A product tracking method for a plastic flower digital workshop

Abstract

The present application provides a product tracking method for a digital workshop for plastic flowers, comprising: obtaining process parameter data of a screw speed curve, a barrel temperature curve, and an injection pressure curve in the production process of products with uneven gradient color bands on plastic flower leaves, and preprocessing the process parameter data to obtain a standardized parameter data set; obtaining quality data of defect types and defect degrees of products with color band defects, associating the color band defect data with corresponding parameter fluctuation feature vectors, and constructing a mapping relationship data set between color band defects and parameter fluctuations; training the mapping relationship data set using a decision tree algorithm to generate a discriminant model for color band defects and parameter fluctuations, and predicting the possibility and degree of color band defects based on the parameter fluctuation feature vectors; and adjusting process parameters if an early warning signal is triggered, the process parameters including screw speed, barrel temperature, and injection pressure are controlled within a preset range, thereby reducing the occurrence of color band defects.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a product tracking method for a plastic flower digital workshop. Background Art

[0002] The uneven gradient color banding on plastic flower leaves is caused by fluctuations in process parameters during the injection molding process. To analyze this issue, we need to focus on three key parameters: the screw speed curve, the barrel temperature curve, and the injection pressure curve. Screw speed affects the shearing and mixing of the plastic melt in the barrel, and speed fluctuations can lead to poor melt uniformity. Barrel temperature determines the degree of plastic melting and fluidity, and fluctuations in the temperature curve can cause unstable melt viscosity. Injection pressure affects the speed and pressure distribution of the melt filling the mold cavity. Variations in the pressure curve can cause uneven filling of the part. To identify the corresponding relationship between parameter fluctuations and color banding defects, it is necessary to carefully record each parameter curve in actual production and sample and analyze defective parts. Through data comparison and statistical analysis, the inherent connections between them can be revealed, providing a basis for precise control of process parameters. Summary of the Invention

[0003] The present invention provides a product tracking method for a plastic flower digital workshop, which mainly includes:

[0004] The process parameter data of the screw speed curve, barrel temperature curve and injection pressure curve in the production process of the plastic flower leaf gradient color uneven product are obtained, and the process parameter data are preprocessed to obtain a standardized parameter data set; the standardized parameter data set is analyzed by a time series analysis algorithm, and the screw speed curve, barrel temperature curve and injection pressure curve are feature extracted to obtain key indicators reflecting the parameter fluctuation characteristics, and a parameter fluctuation feature vector is constructed based on the key indicators; the defect type and defect degree quality data of the color ribbon defective product are obtained, the color ribbon defect data is associated with the corresponding parameter fluctuation feature vector, and a mapping relationship data set between the color ribbon defect and the parameter fluctuation is constructed; the decision tree algorithm is used to train the mapping relationship data set to generate a discriminant model for color ribbon defects and parameter fluctuations. , predict the possibility and degree of ribbon defects based on the parameter fluctuation characteristic vector; in the production process of plastic flower leaf gradient ribbon products, the process parameter data of screw speed curve, barrel temperature curve and injection pressure curve are collected in real time, and the collected data are preprocessed and feature extracted to obtain the parameter fluctuation characteristic vector under the current production state; the parameter fluctuation characteristic vector under the current production state is input into the discrimination model of ribbon defects and parameter fluctuations, and the possibility and degree of ribbon defects are predicted in real time. When the predicted defect possibility or defect degree exceeds the preset threshold, an early warning signal is triggered, indicating that the process parameters are abnormal; if the early warning signal is triggered, the process parameters are adjusted, and the process parameters include screw speed, barrel temperature and injection pressure, and the process parameters are controlled within the preset range, thereby reducing the occurrence of ribbon defects.

[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0006] The present invention discloses a product tracking method for a digital workshop for plastic flowers. In order to solve the quality problem of uneven gradient color bands in the production process of plastic flower leaves, the process parameter data such as screw speed, barrel temperature and injection pressure are collected, and the data are pre-processed and time series analyzed to extract parameter fluctuation characteristics. A mapping relationship is established between these characteristics and the type and degree of color band defects, and a decision tree algorithm is used to train a discriminant model. During the production process, process parameters are collected and features are extracted in real time, and the features are input into the discriminant model to predict the possibility and degree of color band defects. When the prediction result exceeds the preset threshold, an early warning signal is triggered and the process parameters are guided to be adjusted, thereby achieving the prevention and control of color band defects. The present invention realizes real-time monitoring and early warning of product quality by establishing a correlation model between process parameter fluctuations and color band defects, effectively reducing the color band defect rate in the production of plastic flower leaves. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a flowchart of a product tracking method for a plastic flower digital workshop.

[0008] Figure 2 This is a schematic diagram of a product tracking method for a plastic flower digital workshop according to the present invention.

[0009] Figure 3 This is another schematic diagram of a product tracking method for a plastic flower digital workshop according to the present invention. DETAILED DESCRIPTION

[0010] The following will describe the technical solutions in the embodiments of the present invention in detail with reference to the accompanying drawings. The described embodiments are only a part of the embodiments of the present invention.

[0011] like Figure 1-3 In this embodiment, a product tracking method for a plastic flower digital workshop may specifically include:

[0012] Step S101 , obtaining process parameter data of a screw speed curve, a barrel temperature curve, and an injection pressure curve in a production process of a plastic flower leaf product with uneven gradient color bands, and preprocessing the process parameter data to obtain a standardized parameter data set.

[0013] A multi-channel data acquisition board is used to obtain the screw speed signal, barrel temperature signal and injection pressure signal of the injection molding machine, and the signals are filtered by a Butterworth low-pass filter to obtain filtered data; according to the inflection point distribution of the parameter curve in the filtered data, the parameter curve is fitted by a cubic spline interpolation function to obtain characteristic data points; normalization calculation is performed on the characteristic data points, and the normalized data is fitted by a polynomial regression equation, and abnormal values are eliminated according to the mean square error criterion to obtain a characteristic vector sequence; the barrel temperature field distribution data, injection pressure change data and screw speed data in the characteristic vector sequence are used to establish a multivariate linear regression equation, and the combined parameter data set of injection pressure and screw speed at different barrel temperatures is obtained by traversing the parameter space.

