Method for collecting and tracing data in production process of PVC decorative film

Abstract

The invention provides a PVC decorative film production process data acquisition and tracing method, and relates to the technical field of intelligent manufacturing and quality control, and the method comprises the steps: collecting film layer thickness and calendaring process parameters in real time through an online detection system; calculating a microscopic uniformity index MUI on the edge side based on the thickness data, and calculating a process stability index PSI based on the temperature data; inputting the MUI and the PSI into a quality early warning model, and predicting a precision stripe defect risk level; when the early warning exceeds the threshold, an instruction is automatically generated to adjust the temperature or the linear speed of the calendering roller; all process data, control instructions and quality indexes are stored in an associated mode, and a batch quality control file is established; and the model is periodically updated by adopting an online learning mechanism. According to the invention, real-time early warning and closed-loop control of microdefects are realized, quality consistency and yield are improved, and intelligence and stability of production are guaranteed through full-process data tracing and model self-optimization.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and product quality traceability technology, and in particular to a method for data collection and traceability of the production process of PVC decorative film. Background Technology

[0002] PVC decorative film, widely used in home and commercial decoration, relies heavily on its appearance quality and the realism of its surface texture to determine product grade and market competitiveness. During production, a microscopic defect known as "precision lines" easily appears on the film surface. These defects manifest as irregular patterns with wavelengths ranging from 0.1 to 1.0 millimeters. While they do not directly affect physical properties, they severely damage visual appeal and product aesthetics, leading to decreased yield rates and the loss of high-value-added orders.

[0003] Currently, the industry's methods for controlling such micro-defects are relatively lagging and reactive. Most production lines rely on manual sampling or offline laboratory testing after production to detect precision pattern issues. This post-production inspection model introduces significant delays; by the time problems are discovered, a large number of defective products have often already been produced, resulting in a waste of raw materials and labor. Furthermore, manual interpretation suffers from strong subjectivity and poor consistency, making it difficult to accurately classify and quantify defects, and thus failing to meet the stringent quality uniformity requirements of the high-end market.

[0004] At the production process control level, existing methods mainly rely on monitoring and adjusting macroscopic process parameters, such as the average temperature of the calender rolls and the overall linear speed. However, the formation of precision lines is closely related to the thickness uniformity of the film layer at the microscale and the instantaneous fluctuations of process parameters. These microscopic dynamic changes cannot be captured by monitoring macroscopic average values. For example, small but rapid fluctuations in the temperature of the calender rolls, or subtle unevenness in the local thickness of the film layer, can trigger precision lines, but existing control systems cannot detect and respond in a timely manner due to the lack of targeted real-time monitoring indicators.

[0005] Regarding data utilization, the massive amounts of data collected during production, such as thickness, temperature, and speed, are mostly used only for simple historical records and trend displays, failing to undergo in-depth correlation analysis and value mining. There is a lack of effective causal links between process parameter adjustment records, real-time quality indicators, and final product defects, making traceability analysis difficult when quality problems occur, and hindering the accurate identification of the root cause. Simultaneously, control models typically use fixed parameters and cannot self-optimize in response to dynamic factors such as changes in raw material characteristics and equipment operating conditions. Over long-term operation, the accuracy of early warnings gradually declines, requiring frequent manual intervention and model readjustment, increasing maintenance costs and system instability. Summary of the Invention

[0006] To address the technical problems in existing technologies such as lagging detection of microscopic defects, inaccurate control, difficulty in data traceability, and insufficient adaptive capability of quality early warning models in the production process of PVC decorative film, this invention provides a method for data acquisition and traceability in the production process of PVC decorative film.

[0007] The technical solution provided by this invention is as follows:

[0008] This invention provides a method for data collection and traceability in the production process of PVC decorative film, comprising:

[0009] S1: Through an online detection system deployed after the calendering process, the film thickness distribution data during the production process is collected in real time at a sampling frequency of not less than 100Hz, while the process parameters of calendering roll temperature, linear speed and cooling rate are also collected.

