e., position characteristic value) calculated based on the current position parameters exceeds this threshold, it indicates that the current position has deviated from the normal range, thus indicating a deviation. For example, if historical data shows that the position characteristic value of the first principal component normally fluctuates between -2 and 2, the threshold can be set to ±2.5. If the position characteristic value calculated at the current moment is 3, it is determined that the current position has deviated. This deviation can be caused by a variety of factors, such as wear of mechanical components, uneven material distribution, or sudden changes in coating speed. Once a deviation is determined, the system can trigger subsequent corrective actions, such as adjusting the position of the guide rollers and the coating tension, to ensure normal operation and coating quality.

[0063] The fault identification module is configured to calculate a hash value of the operating parameter by a hash function when there is a position deviation, and search the hash table for a corresponding preset fault mode;

[0064] In this embodiment, the fault identification module plays a key role in the automatic coating machine correction control system and is activated when the deviation judgment module determines that there is a position deviation. It takes the working parameters of the coating machine when the deviation occurs (such as coating speed, tension, pressure, coating flow and viscosity, etc.) and the position deviation characteristics as input, and calculates a hash value that can highly condense the current working state through a hash function. The hash value is unique. When designing the system, engineers combine the working parameters corresponding to various possible fault modes based on a large amount of historical fault data and experience. After the hash value is calculated by the hash function, it is stored in the hash table together with the corresponding fault mode information. After the fault identification module calculates the current hash value, it immediately searches for a matching item in the hash table. Once found, it can quickly obtain information such as the specific description of the preset fault mode, cause analysis and solution suggestions, so as to quickly locate the fault and provide support for subsequent effective corrective measures. This not only improves the efficiency of fault diagnosis, but also ensures accuracy with the help of historical data and experience. As new fault data accumulates, the fault identification capability can be further improved.

[0065] Furthermore, during the operation of the automated coating machine, when position deviation occurs, the fault modes found by the fault identification module include various types. The following are the fault modes covered in this embodiment:

[0066] Worn or improperly installed guide rollers: Guide rollers are responsible for guiding the coating material. Over time, wear may occur on the guide roller surface, resulting in uneven friction between the roller and the material, causing the material to deviate from its normal position during operation. Additionally, if the guide roller is not accurately positioned during installation, for example, not parallel to the coating machine's centerline, this can also cause the material to deviate. In this case, the fault manifests as the material gradually deviating from the predetermined path after passing the guide roller. The hash value calculated using the hash function will match the preset fault mode hash value for positional deviation caused by a guide roller problem.

[0067] Loose or faulty transmission components: If the coating machine's transmission system, such as belts, chains, or gears, becomes loose, worn, or damaged, it can lead to unstable power transmission and changes in the speed and direction of the coating material. For example, a loose belt can cause slippage between the belt and the drive pulley, resulting in inconsistent material speeds at different locations and, in turn, positional deviation. This type of fault is reflected in the fault identification module as specific operating parameter changes and positional deviation characteristics, and the corresponding fault mode is matched through hash function calculation.

[0068] Blocked or uneven coating head discharge: The coating head is responsible for applying the coating evenly to the material surface. When the coating head's discharge port is partially blocked, the coating will flow unevenly, resulting in inconsistent coating thickness on the material. This generates lateral forces that cause the material to shift position. In addition to positional deviation, this fault mode is also accompanied by abnormal changes in coating flow and coating pressure. The fault identification module uses a hash function to match the corresponding fault mode based on these combined characteristics.

[0069] Excessively fast or unstable coating speed: When the coating speed is too fast, the inertial force on the material during the coating process increases, making it prone to deviation from the intended path. Furthermore, if the coating speed is unstable, with speeds fluctuating, the material will be subjected to different forces at different speeds, resulting in positional deviation. For example, during acceleration or deceleration, the material may briefly deviate. In this fault mode, abnormal changes in coating speed become a key parameter for fault identification, and a hash function is used to match the preset fault mode of positional deviation caused by coating speed issues.

[0070] Abnormal coating tension: Excessive or insufficient coating tension can cause positional deviation. Excessive tension overstretches the material, making it susceptible to deviation when passing through certain components due to uneven localized force. Excessive tension causes the material to sag, resulting in wrinkles and snaking during operation, leading to loss of positional control. The fault identification module uses a hash function to identify the fault mode caused by abnormal coating tension based on the abnormal tension data collected by the tension sensor and the positional deviation information.

