Image recognition-based blow molded article wall thickness online control method

Abstract

The application provides a kind of based on image recognition's blow molding product wall thickness on-line control method, it is related to on-line control technical field, method includes: type process parameter is formulated, wall thickness adjustment parameter is calculated according to adjusting melt extrusion quantity and die port form, type process parameter is adjusted to obtain forming control parameter, the thickness of target blow molding product in forming process is collected to obtain time series thickness data.Judge whether time series thickness data reaches preset target thickness threshold value, yes, maintain current forming control parameter, otherwise, wall thickness adjustment parameter is optimized.Wall thickness deviation feature is obtained by feature extraction to time series thickness data, and the linkage relationship of wall thickness deviation feature and type process parameter is analyzed to obtain associated correction parameter, and wall thickness adjustment parameter is optimized to obtain wall thickness correction parameter.The application can enhance the directional correction ability of local deviation, improve the wall thickness control precision and stability, adapt to different shape and function products.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of online control technology, and in particular to a blow molding product wall thickness online control method based on image recognition. BACKGROUND

[0002] Blow molding is an important plastic processing method and is widely used in plastic product production. With the continuous improvement of the quality requirements of plastic products, the accurate control of the wall thickness of blow molding products has become one of the key technologies.

[0003] Early machines use two-stage oil pressure to control the core shaft gap, which can only achieve two simple effects of thick or thin, and cannot meet the needs of modern high-precision production. Later, although modern embryo wall thickness control systems are equipped with servo oil pressure systems and computers to form multi-stage control, improving product quality, there are still some problems. For example, it is difficult to accurately adjust the process parameters according to the wall thickness deviation. At the same time, when facing local wall thickness deviation, the traditional control method often has difficulty in accurate correction, resulting in unstable product quality.

[0004] Some automatic transverse thickness control systems based on fuzzy control algorithm, such as automatic air ring, although they have improved the transverse thickness uniformity of the film to some extent, but for highly nonlinear, strongly coupled, time-varying and control uncertainty systems, their precise mathematical model is almost impossible to establish, and the control effect still needs to be improved. In addition, some existing wall thickness control methods also have shortcomings in processing real-time data and dynamic adjustment, and cannot respond to wall thickness changes in the blow molding process in a timely and accurate manner.

[0005] In recent years, with the development of image recognition technology, its application in the field of industrial detection has become more and more widespread. Image recognition technology has the advantages of non-contact, high precision, and strong real-time performance, and can provide more effective means for blow molding product wall thickness online control. However, the application of image recognition technology in blow molding product wall thickness online control is still relatively rare, and a mature technical system has not yet been formed.

[0006] In view of the problems existing in the prior art, the present application provides a blow molding product wall thickness online control method based on image recognition. SUMMARY

[0007] The present application provides a blow molding product wall thickness online control method based on image recognition, which solves the problems of unclear correlation between process parameters and wall thickness deviation and difficulty in accurately correcting local wall thickness deviation in the prior art.

[0008] The present application provides a blow molding product wall thickness online control method based on image recognition, which includes:

[0009] Collect raw material characteristic data, formulate type process parameters according to target blow molding product shape, calculate wall thickness adjustment parameters according to wall thickness regulation method by adjusting melt extrusion amount and die gap shape.

[0010] Adjust type process parameters according to wall thickness adjustment parameters to obtain molding control parameters, collect image data of target blow molding product in molding process, and process image data by using dynamic partition calibration method to obtain time series thickness data.

[0011] Determine whether the time series thickness data reaches the preset target thickness threshold, if yes, maintain the current molding control parameters, otherwise optimize the wall thickness adjustment parameters.

[0012] Obtain wall thickness deviation features by feature extraction on time series thickness data, analyze the linkage relationship between wall thickness deviation features and type process parameters according to sequence pattern mining method to obtain correlation correction parameters, and optimize wall thickness adjustment parameters to obtain wall thickness correction parameters.

[0013] The present application provides a blow molding product wall thickness online control method based on image recognition, and the steps of formulating type process parameters include:

[0014] Determine performance indicators according to product use, obtain shape parameters of target blow molding product, and obtain raw material characteristic data from basic performance and processing adaptability of raw materials.

[0015] Calculate theoretical extrusion amount according to single weight and production speed of target product, calculate initial screw speed according to screw diameter and pitch, and set head temperature of different functional sections according to use function.