[0010] S2: Based on the film thickness distribution data, calculate the microuniformity index MUI on the edge computing device of the production line. The microuniformity index MUI is used to quantify the thickness fluctuation characteristics of the film at the microscale.

[0011] S3: Based on the calender roll temperature data, calculate the process stability index PSI on the edge computing device. The process stability index PSI is used to quantify the fluctuation characteristics of temperature parameters in the calendering process.

[0012] S4: Input the microuniformity index MUI and process stability index PSI into the pre-trained quality warning model to obtain the orange peel texture defect risk warning level; the orange peel texture defect refers to the micro-irregular texture with a wavelength in the range of 0.1-1.0 mm that appears on the film surface;

[0013] S5: When the warning level exceeds the preset threshold, a control command is automatically generated and sent to the calendering equipment control system to adjust the calendering roll temperature or linear speed according to the control cycle.

[0014] S6: Link and store each control command, process parameter adjustment record, real-time collected film thickness distribution data and corresponding MUI and PSI values ​​to establish a complete quality control file for each production batch;

[0015] S7: The quality early warning model adopts an online learning mechanism and regularly updates and optimizes the model parameters using new production data;

[0016] The calculation of the microuniformity index (MUI) and the process stability index (PSI) ensures that the real-time response time is less than 100 milliseconds.

[0017] The beneficial effects of the technical solution provided by this invention include at least the following:

[0018] (1) In this invention, by deploying a high-frequency online detection system and combining it with edge computing devices to calculate the micro-uniformity index and process stability index in real time, the micro-fluctuations in film quality and instantaneous anomalies in process parameters are captured and quantitatively evaluated instantly. This technology transforms the previously lagging and passive manual quality inspection into an advanced and proactive online intelligent diagnosis, which can issue risk warnings at the early stage of precision texture defects. This not only significantly reduces the scrap rate and quality loss caused by batch micro-defects, but also effectively improves the ability to ensure the quality of high-end products, making the quality control of the production process more refined and forward-looking.

[0019] (2) In this invention, a rapid response loop from quality risk identification to automatic adjustment of process parameters is established by constructing an early warning model based on a real-time quality index and automatically generating closed-loop control commands. This system can accurately calculate the optimal adjustment amount of the calender roll temperature or linear speed according to the early warning level, and ensure control stability through anti-oscillation logic. This solves the pain points of traditional control methods being insensitive to micro-defects and having lag in adjustment, realizing proactive intervention and adaptive optimization of the production process, thereby significantly improving the micro-uniformity and overall quality stability of the film product while maintaining the continuity of the production rhythm.

[0020] (3) In this invention, a data traceability system is constructed by associating and storing production data, control instructions, and quality indicators throughout the entire process, and establishing a complete quality control file for each batch. Simultaneously, the online learning mechanism employed by the quality early warning model continuously absorbs new production data to optimize itself, maintaining the long-term reliability of early warning accuracy. This combination not only ensures that every process adjustment and its quality results are traceable, greatly facilitating the root cause analysis of quality problems, but also enables the entire system to continuously evolve and adapt to complex working conditions, reducing reliance on external manual adjustments and enhancing the autonomy and durability of the intelligent manufacturing system. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart illustrating a method for data collection and traceability in the production process of PVC decorative film, provided in an embodiment of the present invention;

[0023] Figure 2A schematic diagram of the pre-training process of the quality early warning model in a data acquisition and traceability method for the production process of PVC decorative film provided in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the process of adjusting the temperature or linear speed of the calendering roller in a data acquisition and traceability method for the production process of PVC decorative film provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0026] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0027] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0028] In embodiments of the present invention, sometimes the subscript is as follows: It may be mistakenly written as a non-subscript form such as W1. When the distinction is not emphasized, the meaning they express is the same.