[0071] Inappropriate paint viscosity: If the viscosity is too high, the paint will have poor fluidity, making it difficult to spread evenly on the substrate and prone to localized accumulation, causing the material to shift position during application due to uneven force. If the viscosity is too low, the paint will flow too quickly on the substrate, making it difficult to control the coating range and also affecting the accuracy of the coating position. The fault identification module uses a hash function to match the corresponding fault mode caused by the paint viscosity problem based on the measured paint viscosity and position deviation.

[0072] Unstable paint flow: Unstable paint flow can lead to uneven coating thickness. Thicker areas of coating can deform locally due to factors such as gravity and drying shrinkage, affecting coating accuracy. The fault identification module uses a hash function to identify fault modes caused by unstable paint flow based on data from the paint flow sensor and position deviation characteristics.

[0073] The fault prediction module uses a machine learning model to train the time mapping relationship between position feature values, operating parameters and fault modes, and is used to predict the position deviation and fault mode of the next time period based on the current operating parameters and position feature values ​​and perform corrective control.

[0074] In this embodiment, the fault prediction module uses a machine learning model to explore the time mapping relationship between position characteristic values, working parameters and fault modes, thereby improving the fault prediction and response capabilities of the automated coating machine. The module first collects a large amount of historical data, covering various working parameters, corresponding position characteristic values ​​and fault modes, and then uses a machine learning model to deeply mine and analyze these data. After repeated training, the model accurately grasps the impact of changes in working parameters on position characteristic values ​​and fault modes. During actual operation, the module uses the trained model to predict position deviations and possible fault modes based on the current working parameters collected in real time, and then implements targeted correction control based on the prediction results, such as adjusting the position of the guide roller, changing the coating speed or adjusting the tension, etc., to avoid faults in advance and ensure stable operation of the coating machine. This not only achieves early warning of faults, but also enables intelligent decision-making on correction measures, reduces dependence on manual experience, and significantly improves production efficiency and product quality.

[0075] Furthermore, the machine learning model includes the following construction steps:

[0076] T1. Collect a time series data set of coating machine operations with each complete coating process as a cycle in the historical period:

[0077] By collecting enough historical data, these data are divided according to each complete coating process. Each complete coating process contains key information of the coating machine from the beginning to the end of coating. The collected data covers the position characteristic values. The comprehensive position information calculated by the previously mentioned deviation judgment module using principal component analysis and other methods reflects the position status of the coating material in the coating machine; the working parameters include coating speed, coating tension, coating pressure, coating flow and coating viscosity, etc., which together describe the working status of the coating machine; the fault mode is a record of whether a fault occurred during the coating process and the specific type of fault, which may be a mechanical fault, process parameter fault, etc.

[0078] T2. Establish a time series model for synchronizing working parameters and position characteristic values ​​during each complete coating process:

[0079] For each complete coating process, the operating parameters and position characteristic values ​​are considered as a data series that changes over time. For example, during a coating process, the coating speed may be adjusted over time, and the position characteristic values ​​will also change accordingly. These changes all follow a certain time series pattern.

[0080] Common time series models, such as ARIMA and SARIMA, can be used to model operating parameters and positional characteristics. By establishing a time series model, it is possible to capture the dynamic trends of these parameters within a coating cycle, as well as their interrelationships and temporal dependencies. A temporal model can help understand how operating parameter adjustments affect positional characteristics, their temporal sequence, and the delayed effects of their interactions, providing more detailed information for subsequent fault prediction.

[0081] Furthermore, the time series model includes the following construction steps:

[0082] T21. Data Preprocessing: For the time series coating machine operating data collected at each complete coating process, we first carefully investigate and remove noise and outliers caused by sensor failures, data transmission errors, etc., and use appropriate interpolation methods to fill missing data to ensure data integrity. Given the different dimensions and value ranges of operating parameters and position feature values, we use Z-score standardization or minimum-maximum normalization methods to bring the data to the same scale. This ensures that each feature is appropriately weighted in subsequent analysis, laying a good foundation for building the ARMA model.