[0014] Specifically, a 16-bit precision multi-channel data acquisition board was used to collect screw speed signals, barrel heating zone temperature signals, and injection pressure signals from the injection molding machine's electrical control cabinet. The sampling frequency was set to 100 Hz, and the sampling interval was 10 milliseconds. The collected data was filtered using a 6th-order Butterworth low-pass filter at a frequency of 20 Hz. The filtered data was stored in a database to generate a time series curve. Based on the quality requirements for the gradient color band of plastic flower leaves, the process parameter curve was read from the database. The inflection point was defined when the slope change between the local maximum and minimum points in the curve exceeded 0.5. A cubic spline interpolation function was used to fit the trend of each parameter. Ten evenly distributed characteristic data points were extracted from the fitted curve to construct a feature matrix. The process parameter data in the feature matrix were normalized, with the normalization interval set between [0, 1]. A 5th-order polynomial regression equation was used to fit the parameter trend. Data points with a parameter change exceeding 0.3 were identified as outliers and removed using the minimum mean square error criterion. After removing the outliers, the data formed a feature vector sequence. From the feature vector sequence, barrel temperature field distribution data, injection pressure variation data, and screw speed data are selected as input variables for the multivariate linear regression equation. A mapping relationship between the three process parameters is established. By traversing the parameter space, the optimal combination of injection pressure and screw speed at different barrel temperatures is obtained, and a standardized parameter data set is constructed. When the multi-channel data acquisition board collects data from the injection molding machine, setting a 16-bit sampling accuracy means that the signal quantization resolution reaches 65,536 discrete values, enabling high-precision digital acquisition of screw speed signals from 0 to 300 rpm, barrel temperature signals from 0 to 400 degrees Celsius, and injection pressure signals from 0 to 200 MPa. A 100Hz sampling frequency corresponds to 100 data points per second, meeting the needs of monitoring rapid parameter changes during the injection molding process. A 6th-order Butterworth low-pass filter can filter out high-frequency interference signals above 20Hz, preserving the effective variation characteristics of the injection molding process parameters. During the production process of plastic flower leaves, the screw speed curve shows a step-like change feature, the injection pressure curve shows a bell-shaped feature that first increases and then decreases, and the barrel temperature curve shows a slow fluctuation feature. When the slope change between adjacent data points exceeds 0.5, it is judged to be an inflection point. The inflection point of the screw speed curve corresponds to a speed change range of 50 to 150 revolutions per minute, the inflection point of the injection pressure curve corresponds to a pressure change range of 80 to 120 MPa, and the inflection point of the barrel temperature curve corresponds to a temperature change range of 220 to 260 degrees Celsius.When normalizing the process parameter data, the maximum screw speed of 300 revolutions per minute is used as the benchmark value, the maximum injection pressure of 200 MPa is used as the benchmark value, and the maximum barrel temperature of 400 degrees Celsius is used as the benchmark value. The normalized data are distributed in the range of 0 to 1. After fitting the parameter change trend with a 5th-order polynomial regression equation, data points with a screw speed change of more than 0.3 corresponding to an actual speed change of 90 revolutions per minute, a pressure change of more than 0.3 corresponding to an actual pressure change of 60 MPa, and a temperature change of more than 0.3 corresponding to an actual temperature change of 120 degrees Celsius are judged as abnormal values. When the multivariate linear regression equation is used to establish the mapping relationship of process parameters, the barrel temperature is used as the independent variable, and a temperature point is taken every 5 degrees in the range of 220 to 260 degrees Celsius. The injection pressure at each temperature point is taken every 5 MPa in the range of 80 to 120 MPa, and the screw speed is taken every 10 revolutions per minute in the range of 50 to 150 revolutions per minute. By traversing these parameter combinations, the optimal matching value of injection pressure and screw speed at different barrel temperatures is obtained, thereby establishing a standardized process parameter data set.

[0015] In step S102 , the normalized parameter data set is analyzed using a time series analysis algorithm, and features are extracted from the screw speed curve, barrel temperature curve, and injection pressure curve to obtain key indicators reflecting parameter fluctuation characteristics, and a parameter fluctuation feature vector is constructed based on the key indicators.

[0016] A time series is constructed according to a sampling interval based on a normalized parameter data set, and the mean fluctuation value and variance fluctuation value of the sequence are obtained through a stationarity test; the autocorrelation function value of the delay point of the time series is calculated, and the spectrum amplitude is obtained from the autocorrelation function value using a fast Fourier transform, and the main frequency point is selected based on the spectrum amplitude; the frequency energy characteristic value is obtained from the main frequency point, and data segments are extracted for the parameter curve, and statistical indicator data is obtained using a multivariate linear regression; the statistical indicator data and the frequency energy characteristic value are normalized, and feature quantities are extracted from the normalized data, and the parameter fluctuation feature vector is obtained in the order of parameter types.