[0029] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0030] Reference manual attached Figure 1 The diagram shows a flowchart of a method for data collection and traceability of the production process of PVC decorative film provided by an embodiment of the present invention.

[0031] This invention provides a method for data collection and traceability in the production process of PVC decorative film, the processing flow of which may include the following steps:

[0032] S1: An online detection system deployed after the calendering process collects film thickness distribution data in real time at a sampling frequency of no less than 100Hz, and also collects process parameters such as calendering roll temperature, linear speed, and cooling rate.

[0033] An online detection system is installed at the exit of the calendering process. This system consists of a high-precision laser thickness gauge array, an infrared temperature sensor group, and an encoder. The laser thickness gauge uses a scanning measurement principle, with eight measurement points evenly arranged along the film width direction. Each measurement point acquires thickness data at a sampling frequency of 200Hz. The temperature of the calender roll is acquired via an embedded PT100 thermocouple, with the measurement point located 5mm below the roll surface. The linear velocity is acquired synchronously with the main drive roll via an absolute encoder. The cooling rate is monitored by two sets of non-contact infrared thermometers at the inlet and outlet of the cooling section. All sensor data is transmitted to the central controller via the PROFINET industrial Ethernet protocol, and time synchronization is achieved using the IEEE 1588 precision clock protocol to ensure that the time deviation of each parameter acquisition is less than 1 millisecond.

[0034] S2: Based on the film thickness distribution data, the microuniformity index MUI is calculated on the edge computing equipment of the production line. The microuniformity index MUI is used to quantify the thickness fluctuation characteristics of the film at the microscale.

[0035] The Microuniformity Index (MUI) is calculated on an edge computing gateway deployed alongside the production line. This gateway utilizes an Intel Xeon D-2145NT processor and is equipped with 16GB of DDR4 memory. The system extracts a continuous sequence of sampling points from the thickness distribution data stream in real time, processing 60 consecutive thickness data points per calculation cycle. During the calculation, the raw thickness data is first filtered using a 3-point moving average, and then the MUI index is calculated using a specially optimized floating-point unit. The calculation result is output in single-precision floating-point format, along with a timestamp, location coordinates, and device identifier, and is transmitted to the quality warning module via shared memory.

[0036] S3: Based on the temperature data of the calender rolls, calculate the process stability index (PSI) on the edge computing device. The process stability index (PSI) is used to quantify the fluctuation characteristics of temperature parameters in the calendering process.

[0037] The process stability index (PSI) is also calculated on the edge computing gateway, but in a different processing thread than the MUI calculation. The system maintains a circular buffer that continuously stores the calender roll temperature data for the most recent 30 seconds, with a sampling interval of 1 second. During PSI calculation, the temperature sequence of 30 consecutive time points is extracted from the buffer, and the temperature standard deviation of the most recent 1000 batches from the historical database is used as a benchmark. Double-precision floating-point arithmetic is used in the calculation process to ensure the accuracy of the exponential function calculation. The PSI value is normalized to the range of 0-1, and a process stability warning is triggered when the value is below 0.6.

[0038] S4: Input the Microuniformity Index (MUI) and Process Stability Index (PSI) into the pre-trained quality warning model to obtain the orange peel texture defect risk warning level. Orange peel texture defects refer to micro-irregular lines with wavelengths in the range of 0.1-1.0 mm that appear on the film surface.

[0039] The pre-trained quality warning model is deployed in the TensorRT inference engine of the edge computing gateway. The model's input layer receives standardized MUI and PSI feature values, and the output layer uses a softmax function to generate probability distributions for three risk levels. During actual inference, the system first performs z-score standardization on the input features and then performs forward propagation computation. The warning level is determined based on the output probability: low risk (0-0.3), medium risk (0.3-0.7), and high risk (0.7-1.0). The inference results are published to the monitoring system via the DDS data distribution service, with an average inference latency of less than 10 milliseconds.