[0083] T22. Feature Engineering: Deeply explore the temporal characteristics of time series data, extract periodic information such as seconds, minutes, and hours, as well as trend features such as moving average and difference that can reflect the long-term trend and short-term fluctuations of the data. At the same time, combine the working parameters and position characteristic values, such as generating new features such as products and ratios, or implementing transformations such as squares and square roots, to deeply explore the potential nonlinear relationships between features, enrich the data feature dimensions, and help the ARMA model better capture data patterns. In view of the characteristics of the coating machine working data, after comprehensive consideration, the ARMA model is more suitable. It can effectively handle data that has both autocorrelation and interference from random factors. Combined with evaluation indicators such as mean square error and mean absolute error, this model will be used in the future to deeply analyze the dynamic changes and mutual connections between working parameters and position characteristic values ​​in time series.

[0084] T23. ARMA Model Training and Tuning: Split the preprocessed and feature-engineered data into a training set and a test set in chronological order with an appropriate ratio (e.g., 80:20). Use the training set to train the ARMA model. By continuously adjusting the autoregressive order p and the moving average order q, minimize loss functions such as the AIC or BIC, enabling the model to accurately learn the time series relationship between operating parameters and positional eigenvalues. Simultaneously, employ strategies such as cross-validation to carefully tune hyperparameters such as p and q to improve the model's predictive performance and generalization capabilities.

[0085] T24. Model Evaluation and Validation: Evaluate the trained ARMA model using the test data set. Model performance is assessed by measuring the deviation between the predicted and actual values ​​using metrics such as mean squared error and mean absolute error. If performance falls short of expectations, adjustments to the model order and feature engineering methods are necessary. Furthermore, cross-validation is used to ensure model stability and reliability, avoid overfitting, and ensure the model accurately predicts the coating machine's operating status in real-world applications.

[0086] Furthermore, the method of predicting the position deviation and failure mode of the next time period based on the current working parameters and position characteristic values ​​and performing correction control includes the following steps:

[0087] E1. Obtain the working parameters and position characteristic values ​​of the current time series, input them into the time series model, and predict the working parameters and position characteristic values ​​of the next time period;

[0088] E2. Based on the position feature value of the next time period, predict whether there will be position deviation in the next time period;

[0089] E3. When a position deviation is predicted for the next time period, the hash value of the working parameters for the next time period is calculated, and the failure mode for the next time period is predicted by searching the hash table;

[0090] E4. Adjust and correct the operating parameters and the position of the coating machine according to the predicted failure mode in the next time period.

[0091] Furthermore, in step E3, the failure mode of the next time period is predicted, and the working parameters and position characteristic values ​​are used as explanatory variables, and the correction plan corresponding to the failure mode is used as the response variable. A random forest model is trained to output the correction plan for the next time period based on the working parameters and position characteristic values ​​of the next time period.

[0092] By using the working parameters and position characteristic values ​​as explanatory variables and the correction plan corresponding to the fault mode as the response variable, a random forest model is trained to output an accurate correction plan based on the working parameters and position characteristic values ​​of the next time period.

[0093] First, a massive amount of historical data is collected, covering the operating parameters of the coating machine under various operating conditions, such as coating speed, tension, pressure, coating flow rate and viscosity, as well as positional features. Each failure mode and its corresponding corrective action are also recorded in detail. For example, when excessive coating speed and abnormal tension cause material position shift and wrinkling, the corrective action involves reducing the coating speed, fine-tuning the tension system, and adjusting the guide roller position. This data forms the cornerstone of the training model. Random forest, a powerful ensemble learning algorithm, consists of numerous decision trees. During training, each decision tree randomly selects a subset of samples and features for learning, effectively avoiding overfitting and enabling each decision tree to discover data patterns from different dimensions. In this scenario, the operating parameters and positional features serve as input, and the model learns the complex relationships between them and the corresponding corrective actions for the failure mode. For example, when fed a specific set of operating parameters and positional features, the model, based on past learning experience, identifies the likely failure mode and recommends a corresponding corrective action. After training is completed, facing the working parameters and position characteristic values ​​of the next time period, the model can quickly output targeted correction plans, prevent failures in advance, reduce downtime and production losses, help the coating machine operate stably and efficiently, and provide operators with a scientific decision-making basis.