[0017] Specifically, a time series was constructed for the normalized parameter data set at a sampling interval of 10 milliseconds. A stationarity test was performed to determine that the mean fluctuation of the sequence was less than 0.1 and the variance fluctuation was less than 0.05. The autocorrelation function values of the delay points 0 to 100 were calculated for the sequences that passed the test, and a data series reflecting the degree of parameter correlation between the screw speed curve, barrel temperature curve, and injection pressure curve at different time scales was obtained. Based on the autocorrelation data series, a fast Fourier transform was performed at a sampling frequency of 100 Hz. The spectrum amplitude was extracted within the frequency range of 0 to 50 Hz, and the top three main frequency points were selected by amplitude sorting. From the main frequency points, the frequency points with amplitudes exceeding twice the average amplitude were selected as the main fluctuation frequency features. The spectrum amplitudes were accumulated within 0.5 Hz of each main frequency point to obtain the frequency energy features. Data segments of 1000 data points, with adjacent segments overlapping by 500 points, were selected from the three parameter curves. Within each data segment, four statistical quantities (mean, standard deviation, skewness, and kurtosis) were extracted. Multiple linear regression was used to calculate the linear relationship between the mean, standard deviation, skewness, and kurtosis of the parameters within the data segment, yielding statistical indicators reflecting the parameter fluctuation characteristics. The extracted frequency characteristics and statistical indicator data were divided into three groups based on parameter type: speed, temperature, and pressure. Each data group was normalized using maximum and minimum values. Six characteristic quantities (dominant frequency, dominant energy, mean, standard deviation, skewness, and kurtosis) were extracted from the normalized data. These were organized into an 18-dimensional parameter fluctuation feature vector, ordered by parameter type. In time series data processing, a 10-millisecond sampling interval corresponds to a 100-Hz sampling frequency. Within the operating ranges of the injection molding machine screw speed (0 to 300 rpm), barrel temperature (0 to 400°C), and injection pressure (0 to 200 MPa), a mean fluctuation of less than 0.1 indicates that the parameter variation is within 10%, and a variance fluctuation of less than 0.05 indicates that the data dispersion is within 5%. The autocorrelation function is used to calculate the parameter correlation within the time range of 0 to 1 second for 0 to 100 delay points. In frequency domain analysis, a fast Fourier transform is performed at a sampling frequency of 100 Hz. Based on the Nyquist sampling theorem, the parameter fluctuation frequency within the range of 0 to 50 Hz can be analyzed. The screw speed curve exhibits periodic fluctuations within the range of 2 to 5 Hz, the barrel temperature curve varies slowly within the range of 0.1 to 0.5 Hz, and the injection pressure curve fluctuates rapidly within the range of 10 to 20 Hz. When the amplitude of a frequency point exceeds twice the average amplitude, it is determined to be the primary fluctuation frequency. For a data segment with 1000 data points corresponding to a time length of 10 seconds, adjacent data segments overlap by 500 points to achieve a smooth transition of 5 seconds of data. Among the statistics extracted within the data segment, the mean reflects the average level of the parameter, the standard deviation represents the fluctuation amplitude of the parameter, the skewness describes the asymmetry of the parameter distribution, and the kurtosis reflects the sharpness of the parameter distribution. The quantitative relationship between the statistics is established through multiple linear regression.The parameter fluctuation characteristic vector contains 18 characteristic quantities, among which the frequency value of the main frequency point reflects the periodic characteristics of the parameter fluctuation, the energy value of the main frequency point characterizes the intensity of the parameter fluctuation, the mean reflects the stability level of the parameter, the standard deviation characterizes the fluctuation range of the parameter, the skewness describes the inclination direction of the parameter distribution, and the kurtosis reflects the concentration degree of the parameter distribution. The screw speed characteristic quantity is in 1 to 6 dimensions, the temperature characteristic quantity is in 7 to 12 dimensions, and the pressure characteristic quantity is in 13 to 18 dimensions. The characteristic quantity distribution is in the range of 0 to 1 through maximum and minimum value normalization.

[0018] Step S103 , obtaining the defect type and defect degree quality data of the ribbon defect product, associating the ribbon defect data with the corresponding parameter fluctuation feature vector, and constructing a mapping relationship data set between the ribbon defect and the parameter fluctuation.

[0019] Acquire an image of a product with color ribbon defects, obtain a denoised image through Gaussian noise reduction processing, calculate the defect area ratio, morphological parameters and color features based on the denoised image, and obtain quality feature data; add the production batch number and timestamp mark to the quality feature data to construct a defect feature record; extract the feature vectors of adjacent time points from the parameter fluctuation feature vector database according to the timestamp, and perform a weighted average operation on the feature vectors using a weight coefficient to obtain the comprehensive feature vector; if the comprehensive feature vector meets the linear classifier mapping condition, establish a mapping relationship between the comprehensive feature vector and the defect type; generate an associated data record based on the mapping relationship and the defect feature record.

[0020] Specifically, images of defective ribbon products with a resolution of 2048×1536 pixels and a frame rate of 30 frames per second were acquired from the production line's image acquisition equipment. After Gaussian noise reduction, the images were converted into 256-level grayscale images. Region segmentation was performed by setting a grayscale value difference threshold of 8 as the growth criterion. The number of pixels within the segmented region was calculated to obtain the defect area ratio. The grayscale mean and standard deviation were extracted from the defective region to construct a defect severity index. Edges were extracted based on the grayscale threshold of the defective region. The edge perimeter, area ratio, roundness, and aspect ratio were calculated as morphological feature parameters. A histogram of 16 grayscale levels within the ribbon defect region was constructed. The mean, variance, skewness, and kurtosis of the histogram were calculated as color distribution features. A brightness difference score was calculated compared to the standard ribbon image. The product production batch number and production timestamp were added to the defect data to construct quality feature data including defect area, morphological parameters, and color features. Based on the production timestamp of the defective product, five vectors before and after the corresponding time point were extracted from the parameter fluctuation feature vector database. Weight coefficients were set according to the time interval, with a weight ratio of 0.8 for adjacent vectors. These 10 feature vectors were weighted averaged to obtain an 18-dimensional comprehensive feature vector. A time-series correspondence was established between the quality feature data and the comprehensive feature vector. A linear classifier was used to establish a mapping relationship between the parameter fluctuation feature vector and the color ribbon defect type. A multivariate regression equation was used to establish a mapping relationship between the feature vector and the defect severity index. These two mapping relationships were combined with the quality feature data to form associated data records, which were stored in chronological order to form a data set that maps color ribbon defects to parameter fluctuations. During image acquisition, a 2048×1536 pixel resolution corresponds to an actual ribbon size of 200×150 mm, with each pixel corresponding to 0.1 mm of actual length. A sampling frequency of 30 frames per second (fps) meets the production line's operating speed of 2 meters per second. Gaussian noise reduction uses a 5×5 pixel window to eliminate image noise. The 256 grayscale levels are sufficient to capture subtle differences in ribbon gradients, and a grayscale difference threshold of 8 corresponds to a 3% brightness change. The defect area percentage is calculated by dividing the number of pixels by the total number of pixels in the image. For defect area feature extraction, the edge perimeter reflects the complexity of the defect outline, the area ratio indicates the proportion of the defect area to the entire ribbon area, the circularity describes the degree of the defect shape's proximity to a circle, and the aspect ratio reflects the direction of the defect's extension. The 16 grayscale levels correspond to 16 histogram counting intervals, with grayscale values from 0 to 255 counted every 16th value. The histogram features reflect the color distribution pattern within the defect area, and the production batch number and timestamp uniquely identify the defect data. In the temporal correlation of the parameter fluctuation feature vectors, the five vectors before and after the defect occurrence time correspond to an actual time span of 100 milliseconds. The weight ratio of 0.8 between adjacent vectors reflects the characteristic that the time correlation decays with increasing distance. The 18-dimensional comprehensive feature vector contains 6 feature quantities each of the speed group, temperature group, and pressure group. The weighted average maintains the dimensional consistency of the feature quantities of each group.In the mapping relationship dataset, the linear classifier maps the parameter fluctuation feature vector to three defect types: uneven ribbon, broken ribbon, and deformed ribbon. The multivariate regression equation maps the feature vector to a defect severity index ranging from 0 to 1. In the defect severity index, 0 indicates no defect and 1 indicates the most serious defect. The associated data records contain five data items: production time, defect type, defect severity, feature vector, and quality characteristics, which are stored in ascending order of timestamps to form a dataset.