[0040] S5: When the warning level exceeds the preset threshold, a control command is automatically generated and sent to the calendering equipment control system to adjust the temperature or linear speed of the calendering rolls according to the control cycle.

[0041] The control command generation module adopts an event-driven architecture, triggering control logic immediately upon receiving a high-risk warning. The system maintains a priority task queue, with the control command generation task assigned the highest real-time priority. The command generation process first queries the current process parameter status, then calculates the adjustment amount based on the warning level. For temperature adjustment, an incremental PID algorithm is used; for linear speed adjustment, a feedforward control strategy is employed. The generated control commands are encapsulated in the ISO / IEC 9506 MMS protocol format and sent to the calendering equipment controller via the OPC UA interface, with a fixed command cycle of 100 milliseconds.

[0042] S6: Link and store each control command, process parameter adjustment record, real-time collected film thickness distribution data and corresponding MUI and PSI values ​​to establish a complete quality control file for each production batch.

[0043] The data logging system is built on the InfluxDB time-series database and deployed using a distributed architecture. Each data point contains fields such as nanosecond-level timestamp, device ID, parameter type, and value. Control command records, in addition to basic parameters, also include command hash values, digital signatures, and receipt status. The system creates an independent namespace for each production batch, organizing data according to a hierarchical structure of "factory-production line-date-batch number." Quality control archives automatically generate summary reports in JSON format, including key parameter statistics, adjustment operation logs, and quality assessment matrices, ensuring data integrity through the SHA-256 algorithm.

[0044] S7: The quality early warning model adopts an online learning mechanism, and regularly updates and optimizes the model parameters using new production data.

[0045] The online learning system employs a microservice architecture, with model update tasks orchestrated via Kubernetes. The system automatically triggers a model update process every 24 hours, extracting new production data from the data lake accumulated over the past 24 hours. The update process utilizes an incremental learning algorithm, fine-tuning the parameters of the fully connected layers using the Adam optimizer while maintaining the original model structure. Model performance monitoring tracks accuracy, precision, and recall metrics in real time. An automatic version rollback mechanism is triggered when performance degradation exceeds 5%, ensuring the stability of the production system.

[0046] Among them, the calculation of the microuniformity index (MUI) and the process stability index (PSI) ensures that the real-time response time is less than 100 milliseconds.

[0047] In one possible implementation, the step of calculating the micro-uniformity index MUI in S2 includes:

[0048] S201: Obtain the film thickness measurement values ​​of N consecutive sampling points in the detection area after the calendering process, where N≥50;

[0049] S202: Calculate the micro-uniformity index MUI according to the following formula:

[0050]

[0051] Where N is the number of consecutive sampling points, which is dimensionless; The thickness of the film at the i-th sampling point is expressed in micrometers. σ is the average film thickness at N sampling points, in micrometers; σ is the standard deviation of the thickness at N sampling points, in micrometers.

[0052] When calculating the Microuniformity Index (MUI), the system first extracts 60 consecutive sampling points from the thickness data stream, corresponding to a complete scan cycle in the transverse detection region of the film layer. The thickness value of each sampling point is first processed by Kalman filtering to eliminate measurement noise and retain true signal characteristics. During the calculation, the system first calculates the weighted average of these 60 points as the baseline thickness, with the weights determined according to the position of the sampling points in the film width direction. Next, the ratio of the absolute value of the thickness change rate of adjacent points to the average thickness is calculated, and the standard deviation of this set of data is calculated using the Welford online algorithm. All intermediate calculation results are in 32-bit floating-point format, and the final MUI value retains four significant digits.

[0053] In one possible implementation, the step of calculating the process stability index PSI in S3 includes:

[0054] S301: Obtain the calender roll temperature data for M consecutive time points in the current production batch, where M≥30;

[0055] S302: Calculate the process stability index (PSI) according to the following formula:

[0056]

[0057] Where M is the number of time points, which is dimensionless; The temperature of the calender roll at the j-th time point is expressed in degrees Celsius. This represents the average temperature over M time points, in degrees Celsius. This represents the historical standard deviation of the calender roll temperature, in degrees Celsius. This is the target set value for the temperature of the calender rolls, in degrees Celsius. This represents the allowable fluctuation range of the calender roll temperature, expressed in degrees Celsius.