[0016] Set blow molding pressure and cooling water temperature according to raw material characteristic data, calculate initial gap of die according to target wall thickness and blow ratio of product, and correct type process parameters by combining historical optimal parameters of similar products.

[0017] The present application provides a blow molding product wall thickness online control method based on image recognition, and the steps of calculating wall thickness adjustment parameters include:

[0018] Calculate melt volume extrusion amount in current production according to initial screw speed.

[0019] Collect average die gap, die gap distribution and die outlet diameter as die key parameters.

[0020] Calculate target required extrusion amount and target required gap with unchanged die shape according to preset target thickness threshold.

[0021] Extrude raw materials according to target required extrusion amount, and correct local wall thickness deviation area according to target required gap to obtain die gap distribution adjustment scheme.

[0022] Detecting the head pressure data in the blow molding process, combining the initial screw rotation speed to calculate the real-time melt viscosity, when exceeding the preset reference viscosity, adjusting the target demand extrusion amount and the target demand gap to obtain the wall thickness adjustment parameter.

[0023] The present application provides a blow molding product wall thickness online control method based on image recognition, and the step of adjusting the molding control parameter comprises:

[0024] According to the corresponding association of the wall thickness adjustment parameter and the type process parameter, a parameter mapping table is constructed, and the influence amplitude of the unit change of the type process parameter on the wall thickness is determined to form a sensitivity coefficient table.

[0025] Based on the parameter mapping table and the sensitivity coefficient table, the wall thickness adjustment parameter is converted into the adjustment value of the process parameter.

[0026] According to the influence of directness and response speed, each parameter in the wall thickness adjustment parameter is sorted, and the adjustment rate is set according to the single-step maximum adjustment amount.

[0027] The adjustment value is superimposed with the type process parameter, and the molding control parameter is generated according to the equipment safety and the process feasible range.

[0028] The present application provides a blow molding product wall thickness online control method based on image recognition, and the step of obtaining the time sequence thickness data comprises:

[0029] Based on the blow molding product moving speed and the minimum detection accuracy, the sampling frequency is calculated, and the detection area is divided according to the blow molding product structure and the circumferential direction.

[0030] According to the sampling frequency and the detection area, image data is collected and preprocessed to obtain preprocessed thickness data.

[0031] Taking the pulse of the main encoder of the production line as the reference time, the preprocessed thickness data collected by different sensors is calibrated, the data segment is divided according to the molding cycle for intermittent production, and the time sequence thickness data is obtained by removing outliers and continuity verification.

[0032] The present application provides a blow molding product wall thickness online control method based on image recognition, and the step of extracting the wall thickness deviation feature comprises:

[0033] According to the preset target wall thickness threshold as the reference, the absolute deviation and the relative deviation of each sampling point are calculated to define the deviation, and the time sequence thickness data is sliced to form a regionalized sub-data set according to the detection area.

[0034] From the time sequence point of view, the trend feature, the fluctuation feature and the burst feature of the wall thickness deviation are analyzed to obtain the dynamic change law.

[0035] From the perspective of the spatial distribution of the blow molding product, the position characteristics of the wall thickness deviation are analyzed, and local defects and regional differences are identified.

[0036] The dynamic change rule, local defects, and regional differences are integrated to form the wall thickness deviation characteristics.

[0037] The present application provides a blow molding product wall thickness online control method based on image recognition, and the steps for obtaining the dynamic change rule include:

[0038] The average deviation and cumulative deviation in the preset period are calculated, and the trend characteristics are obtained by linear regression fitting the deviation change curve over time.

[0039] The standard deviation and range of the regionalized sub-data set are calculated, and the Fourier transform is performed to extract the fluctuation frequency and corresponding amplitude to obtain the fluctuation characteristics.

[0040] The deviation peak value and duration exceeding the preset target wall thickness threshold are extracted as the deviation time, and the deviation change amount of adjacent data points is calculated as the burstiness characteristics.

[0041] The trend characteristics, fluctuation characteristics, and burstiness characteristics are integrated to obtain the dynamic change rule.

[0042] The present application provides a blow molding product wall thickness online control method based on image recognition, and the steps for obtaining the associated correction parameters include:

[0043] The type process parameters and wall thickness deviation characteristics are divided into discrete states according to the normal range, and the adjustment action and burstiness characteristics are sequentially encoded to form the associated sequence pair.