[0021] Step S104 , using a decision tree algorithm to train the mapping relationship data set, generating a discriminant model between color ribbon defects and parameter fluctuations, and predicting the possibility and degree of color ribbon defects based on the parameter fluctuation feature vector.

[0022] Obtain defect type labels, defect degree indicators and parameter fluctuation feature vectors sorted by production time in the mapping relationship data set, perform undersampling processing based on the feature vectors to obtain a training data matrix and a verification data matrix; calculate the Gini coefficient for the parameter fluctuation feature vector, if the Gini coefficient is greater than the threshold, determine the corresponding feature quantity as a candidate split feature, and use the quantile to calculate the sample entropy of the candidate split feature to obtain the optimal split node; perform grid search on the decision tree parameters based on the training data matrix, calculate the mean square error through the verification data matrix, and determine the parameter combination corresponding to the minimum mean square error to obtain the optimized decision tree parameters; use the optimized decision tree parameters to train a decision tree model, calculate the leaf node probability distribution for the decision tree model, and determine the defect type and defect degree values with the highest confidence through the leaf node probability distribution.

[0023] Specifically, defect type labels, defect severity indicators, and parameter fluctuation feature vectors were read from a mapping dataset in chronological order. Undersampling was performed to equalize the number of samples for each defect type. After data standardization, the training and validation data were divided into training and validation data in an 8:2 ratio. All feature vectors were normalized to a value distribution between 0 and 1, and training and validation data matrices were constructed. Based on the distribution pattern of the parameter fluctuation feature vectors, the Gini coefficient of each feature was calculated. Features with a Gini coefficient greater than 0.3 were selected as candidate split features. Ten quantiles were used as candidate split thresholds. For each candidate split point, the sample entropy of the left and right subtrees was calculated. The feature and threshold combination with the largest entropy gain was selected as the splitting node, generating a decision tree branch that describes the parameter fluctuation pattern. A grid search was used to optimize the decision tree parameters, searching for parameter combinations with a maximum tree depth between 3 and 8 and a minimum number of sample splits between 10 and 50. The prediction error of the validation data for each parameter set was calculated using mean squared error as the evaluation function. The parameter combination with the smallest error was selected to train the decision tree, establishing the defect type discrimination rule and the defect severity prediction rule, respectively. For the trained decision tree, the probability distribution and numerical prediction results of the categories of all leaf nodes are calculated. The real-time parameter fluctuation feature vector is input into the decision tree. The path of the feature vector to each leaf node is obtained by traversing the decision tree nodes. The confidence level of each leaf node is calculated, and the defect type with the highest confidence level and the corresponding defect severity value are output. The mapping relationship dataset contains three typical defect types: uneven ribbon, broken ribbon, and deformed ribbon. The original number of samples for each defect type is 1000, 600, and 400, respectively. Through undersampling, the number of samples for each type is adjusted to 400, totaling 1200 training samples. The data is divided into 960 training data and 240 validation data in an 8:2 ratio. The 18-dimensional feature vector is normalized so that the three groups of feature values for screw speed, barrel temperature, and injection pressure are distributed between 0 and 1. The Gini coefficient reflects the degree of uneven data distribution of feature quantities. In the parameter fluctuation feature vector, the Gini coefficient of the screw speed fluctuation frequency feature is 0.42, the Gini coefficient of the fluctuation energy feature is 0.38, the mean feature Gini coefficient of the barrel temperature is 0.35, the standard deviation feature Gini coefficient is 0.33, the kurtosis feature Gini coefficient of the injection pressure is 0.36, and the skewness feature Gini coefficient is 0.32. These feature quantities are selected as candidate split features, and 9 candidate split thresholds are set for each feature according to the quantiles of 10%, 20%, 30% to 90%.During decision tree parameter optimization, the maximum tree depth increased from 3 to 8 layers, and the minimum number of sample splits increased from 10 to 50, resulting in a total of 30 parameter combinations (6×5). Taking color band unevenness defects as an example, a tree depth of 4 layers and a minimum number of splits of 20 achieved the lowest mean squared error (MSE) of the validation data prediction, with a prediction accuracy of 85%. The corresponding decision tree contains 15 split nodes and 16 leaf nodes, each storing the probability distribution of the defect type and the predicted defect severity. During real-time prediction, the parameter fluctuation eigenvector is propagated down the decision tree's split nodes layer by layer. When the eigenvalue is less than the split threshold, the left subtree is selected; when it is greater than the split threshold, the right subtree is selected. Finally, a leaf node is reached. The category with the highest probability of the defect type in that leaf node is used as the prediction result, and the predicted defect severity value for that leaf node is also output. The confidence level of the prediction result is determined by the proportion of the number of samples in that leaf node to the total number of samples.