[0058] The Process Stability Index (PSI) is calculated based on a calender roll temperature sequence over 30 consecutive time points, with data sourced from the production line's SCADA system's real-time database. During calculation, the system first compensates for the raw readings using the temperature sensor's calibration coefficients, then employs a sliding window approach to calculate the statistical characteristics of the temperature sequence. Historical standard deviation data is obtained from process parameter archives stored in the production line's MES system; the system automatically selects historical data for the same product specifications as a reference benchmark. The PSI calculation uses a double exponential smoothing method to handle the trend component of the temperature sequence, ensuring good sensitivity to both slow drifts and rapid fluctuations.

[0059] In one possible implementation, such as Figure 2 As shown, the quality early warning model in S4 is pre-trained through the following steps:

[0060] S401: Collect historical production data, including film thickness distribution data, process parameter data, and corresponding orange peel defect level data;

[0061] S402: Calculate the microuniformity index MUI based on historical film thickness data;

[0062] S403: Calculate the process stability index (PSI) based on historical process parameter data;

[0063] S404: Using MUI and PSI as input features and orange peel texture defect level as output label, a gradient boosting decision tree model is trained to obtain a quality early warning model.

[0064] The quality early warning model was trained using the TensorFlow framework. Training data came from historical production records in the factory's data platform, containing 8,000 complete sets of production process data. During data preprocessing, the system first performed outlier detection and missing value imputation on the raw thickness distribution data, then calculated the MUI and PSI features for each set. The feature engineering stage also generated derived features such as the first-order difference of the MUI and the sliding variance of the PSI. Model training employed a distributed computing architecture, running in parallel on four NVIDIA T4 GPUs, and the training process converged after approximately six hours.

[0065] In one possible implementation, the step of generating control commands in S5 includes:

[0066] S501: Determine the optimization direction based on the current MUI and PSI values ​​using the quality early warning model;

[0067] S502: Based on the optimization direction and the current process parameter status, calculate the adjustment amount of the calender roll temperature or linear speed. The adjustment amount must be verified to ensure that it does not exceed the safe operating range of the equipment.

[0068] S503: The adjustment amount is encapsulated into a standard control command format and issued. The control command is accompanied by a unique serial number. After the control system executes the command, it must return a confirmation signal.

[0069] The control command generation module employs a hybrid decision-making mechanism based on rules and models. Upon receiving MUI and PSI values, the system first uses a rule engine to make a preliminary judgment to determine the type of process parameter that needs adjustment. Then, it calls the API interface of the quality early warning model to obtain detailed optimization suggestions. The calculation of the adjustment amount comprehensively considers the current equipment status, process constraints, and quality objectives, and uses a constrained optimization algorithm to solve the problem. Each command is digitally signed immediately after generation and transmitted to the equipment controller via HTTPS protocol to ensure the integrity and security of the command.

[0070] In one possible implementation, such as Figure 3 As shown, the specific steps for adjusting the temperature or linear speed of the calender roll according to the control cycle in S5 include:

[0071] S511: When the warning level exceeds the preset threshold, the anti-oscillation judgment logic is activated. The final control command is only generated when the warning level exceeds the threshold for multiple consecutive calculation cycles or when the MUI and PSI indicators continue to deteriorate.

[0072] S512: After the control command is issued, wait for the system to stabilize for a predetermined time;

[0073] S513: Reacquire film thickness data and calculate MUI value. If the MUI value improves to a safe range, maintain the current parameters; otherwise, trigger further processing.