[0044] The PrefixSpan algorithm is used to mine the frequent patterns in the associated sequence pair.

[0045] The frequent patterns are converted into process parameter correction rules, including trigger conditions, associated process parameters, correction actions, and confidence levels.

[0046] Based on the statistical relationship between the process parameter change amount and the deviation change amount in the frequent patterns, the correction coefficient is calculated.

[0047] When multiple process parameter correction rules are triggered simultaneously, the process parameter correction rule with the maximum correction coefficient is selected as the associated correction parameter.

[0048] The present application provides a blow molding product wall thickness online control method based on image recognition, and the steps for obtaining the frequent patterns include:

[0049] All associated sequence pairs are traversed, all item prefix patterns are extracted to calculate the support, and item prefix patterns exceeding the preset minimum support are extracted to form the initial frequent sequence.

[0050] A projection database is generated for each initial frequent sequence, and all suffix items of the current initial frequent sequence are combined to form a candidate item pattern.

[0051] For each candidate item pattern, the number of occurrences in the projection database is counted, and frequent item patterns are screened out.

[0052] For each frequent item pattern, the projection database is continuously constructed, the prefix pattern is recursively expanded, and long sequence frequent patterns are generated.

[0053] Short patterns that do not meet the preset length in the long sequence frequent patterns are removed, and the frequent patterns are sorted by length to form short to long frequent patterns.

[0054] The present application provides a blow molding product wall thickness online control method based on image recognition, and the steps of optimizing the wall thickness correction parameter include:

[0055] A correction mapping table is constructed according to the mapping relationship between the associated correction parameter and the wall thickness adjustment parameter, and a plurality of adjustment amounts are obtained by sorting the influence weight of the associated correction parameter on the wall thickness.

[0056] Each level of adjustment amount is adjusted according to the associated correction parameter and the correction coefficient to obtain a plurality of adjustment parameters.

[0057] It is judged whether the wall thickness adjusted by the multi-level adjustment parameter reaches the preset target thickness threshold, if yes, the current multi-level adjustment parameter is taken as the wall thickness correction parameter, otherwise the correction coefficient is recalculated.

[0058] The present application provides a blow molding product wall thickness online control method based on image recognition, which realizes real-time monitoring of wall thickness changes through an online control method based on image recognition, and adjusts process parameters according to real-time data to ensure wall thickness uniformity, solves the defects that traditional blow molding product wall thickness control methods usually rely on experience parameter setting and are difficult to adjust in real time, resulting in poor wall thickness uniformity and high scrap rate. Through dynamic partition calibration method and real-time melt viscosity detection, the wall thickness adjustment parameter is adjusted in real time, and the real-time performance of wall thickness control is improved. Through partition detection and feature extraction, independent detection and adjustment are carried out for different functional sections and local areas, and the wall thickness uniformity of complex shape products is ensured, which solves the defects that traditional methods are difficult to accurately control the wall thickness of blow molding products with complex shape, resulting in local wall thickness deviation. Through real-time monitoring and adjustment, the wall thickness uniformity is ensured, the scrap rate is reduced, and the product quality is improved. And using partition detection and feature extraction, the wall thickness uniformity of complex shape products is ensured, and products with different shapes and functions are adapted. The automatic extraction and continuous optimization of process knowledge are realized, which significantly improves the intelligentization and automation level of the blow molding process, and has important industrial application value and popularization prospect. BRIEF DESCRIPTION OF DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the present application or the prior art, the accompanying drawings required by the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description are some embodiments of the present application, and other drawings can be obtained by a person of ordinary skill in the art without creative effort.

[0060] Figure 1 is one of the flowcharts of the image recognition-based blow molding product wall thickness online control method provided by the embodiments of the present application;

[0061] Figure 2 is another flowchart of the image recognition-based blow molding product wall thickness online control method provided by the embodiments of the present application;

[0062] Figure 3 is a third flowchart of the image recognition-based blow molding product wall thickness online control method provided by the embodiments of the present application;

[0063] Figure 4 is a fourth flowchart of the image recognition-based blow molding product wall thickness online control method provided by the embodiments of the present application;

[0064] Figure 5 is a correction mapping table of the image recognition-based blow molding product wall thickness online control method provided by the embodiments of the present application. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the accompanying drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the protection scope of the present application.