[0024] Step S105 , during the production process of the plastic flower leaf gradient color ribbon product, process parameter data of the screw speed curve, barrel temperature curve, and injection pressure curve are collected in real time, and the collected data are preprocessed and feature extracted to obtain the parameter fluctuation feature vector under the current production state.

[0025] A multi-channel data acquisition card is used to acquire the screw speed signal, barrel temperature signal and injection pressure signal of the injection molding machine control cabinet, and a data record containing a timestamp, a channel number and a numerical value is generated according to the signal. The data record is stored in a circular buffer area according to the first-in-first-out principle; a data block is read from the circular buffer area, the data block is filtered and denoised by a Butterworth low-pass filter, and the data block is completed by cubic spline interpolation to obtain a continuous parameter curve; a power spectral density analysis is performed on the continuous parameter curve, and the main frequency point eigenvalues and statistical eigenvalues of the continuous parameter curve are extracted to obtain a characteristic data sequence; for the characteristic data sequence, the eigenvalues of the screw speed group, the barrel temperature group and the injection pressure group are extracted, and if the eigenvalue fluctuation amplitude exceeds a preset threshold, the eigenvalue is organized into a parameter fluctuation eigenvector.

[0026] Specifically, a 16-bit precision multi-channel data acquisition card was used to collect screw speed signals, barrel temperature signals, and injection pressure signals from the injection molding machine control cabinet. The sampling frequency was set to 100 Hz and the sampling accuracy was 0.01. A synchronous clock signal was used to control the simultaneous sampling of the three channels. The collected data was organized into data records based on timestamp, channel number, and value. Every 1000 records formed a data block, which was stored in a circular buffer on a first-in, first-out basis. The buffer had a capacity of 10 data blocks. The latest data block was read from the circular buffer, and the three parameter signals were filtered and denoised using a 6th-order Butterworth low-pass filter with a cutoff frequency of 20 Hz. The data were segmented into fixed-length segments of 1000 data points, and cubic spline interpolation was used to fill in missing data points. A continuous parametric curve was constructed with 500 overlapping points between data segments. Zero-mean normalization was applied to the curve to generate a normalized parametric curve. A power spectral density analysis was performed on the normalized parameter curve. The three frequency points with the largest amplitudes within the 0 to 50 Hz frequency range were selected as the dominant frequency points. The power and phase values corresponding to the dominant frequency points were calculated. The mean, standard deviation, maximum, minimum, and peak-to-valley ratio of the curve were also calculated as statistics. The eigenvalues were updated and calculated every 100 data points to generate a continuous feature data sequence. For this feature data sequence, the power spectral eigenvalues and statistical eigenvalues of the screw speed group, barrel temperature group, and injection pressure group were extracted. All eigenvalues were normalized to their maximum and minimum values. From each feature group, the eigenvalues with a fluctuation amplitude exceeding 0.1 were selected and organized into an 18-dimensional parameter fluctuation feature vector in the order of speed, temperature, and pressure. During data acquisition, 16-bit precision corresponds to 65,536 quantization levels, enabling the acquisition of screw speed signals from 0 to 300 rpm with an accuracy of 0.01 rpm, barrel temperature signals from 0 to 400°C with an accuracy of 0.01°C, and injection pressure signals from 0 to 200 MPa with an accuracy of 0.01 MPa. A 100 Hz sampling frequency corresponds to a 10 millisecond sampling interval, meeting the requirements for monitoring rapidly changing injection molding parameters. A data block of 1,000 records corresponds to 10 seconds of real-time data. During data preprocessing, a 6th-order Butterworth low-pass filter with a flat passband response below a cutoff frequency of 20 Hz filters out high-frequency noise while retaining the key features of parameter changes. A 1,000-point data segment corresponds to 10 seconds of real-time time, and 500-point overlap between adjacent data segments ensures smooth transitions in feature extraction. Cubic spline interpolation uses the curvature information of the four preceding and following data points to estimate missing values, and zero-mean normalization eliminates the impact of differences in parameter base values. Power spectrum density analysis shows that the screw speed signal fluctuates periodically in the frequency range of 2 to 5 Hz. The main frequency point corresponds to the fundamental frequency and harmonic components of the screw rotation. The barrel temperature signal changes slowly in the range of 0.1 to 0.5 Hz, corresponding to the response characteristics of the heating control. The injection pressure signal fluctuates rapidly in the range of 10 to 20 Hz, reflecting the dynamic change of the melt pressure in the cylinder.The feature vector contains 18 dimensions, among which the screw speed group feature contains 3 main frequency power values and 5 statistics, the barrel temperature group feature contains 3 main frequency power values and 5 statistics, and the injection pressure group feature contains 3 main frequency power values and 5 statistics. The feature value with a fluctuation amplitude exceeding 0.1 indicates that the parameter has changed significantly during the observation time. Through maximum and minimum value normalization, the numerical values of all feature values are distributed in the range of 0 to 1, which is convenient for subsequent defect prediction and judgment.

[0027] In step S106, the parameter fluctuation feature vector under the current production status is input into the discrimination model of ribbon defects and parameter fluctuations, and the possibility and degree of ribbon defects are predicted in real time. When the predicted defect possibility or defect degree exceeds the preset threshold, an early warning signal is triggered to indicate that the process parameters are abnormal.

[0028] A periodic data stream transmission method is used to obtain a production status parameter feature vector, wherein the feature vector includes a speed feature, a temperature feature, and a pressure feature; a prediction probability value and a prediction degree value are obtained through a discriminant function based on the feature vector, and a prediction record is generated; a confidence value is calculated for the prediction record, and the prediction record is filtered according to the confidence value, and a prediction result data packet is generated according to the prediction record, wherein the prediction result data packet includes a type probability distribution, a degree value, and a confidence value; the prediction result data packet is compared with a preset threshold value, and if the prediction probability value exceeds the probability threshold or the prediction degree value exceeds the numerical threshold, an orange warning signal is triggered; if the continuous prediction results exceed the threshold, a red warning signal is triggered; a parameter fluctuation feature vector is extracted from the warning signal, and the number of abnormal parameters is determined by the feature vector. The warning level is corrected according to the number of abnormal parameters, and a corrected warning signal is generated and sent to the control terminal.