[0094] Furthermore, the random forest model includes the following construction steps:

[0095] P1. Feature selection and processing: Taking working parameters and position eigenvalues ​​as explanatory variables, they need to be processed. The first is feature selection. Not all working parameters have the same important impact on failure modes and correction solutions. Some parameters may be redundant or have low correlation. Excluding them can make the model more focused on key information and improve training efficiency. At the same time, since the value ranges of different working parameters and position eigenvalues ​​may vary greatly, normalization or standardization is required. Taking coating speed and paint viscosity as an example, if they are not processed, their different magnitudes will make the model more inclined to focus on features with large values ​​during training, and ignore features with small values, resulting in deviations in model training. By converting them to the same dimensional range, such as normalizing them to [0,1], the model can consider the impact of each feature equally and learn the relationship between them and correction solutions more accurately.

[0096] P2. Random Forest Model Training: The random forest model is a powerful ensemble learning algorithm, whose core lies in the collaborative decision-making of multiple decision trees. Each decision tree randomly extracts samples and features from the training data for training, allowing different decision trees to learn the data characteristics from different perspectives. Each decision tree will find different decision paths based on the different values ​​of the input operating parameters and position feature values, ultimately forming a judgment on the correction method. For example, one decision tree may discover that when the coating speed is high and the position feature values ​​deviate from a certain range, it means that the material is moving too fast and the position is deviating. In this case, corrective measures such as reducing the speed and adjusting the guide roller position are necessary. Meanwhile, another decision tree may discover from the combined characteristics of coating viscosity and coating tension that the material may be experiencing positional deviation and uneven coating due to high viscosity and low tension, and thus derive the correction method of adding diluent and adjusting the tension. The integrated decision-making of multiple decision trees enables the model to learn very complex mapping relationships, avoiding the one-sidedness that may occur in a single decision tree, and improving the model's robustness and generalization ability.

[0097] P3. Hyperparameter adjustment: Hyperparameters play an important regulatory role in the random forest model, and they determine the structure and performance of the model. Hyperparameters such as the number of decision trees, the maximum depth of each tree, and the minimum number of samples required for node splitting all need to be carefully adjusted. Through cross-validation, we can try different hyperparameter combinations, divide the data into multiple parts, one part for training and one part for validation, and compare the performance of the model on the validation set under different hyperparameter combinations. For example, when adjusting the number of decision trees, too few decision trees may lead to insufficient learning ability of the model, while too many decision trees may lead to overfitting, making the model too dependent on training data and losing the ability to generalize to new data. Through continuous attempts, a suitable hyperparameter combination is found so that the model has a good learning effect on the training set, while maintaining good performance on the validation set, ensuring that the model can accurately predict the correction solution in practical applications.

[0098] P4. Model evaluation and application: Use an independent test set to evaluate the trained model. For classified correction schemes, use indicators such as accuracy, recall rate, and F1 score. They can measure whether the correction scheme predicted by the model is accurate and whether it can cover various possible situations; for regression-type correction schemes (if the correction scheme includes continuous numerical adjustments), use indicators such as mean square error and mean absolute error to evaluate the deviation between the correction scheme predicted by the model and the actual situation. When the evaluation results show that the model performs well, it can be applied to actual operations. When the working parameters and position characteristic values ​​of the next time period are input, the model can output a specific correction scheme based on the existing learning results, such as giving detailed operational suggestions such as "increasing the coating pressure, adjusting the guide roller angle", etc., to provide timely and effective decision support for the stable operation of the coating machine.

[0099] Furthermore, the evaluation index is calculated as follows:

[0100]

[0101] Where ACC is the accuracy rate, PPV is the precision rate, TRP is the recall rate, F1 is the F1 value, TP is the true positive rate, TN is the true negative rate, FP is the false positive rate, and FN is the false negative rate.