[0074] The anti-oscillation control logic is implemented using a finite state machine, comprising four states: "monitoring," "judgment," "execution," and "verification." The system maintains a warning level queue of length 10. Control commands are only generated when the proportion of high-risk warnings in the queue exceeds 70% and persists for three consecutive control cycles. After the control command is issued, the system enters a waiting state, during which the stability of process parameters is continuously monitored. The verification phase employs a statistical process control-based approach, evaluating the adjustment effect by calculating the moving range of the MUI and the process capability index.

[0075] In one possible implementation, when obtaining the film thickness measurement value in S201, an abnormal data processing step is also included:

[0076] S2011: When the thickness values ​​of multiple consecutive sampling points exceed the allowable range of the process, they are automatically marked as abnormal data segments;

[0077] S2012: Repair the abnormal data segment using an interpolation algorithm based on adjacent valid data, and record the time and location information of the abnormality.

[0078] Anomaly detection employs a density-based LOF (List of Elements) algorithm to identify outliers in thickness data. When consecutive outliers are detected, the system initiates a multi-level processing procedure: first, it attempts to repair the outliers using linear interpolation of neighboring valid data; if the repaired data still exceeds a reasonable range, the area is marked as requiring manual intervention. All anomaly handling records are written to the quality event log, including the anomaly type, handling method, and personnel information, forming a complete anomaly tracing chain.

[0079] In one possible implementation, when acquiring the calender roll temperature data in step S301, a data verification step is also included:

[0080] S3011: Check if the sensor readings are within the preset reasonable range;

[0081] S3012: Compare the temperature change rate between adjacent sampling points. If the change rate exceeds the set threshold, start the sensor diagnostic program.

[0082] The temperature data verification system implements a triple verification mechanism. The first level of verification is completed at the sensor end, eliminating obviously abnormal readings through hardware filters. The second level of verification is performed in the data acquisition module, using a physical model-based method to detect data points that do not conform to the laws of heat conduction. The third level of verification is executed at the application layer, identifying faulty sensors by comparing the consistency of readings from multiple adjacent sensors. When an abnormal data is detected, the system automatically switches to a backup sensor or activates a data reconstruction algorithm.

[0083] In one possible implementation, when training the gradient boosting decision tree model in S404, a five-fold cross-validation method is used, the training dataset is divided into a training set, a validation set and a test set in proportion, and an early stopping method is used to prevent overfitting.

[0084] Gradient boosting decision tree training is implemented using the XGBoost algorithm, and hyperparameter optimization is performed using Bayesian optimization. During training, the system monitors the loss function curves of the training and validation sets in real time, automatically adjusting the regularization parameters when overfitting is detected. Model evaluation employs hierarchical cross-validation to ensure that each defect level is sufficiently representative in both the training and validation sets. The final model is comprehensively evaluated using AUC-ROC curves and precision-recall curves, selecting the version that achieves the best balance between recall and precision.

[0085] In one possible implementation, the online learning mechanism in S7 specifically includes:

[0086] S701: Periodically and automatically initiates the model update process, collecting new production data to form an incremental training set;

[0087] S702: During the update process, the model retains the original network structure and fine-tunes the weight parameters;

[0088] S703: After each update, the model performance is automatically tested on the validation set. If the performance drops below a set threshold, the system rolls back to the previous version.

[0089] The online learning system employs a canary deployment strategy, with new models initially tested on a single production line. During the trial run, the system runs both the old and new models in parallel, comparing their prediction results with actual quality data. Model performance evaluation utilizes a dynamic threshold mechanism, comprehensively considering multiple dimensions such as accuracy, response time, and resource consumption. Only when the new model outperforms the old model across all evaluation metrics will it be gradually rolled out to other production lines. Each model update generates a complete change report, including performance comparison data, parameter change records, and rollback plans.