[0029] Specifically, the discriminant function uses an online data stream transmission method with a 100-millisecond cycle to transmit the parameter fluctuation feature vector of the current production state. The feature vectors are arranged in the order of speed, temperature, and pressure features. Probabilistic statistics are used to calculate the proportion of each leaf node sample in each defect type as the predicted probability. The mean defect severity of the leaf node samples is calculated as the predicted severity. A data record is generated containing the prediction timestamp, predicted probability, and predicted severity. A confidence value is calculated based on the proportion of the number of leaf node samples to the total number of samples. Prediction records with a confidence value below 0.8 are eliminated. The remaining prediction records are sorted in descending order by type probability. The type with the highest probability is selected as the predicted type, and the corresponding defect severity value is extracted as the predicted value. A prediction result data packet is generated, containing the type probability distribution, severity value, and confidence value. For each prediction result data packet, the defect type probability is compared with a probability threshold of 0.8, and the defect severity value is compared with a numerical threshold of 0.6. If the predicted probability exceeds the probability threshold or the predicted value exceeds the numerical threshold, an orange alert is triggered. If the prediction results exceed the threshold for five consecutive times, a red alert is triggered. The alert level, trigger time, predicted data, and feature vector are recorded. Parameter fluctuation feature vectors are extracted from the warning data. Parameters with fluctuation amplitudes exceeding 0.3 in the speed, temperature, and pressure characteristics are identified as abnormal parameters. The warning level is adjusted based on the number of abnormal parameters. A warning signal is generated, including the warning level, abnormal parameters, fluctuation amplitude, prediction type, and prediction degree. This signal is then transmitted to the control terminal via a serial communication bus at a baud rate of 9600. During real-time parameter prediction, a 100-millisecond data transmission cycle corresponds to a 10-Hz update frequency, meeting the 2-meter-per-second operating speed requirement of the plastic flower leaf production line. The 18-dimensional feature vector includes 6 features from the speed group, 6 features from the temperature group, and 6 features from the pressure group. Leaf node sample statistics show that unevenness accounts for 45%, fracture accounts for 35%, and deformation accounts for 20%. Prediction probabilities are calculated based on the sample proportions. In the prediction result screening, a confidence threshold of 0.8 means that the leaf node must contain at least 100 training samples to be statistically significant. Taking the uneven ribbon defect as an example, in the prediction records with a confidence level exceeding 0.8, the typical prediction probability distribution is 0.85 for the uneven type, 0.1 for the fracture type, and 0.05 for the deformation type. The corresponding defect severity values are distributed between 0.3 and 0.8, and the confidence values are distributed between 0.8 and 0.95. In the early warning trigger mechanism, a probability threshold of 0.8 corresponds to an 85% defect prediction probability, and a numerical threshold of 0.6 corresponds to a moderate defect. An orange warning indicates a single prediction result exceeds the limit, and a red warning indicates five consecutive predictions that exceed the limit. In actual production, uneven ribbons caused by screw speed fluctuations usually trigger orange warnings, while ribbon fractures caused by abnormal barrel temperature usually trigger red warnings.In the judgment of abnormal parameters, the fluctuation amplitude threshold of 0.3 corresponds to a 30% change in the parameter. Under actual working conditions, when the speed fluctuation exceeds 90 revolutions per minute, the temperature fluctuation exceeds 120 degrees Celsius, and the pressure fluctuation exceeds 60 MPa, it will be judged as an abnormal parameter. When two abnormal parameters are detected, the orange warning will be upgraded to a red warning, and 10 groups of warning data will be transmitted to the control terminal per second through serial communication at a baud rate of 9600.

[0030] Step S107: If the warning signal is triggered, the process parameters are adjusted, including screw speed, barrel temperature and injection pressure, to control the process parameters within a preset range, thereby reducing the occurrence of ribbon defects.

[0031] The abnormal parameter name, current parameter value and fluctuation amplitude are obtained from the early warning signal, the abnormal parameters are prioritized according to the fluctuation amplitude, and the target interval value is obtained by reading the preset parameter range library; a digital adjustment instruction is generated according to the parameter correction amount of the abnormal parameter, the parameter correction amount is calculated by a proportional-integral controller, and the target value of the next adjustment cycle is obtained by an incremental adjustment method; the actual value and the target value of the abnormal parameter are read from the actuator, and a secondary correction is performed on the parameter whose deviation between the actual value and the target value exceeds the target interval threshold by a decreasing step size method; the abnormal parameter is monitored online, and when the fluctuation amplitude of the abnormal parameter is less than the preset threshold, steady-state compensation is performed by reducing the proportional coefficient and increasing the integral time constant.