[0102] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. An automatic coating machine correction control system, characterized by: It includes a data acquisition module, a deviation judgment module, a fault identification module and a fault prediction module; the data acquisition module, the deviation judgment module, the fault identification module and the fault prediction module are communicatively connected, wherein: The data acquisition module is used to monitor and collect position parameters and working parameters of multiple position sensors of the coating machine in real time; The deviation judgment module is used to obtain position characteristic values ​​of multiple position parameters and judge whether there is a deviation in the current position; The fault identification module is configured to calculate a hash value of the operating parameter by a hash function when there is a position deviation, and search the hash table for a corresponding preset fault mode; The fault prediction module uses a machine learning model to train the time mapping relationship between position feature values, operating parameters and fault modes, and is used to predict the position deviation and fault mode of the next time period based on the current operating parameters and position feature values ​​and perform correction control; The working parameters include coating speed, coating tension, coating pressure, coating flow rate and coating viscosity; The method of obtaining the position characteristic values ​​of the plurality of position parameters and determining whether the current position has a deviation is specifically implemented by using a principal component analysis method, including the following steps: S1. Collect position parameter data samples of multiple position sensors in historical periods; S2. Reduce the dimensions of different position parameters through principal component analysis to obtain several principal components; S3. Selecting a calculation formula for the first principal component to calculate positional eigenvalues ​​of multiple positional parameters; S4. When the current position feature value exceeds a preset position feature value threshold, it is determined that there is a deviation in the current position; The failure modes include worn or improperly installed guide rollers, loose or faulty transmission components, clogged or uneven coating head discharge, excessive or unstable coating speed, abnormal coating tension, inappropriate coating viscosity, and unstable coating flow; The method of predicting the position deviation and failure mode of the next time period based on the current working parameters and position characteristic values ​​and performing correction control includes the following steps: E1. Obtain the working parameters and position characteristic values ​​of the current time series, input them into the time series model, and predict the working parameters and position characteristic values ​​of the next time period; E2. Based on the position feature value of the next time period, predict whether there will be position deviation in the next time period; E3. When a position deviation is predicted for the next time period, the hash value of the working parameters for the next time period is calculated, and the failure mode for the next time period is predicted by searching the hash table; E4. Adjust and correct the operating parameters and the position of the coating machine according to the predicted failure mode in the next time period.

2. The automatic coating machine correction control system according to claim 1, characterized in that: The machine learning model includes the following construction steps: T1. Collect a time series coating machine working data set with each complete coating process as a cycle in the historical period; T2. Establish a time series model for synchronizing working parameters and position characteristic values ​​during each complete coating process.

3. The automatic coating machine correction control system according to claim 2, characterized in that: The time series model includes the following construction steps: T21. Data preprocessing: Remove noise and outliers from the collected time series coating machine working data, and standardize the data using the Z-score standardization method; T22. Feature Engineering: Mining the temporal characteristics of time series data, combining working parameters with position feature values, and mining potential nonlinear relationships between features; T23, ARMA model training and tuning: The preprocessed and feature-engineered data is divided into training and test sets in chronological order and at a specific ratio. The ARMA model is trained using the training set. By continuously adjusting the autoregressive order p and the moving average order q, the loss function is minimized, allowing the model to accurately learn the time series relationship between the working parameters and the position eigenvalues. The autoregressive order p and the moving average order q are optimized using cross-validation. T24. Model evaluation and validation: Use the test set data to evaluate the trained ARMA model, use the mean square error and mean absolute error to measure the deviation between the predicted value and the actual value, and judge the model performance; if it does not meet expectations, return to adjust the feature engineering, autoregressive order p, and moving average order q.

4. The automatic coating machine correction control system according to claim 1, characterized in that: In step E3, the failure mode of the next time period is predicted, and the working parameters and position characteristic values ​​are used as explanatory variables, and the correction plan corresponding to the failure mode is used as the response variable. A random forest model is trained to output the correction plan for the next time period based on the working parameters and position characteristic values ​​of the next time period.

5. The automatic coating machine deviation correction control system according to claim 4, characterized in that: The random forest model includes the following construction steps: P1. Feature selection and processing: Select the working parameters and position characteristic values ​​related to the correction scheme labels in the historical period as features; P2. Random forest model training: Establish multiple decision trees between working parameters and position feature values ​​and correction solutions. Each decision tree randomly extracts samples and features from the training data for training. P3. Hyperparameter Tuning: Adjust the number of decision trees, maximum depth, and minimum number of samples required for node splitting, and optimize the random forest model through cross-validation. P4. Model evaluation and application: Use an independent test set to evaluate the trained model, using accuracy, recall, and F1 score as evaluation metrics.

Images