[0090] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0091] (1) In this invention, by deploying a high-frequency online detection system and combining it with edge computing devices to calculate the micro-uniformity index and process stability index in real time, the micro-fluctuations in film quality and instantaneous anomalies in process parameters are captured and quantitatively evaluated instantly. This technology transforms the previously lagging and passive manual quality inspection into an advanced and proactive online intelligent diagnosis, which can issue risk warnings at the early stage of precision texture defects. This not only significantly reduces the scrap rate and quality loss caused by batch micro-defects, but also effectively improves the ability to ensure the quality of high-end products, making the quality control of the production process more refined and forward-looking.

[0092] (2) In this invention, a rapid response loop from quality risk identification to automatic adjustment of process parameters is established by constructing an early warning model based on a real-time quality index and automatically generating closed-loop control commands. This system can accurately calculate the optimal adjustment amount of the calender roll temperature or linear speed according to the early warning level, and ensure control stability through anti-oscillation logic. This solves the pain points of traditional control methods being insensitive to micro-defects and having lag in adjustment, realizing proactive intervention and adaptive optimization of the production process, thereby significantly improving the micro-uniformity and overall quality stability of the film product while maintaining the continuity of the production rhythm.

[0093] (3) In this invention, a data traceability system is constructed by associating and storing production data, control instructions, and quality indicators throughout the entire process, and establishing a complete quality control file for each batch. Simultaneously, the online learning mechanism employed by the quality early warning model continuously absorbs new production data to optimize itself, maintaining the long-term reliability of early warning accuracy. This combination not only ensures that every process adjustment and its quality results are traceable, greatly facilitating the root cause analysis of quality problems, but also enables the entire system to continuously evolve and adapt to complex working conditions, reducing reliance on external manual adjustments and enhancing the autonomy and durability of the intelligent manufacturing system.

[0094] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0095] The following points need to be explained:

[0096] (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0097] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0098] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0099] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for data collection and traceability in the production process of PVC decorative film, characterized in that, include: S1: Through an online detection system deployed after the calendering process, the film thickness distribution data during the production process is collected in real time at a sampling frequency of not less than 100Hz, while the process parameters of calendering roll temperature, linear speed and cooling rate are also collected. S2: Based on the film thickness distribution data, calculate the microuniformity index MUI on the edge computing device of the production line. The microuniformity index MUI is used to quantify the thickness fluctuation characteristics of the film at the microscale. S3: Based on the calender roll temperature data, calculate the process stability index PSI on the edge computing device. The process stability index PSI is used to quantify the fluctuation characteristics of temperature parameters in the calendering process. S4: Input the microuniformity index MUI and process stability index PSI into the pre-trained quality early warning model to obtain the orange peel texture defect risk warning level; The orange peel texture defect refers to microscopic irregular patterns with a wavelength in the range of 0.1-1.0 mm that appear on the surface of the film layer; S5: When the warning level exceeds the preset threshold, a control command is automatically generated and sent to the calendering equipment control system to adjust the calendering roll temperature or linear speed according to the control cycle. S6: Link and store each control command, process parameter adjustment record, real-time collected film thickness distribution data and corresponding MUI and PSI values ​​to establish a complete quality control file for each production batch; S7: The quality early warning model adopts an online learning mechanism and regularly updates and optimizes the model parameters using new production data; The calculation of the microuniformity index (MUI) and the process stability index (PSI) ensures that the real-time response time is less than 100 milliseconds.

2. The method for data collection and traceability in the production process of PVC decorative film according to claim 1, characterized in that, The steps for calculating the micro-uniformity index MUI in S2 include: S201: Obtain the film thickness measurement values ​​of N consecutive sampling points in the detection area after the calendering process, where N≥50; S202: Calculate the micro-uniformity index MUI according to the following formula: Where N is the number of consecutive sampling points, which is dimensionless; The thickness of the film at the i-th sampling point is expressed in micrometers. σ is the average film thickness at N sampling points, in micrometers; σ is the standard deviation of the thickness at N sampling points, in micrometers.