[0032] Specifically, the system extracts the abnormal parameter name, current value, and fluctuation amplitude from the warning signal. The system then retrieves the target ranges of 50 to 150 rpm for screw speed, 220 to 260°C for barrel temperature, and 80 to 120 MPa for injection pressure by reading a preset parameter range library. The abnormal parameters are prioritized based on the fluctuation amplitude, and a feedback controller with a proportional coefficient of 0.6 and an integral time constant of 10 seconds is used to calculate parameter corrections. 16-bit digital adjustment commands are generated based on the parameter corrections, with adjustment steps set to 5 rpm for screw speed, 2°C for barrel temperature, and 2 MPa for injection pressure. The system calculates the target value for the next adjustment cycle for each parameter using incremental adjustment. These adjustment commands are then sent to the actuators in descending order of parameter response speed. The adjustment deviation is calculated by reading the actual and target values from the actuators. Parameters with deviations exceeding 5% of the target range undergo secondary corrections. This secondary correction uses a decreasing step size, with an initial step size of 50% of the baseline step size. Correction is terminated when the deviation is less than 2% for three consecutive adjustments. Steady-state errors are eliminated through integral compensation. The adjusted process parameters are monitored online with a sampling period of 100 milliseconds. The real-time parameter values are read and compared with the target range. Once the parameters enter the target range and the fluctuation amplitude is less than 1%, steady-state control is implemented. The proportional coefficient of the feedback controller is reduced to 0.3, and the integral time constant is increased to 20 seconds, achieving steady-state compensation of the parameters. During the parameter adjustment process, the preset ranges are determined based on experience in the production process of plastic flower leaves. The screw speed range of 50 to 150 rpm corresponds to the material plasticization rate, the barrel temperature range of 220 to 260 degrees Celsius corresponds to the material molten state, and the injection pressure range of 80 to 120 MPa corresponds to the melt filling pressure. Parameters are prioritized by response time, from fastest to slowest, with a pressure response time of 0.1 second, a speed response time of 0.5 second, and a temperature response time of 5 seconds. Digital adjustment commands use 16-bit precision. A speed adjustment value of 5 rpm corresponds to a digital value of 327, a temperature adjustment value of 2°C corresponds to a digital value of 131, and a pressure adjustment value of 2 MPa corresponds to a digital value of 262. In incremental adjustment, the target value for each adjustment cycle is an increment added to the current value. The injection pressure is first adjusted to 90 MPa, the screw speed is then adjusted to 100 rpm, and the barrel temperature is finally adjusted to 240°C. Parameter adjustment deviation is calculated using the difference between the target and actual values. A secondary correction is triggered when the screw speed deviation exceeds 5 rpm, the barrel temperature deviation exceeds 10°C, or the injection pressure deviation exceeds 4 MPa. The step sizes of the secondary correction are reduced to 2.5 rpm, 1°C, and 1 MPa, respectively. Three consecutive adjustments reduce the speed deviation to below 1 rpm, the temperature deviation to below 2°C, and the pressure deviation to below 0.5 MPa.Sampling monitoring during the steady-state control stage showed that the injection pressure fluctuated by 0.5 MPa near 90 MPa, the screw speed fluctuated by 0.8 rpm near 100 rpm, and the barrel temperature fluctuated by 1.5 degrees Celsius near 240 degrees Celsius. After the feedback controller parameters were adjusted, the pressure fluctuation dropped to 0.2 MPa, the speed fluctuation dropped to 0.3 rpm, and the temperature fluctuation dropped to 0.8 degrees Celsius, achieving stable control of the parameters.

[0033] Obviously, those skilled in the art may make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include such modifications and variations.

Claims

1. A product tracking method for a plastic flower digital workshop, characterized in that: The method comprises: The process parameter data of the screw speed curve, barrel temperature curve and injection pressure curve in the production process of the plastic flower leaf gradient color uneven product are obtained, and the process parameter data are preprocessed to obtain a standardized parameter data set; the standardized parameter data set is analyzed by a time series analysis algorithm, and the screw speed curve, barrel temperature curve and injection pressure curve are feature extracted to obtain key indicators reflecting the parameter fluctuation characteristics, and a parameter fluctuation feature vector is constructed based on the key indicators; the defect type and defect degree quality data of the color ribbon defective product are obtained, the color ribbon defect data are associated with the corresponding parameter fluctuation feature vector, and a mapping relationship data set between the color ribbon defect and the parameter fluctuation feature vector is constructed; the decision tree algorithm is used to train the mapping relationship data set to generate a discrimination between color ribbon defects and parameter fluctuations. The model predicts the possibility and degree of ribbon defects based on the parameter fluctuation eigenvector; in the production process of plastic flower leaf gradient ribbon products, the process parameter data of the screw speed curve, barrel temperature curve and injection pressure curve are collected in real time, and the collected data are preprocessed and feature extracted to obtain the parameter fluctuation eigenvector under the current production state; the parameter fluctuation eigenvector under the current production state is input into the discrimination model of ribbon defects and parameter fluctuations, and the possibility and degree of ribbon defects are predicted in real time. When the predicted defect possibility or defect degree exceeds the preset threshold, an early warning signal is triggered, indicating that the process parameters are abnormal; if the early warning signal is triggered, the process parameters are adjusted, including the screw speed, barrel temperature and injection pressure, to control the process parameters within the preset range, thereby reducing the occurrence of ribbon defects.

2. The method according to claim 1, characterized in that The process parameter data of the screw speed curve, the barrel temperature curve, and the injection pressure curve in the production process of the plastic flower leaf gradient color band uneven product are obtained, and the process parameter data are preprocessed to obtain a standardized parameter data set, including: A multi-channel data acquisition board is used to obtain a screw speed signal, a barrel temperature signal, and an injection pressure signal of an injection molding machine, and the screw speed signal, the barrel temperature signal, and the injection pressure signal of the injection molding machine are filtered using a Butterworth low-pass filter to obtain filtered data; According to the inflection point distribution of the parameter curve in the filtered data, a cubic spline interpolation function is used to fit the parameter curve to obtain characteristic data points; Perform normalization calculation on the characteristic data points, fit the normalized data through a polynomial regression equation, and eliminate outliers according to the mean square error criterion to obtain a characteristic vector sequence; The barrel temperature field distribution data, injection pressure change data and screw speed data in the characteristic vector sequence are used to establish a multivariate linear regression equation, and a combined parameter data set of injection pressure and screw speed at different barrel temperatures is obtained by traversing the parameter space.

3. The method according to claim 1, characterized in that The normalized parameter data set is analyzed by a time series analysis algorithm, and features are extracted from the screw speed curve, the barrel temperature curve, and the injection pressure curve to obtain key indicators reflecting parameter fluctuation characteristics. A parameter fluctuation feature vector is constructed based on the key indicators, including: Construct a time series based on the normalized parameter data set according to the sampling interval, and obtain the mean fluctuation value and variance fluctuation value of the series through the stationarity test; Calculate the autocorrelation function value of the delay point for the time series, use fast Fourier transform to obtain the spectrum amplitude from the autocorrelation function value, and select the main frequency point according to the spectrum amplitude; Obtaining frequency energy characteristic values from the main frequency point, extracting data segments from the parameter curve, and obtaining statistical index data using multiple linear regression; The statistical index data and the frequency energy characteristic values are normalized, characteristic quantities are extracted from the normalized data, and parameter fluctuation characteristic vectors are obtained according to the order of parameter types.