3. The method for data collection and traceability of the production process of PVC decorative film according to claim 1, characterized in that, The step of calculating the process stability index PSI in S3 includes: S301: Obtain the calender roll temperature data for M consecutive time points in the current production batch, where M≥30; S302: Calculate the process stability index (PSI) according to the following formula: Where M is the number of time points, which is dimensionless; The temperature of the calender roll at the j-th time point is expressed in degrees Celsius. This represents the average temperature over M time points, in degrees Celsius. This represents the historical standard deviation of the calender roll temperature, in degrees Celsius. This is the target set value for the temperature of the calender rolls, in degrees Celsius. This represents the allowable fluctuation range of the calender roll temperature, expressed in degrees Celsius.

4. The method for data collection and traceability of the production process of PVC decorative film according to claim 1, characterized in that, The quality early warning model in S4 is pre-trained through the following steps: S401: Collect historical production data, including film thickness distribution data, process parameter data, and corresponding orange peel defect level data; S402: Calculate the micro uniformity index MUI based on the historical film thickness data; S403: Calculate the process stability index (PSI) based on the historical process parameter data; S404: Using MUI and PSI as input features and orange peel texture defect level as output label, a gradient boosting decision tree model is trained to obtain a quality early warning model.

5. The method for data collection and traceability in the production process of PVC decorative film according to claim 1, characterized in that, The step of generating control commands in S5 includes: S501: Determine the optimization direction based on the current MUI and PSI values ​​using the quality early warning model; S502: Based on the optimization direction and the current process parameter status, calculate the adjustment amount of the calender roll temperature or linear speed. The adjustment amount must be verified to ensure that it does not exceed the safe operating range of the equipment. S503: The adjustment amount is encapsulated into a standard control command format and issued. The control command is accompanied by a unique serial number. After the control system executes the command, it needs to return a confirmation signal.

6. The method for data collection and traceability of the production process of PVC decorative film according to claim 1, characterized in that, The specific steps in S5 for adjusting the temperature or linear speed of the calendering roll according to the control cycle include: S511: When the warning level exceeds the preset threshold, the anti-oscillation judgment logic is activated. The final control command is only generated when the warning level exceeds the threshold for multiple consecutive calculation cycles or when the MUI and PSI indicators continue to deteriorate. S512: After the control command is issued, wait for the system to stabilize for a predetermined time; S513: Reacquire film thickness data and calculate MUI value. If the MUI value improves to a safe range, maintain the current parameters; otherwise, trigger further processing.

7. The method for data collection and traceability in the production process of PVC decorative film according to claim 2, characterized in that, When obtaining the film thickness measurement value in step S201, an abnormal data processing step is also included: S2011: When the thickness values ​​of multiple consecutive sampling points exceed the allowable range of the process, they are automatically marked as abnormal data segments; S2012: Repair the abnormal data segment using an interpolation algorithm based on adjacent valid data, and record the time and location information of the abnormality.

8. The method for data collection and traceability in the production process of PVC decorative film according to claim 3, characterized in that, When acquiring the calender roll temperature data in step S301, a data verification step is also included: S3011: Check if the sensor readings are within the preset reasonable range; S3012: Compare the temperature change rate between adjacent sampling points. If the change rate exceeds the set threshold, start the sensor diagnostic program.

9. The method for data collection and traceability in the production process of PVC decorative film according to claim 4, characterized in that: In the S404 method, when training the gradient boosting decision tree model, a five-fold cross-validation method is used. The training dataset is divided into training set, validation set and test set according to the proportion, and overfitting is prevented by early stopping.

10. The method for data collection and traceability in the production process of PVC decorative film according to claim 1, characterized in that, The online learning mechanism in S7 specifically includes: S701: Periodically and automatically initiates the model update process, collecting new production data to form an incremental training set; S702: During the update process, the model retains the original network structure and fine-tunes the weight parameters; S703: After each update, the model performance is automatically tested on the validation set. If the performance drops below a set threshold, the system rolls back to the previous version.

Images