4. The method according to claim 1, wherein The method of obtaining defect type and defect degree quality data of ribbon defective products, associating the ribbon defect data with corresponding parameter fluctuation feature vectors, and constructing a mapping relationship data set between the ribbon defects and the parameter fluctuation feature vectors includes: Acquire an image of a product with color ribbon defects, obtain a noise-reduced image through Gaussian noise reduction processing, calculate the area ratio, morphological parameters, and color characteristics of the defective area based on the noise-reduced image, and obtain quality feature data; Adding a production batch number and a timestamp to the quality characteristic data to construct a defect characteristic record; Extracting feature vectors of adjacent time points from a parameter fluctuation feature vector database according to the timestamps, and performing a weighted average operation on the feature vectors using a weight coefficient to obtain a comprehensive feature vector; If the comprehensive feature vector satisfies the linear classifier mapping condition, a mapping relationship between the comprehensive feature vector and the defect type is established; An associated data record is generated according to the mapping relationship and the defect feature record.

5. The method according to claim 1, wherein The decision tree algorithm is used to train the mapping relationship data set to generate a discriminant model of ribbon defects and parameter fluctuations, and the possibility and degree of ribbon defects are predicted based on the parameter fluctuation feature vector, including: Obtain defect type labels, defect severity indices, and parameter fluctuation feature vectors sorted by production time in the mapping relationship dataset, and perform undersampling processing based on the feature vectors to obtain a training data matrix and a verification data matrix; Calculate the Gini coefficient for the parameter fluctuation eigenvector. If the Gini coefficient is greater than a threshold, determine the corresponding feature as a candidate split feature. Calculate the sample entropy of the candidate split feature using quantiles to obtain the optimal split node. Performing a grid search on the decision tree parameters according to the training data matrix, calculating the mean square error through the verification data matrix, and determining the parameter combination corresponding to the minimum mean square error to obtain the optimized decision tree parameters; The optimized decision tree parameters are used to train a decision tree model, a leaf node probability distribution is calculated for the decision tree model, and the defect type and defect degree value with the highest confidence are determined through the leaf node probability distribution.

6. The method according to claim 1, characterized in that In the production process of the plastic flower leaf gradient ribbon product, the process parameter data of the screw speed curve, the barrel temperature curve and the injection pressure curve are collected in real time, and the collected data are preprocessed and feature extracted to obtain the parameter fluctuation feature vector under the current production state, including: A multi-channel data acquisition card is used to acquire a screw speed signal, a barrel temperature signal, and an injection pressure signal from an injection molding machine control cabinet; a data record including a timestamp, a channel number, and a value is generated based on the screw speed signal, the barrel temperature signal, and the injection pressure signal from the injection molding machine control cabinet; and the data record is stored in a circular buffer according to a first-in-first-out principle. Reading a data block from the circular buffer, filtering and denoising the data block using a Butterworth low-pass filter, and completing the data block using cubic spline interpolation to obtain a continuous parametric curve; Performing power spectrum density analysis on the continuous parameter curve, extracting the main frequency point eigenvalues and statistical eigenvalues of the continuous parameter curve, and obtaining a characteristic data sequence; For the characteristic data sequence, characteristic values of the screw speed group, the barrel temperature group and the injection pressure group are extracted. If the characteristic value fluctuation amplitude exceeds a preset threshold, the characteristic value is organized into a parameter fluctuation characteristic vector.

7. The method according to claim 1, characterized in that The parameter fluctuation feature vector under the current production state is input into the discriminant model of ribbon defects and parameter fluctuations to predict the possibility and degree of ribbon defects in real time. When the predicted defect possibility or defect degree exceeds a preset threshold, an early warning signal is triggered to indicate that the process parameters are abnormal, including: A production status parameter feature vector is obtained by using a periodic data stream transmission method, wherein the feature vector includes a rotation speed feature, a temperature feature, and a pressure feature; Obtain a prediction probability value and a prediction degree value based on the feature vector through a discriminant function, and generate a prediction record; Calculating a confidence value for the prediction record, filtering the prediction record by the confidence value, and generating a prediction result data packet based on the prediction record, the prediction result data packet including a type probability distribution, a degree value, and a confidence value; Comparing the prediction result data packet with a preset threshold, if the prediction probability value exceeds the probability threshold or the prediction degree value exceeds the numerical threshold, an orange warning signal is triggered; If the continuous prediction results exceed the threshold, a red warning signal is triggered; A parameter fluctuation feature vector is extracted from the warning signal, the number of abnormal parameters is determined by the feature vector, the warning level is corrected according to the number of abnormal parameters, and a corrected warning signal is generated and sent to the control terminal.

8. The method according to claim 1, characterized in that If the warning signal is triggered, the process parameters are adjusted, including screw speed, barrel temperature and injection pressure, to control the process parameters within the preset range, thereby reducing the occurrence of ribbon defects, including: Obtain the abnormal parameter name, current value and fluctuation amplitude from the warning signal, prioritize the abnormal parameters according to the fluctuation amplitude, and obtain the target interval value by reading the preset parameter range library; Generate a digital adjustment instruction according to the parameter correction amount of the abnormal parameter, calculate the parameter correction amount using a proportional-integral controller, and obtain the target value of the next adjustment cycle through an incremental adjustment method; Reading the actual value and the target value of the abnormal parameter from the actuator, and performing secondary correction in a decreasing step manner on the parameter whose deviation between the actual value and the target value exceeds the target interval threshold; The abnormal parameters are monitored online, and when the fluctuation amplitude of the abnormal parameters is less than a preset threshold, steady-state compensation is performed by reducing the proportional coefficient and increasing the integral time constant.

Images