[0066] The present application will be described below with reference to the accompanying drawings. Figures 1-5 An image recognition-based blow molding product wall thickness online control method is described.

[0067] As shown in the accompanying drawings, Figure 1 the image recognition-based blow molding product wall thickness online control method provided by the embodiments of the present application comprises:

[0068] Collecting raw material characteristic data, combining target blow molding product shape to formulate type process parameters, and calculating wall thickness adjustment parameters according to adjustment of melt extrusion amount and die shape by wall thickness regulation method.

[0069] The step of formulating type process parameters comprises:

[0070] According to the performance index of the product use, the shape parameters of the target blow molding product are obtained, and the material characteristic data are obtained by analyzing the basic performance and processing adaptability of the raw materials.

[0071] Basic performance detection: the melt flow rate can be detected by a melt flow rate instrument, and the fluidity of the material under standard temperature and load is recorded.

[0072] The molecular weight distribution is analyzed by gel permeation chromatography to obtain the weight average molecular weight and the molecular weight distribution index (PDI), and the material with PDI≤3 is preferentially selected to ensure the processing stability.

[0073] The melting temperature and crystallinity of the material are determined by a differential scanning calorimeter to determine the processing temperature range, and the processing temperature needs to be higher than the melting temperature.

[0074] Processing adaptability detection: for hygroscopic materials, the water content is detected by a Karl Fischer moisture meter, and the water content is required to be ≤0.02% to avoid bubbles during processing. The viscosity curve of the melt under different shear rates and temperatures is determined by a rotary rheometer to determine the correlation between viscosity, temperature and shear rate.

[0075] According to the single weight and production speed of the target product, the theoretical extrusion amount is calculated, and the initial screw rotation speed is calculated combined with the screw diameter and pitch. According to the use function, the head temperature of different functional segments is set segmentally.

[0076] According to the material characteristic data, the blow molding pressure and cooling water temperature are set, the initial gap of the die is calculated according to the target wall thickness and blow ratio of the product, and the type process parameters are obtained by correcting the historical optimal parameters of similar products.

[0077] As shown in Figure 2 The steps of calculating the wall thickness adjustment parameter include:

[0078] According to the initial screw rotation speed, the melt volume extrusion amount in the current production is calculated, which is expressed by the formula:

[0079]

[0080] In the formula, is the initial screw rotation speed, is the displacement per revolution of the screw, is the melt flow efficiency coefficient, is the melt volume extrusion amount.

[0081] The average gap of the die, the die gap distribution and the die outlet diameter are collected as the key parameters of the die.

[0082] According to the preset target thickness threshold, the target demand extrusion amount and the target demand gap of the die shape are calculated, which is expressed by the formula:

[0083]

[0084]

[0085] wherein, is the target demand extrusion amount, is the target demand gap, is the preset target thickness threshold, is the material flow rate at the die, is the die diameter, is the initial gap, is the correction coefficient, is the current extrusion amount.

[0086] According to the target demand extrusion amount, the raw material is extruded, and the local wall thickness deviation area is corrected in combination with the target demand gap to obtain a die gap distribution adjustment scheme.

[0087] The head pressure data in the blow molding process is detected, and the real-time melt viscosity is calculated in combination with the initial screw speed. When the preset reference viscosity is exceeded, the target demand extrusion amount and the target demand gap are adjusted to obtain a wall thickness adjustment parameter.

[0088] According to the wall thickness adjustment parameter, the type process parameter is adjusted to obtain a molding control parameter, the image data of the target blow molding product in the molding process is collected, and the dynamic partition calibration method is used to process the image data to obtain time sequence thickness data.

[0089] The step of adjusting to obtain the molding control parameter comprises:

[0090] According to the corresponding association between the wall thickness adjustment parameter and the type process parameter, a parameter mapping table is constructed, and the influence amplitude of the unit change of the type process parameter on the wall thickness is determined to form a sensitivity coefficient table.

[0091] Based on the parameter mapping table and the sensitivity coefficient table, the wall thickness adjustment parameter is converted into an adjustment value of the process parameter.

[0092] Screw speed adjustment: the screw speed adjustment amount is calculated according to the melt extrusion amount adjustment amount and the screw displacement.

[0093] Die total opening adjustment: the die total opening adjustment amount is calculated according to the die average gap adjustment amount and the die mechanical transmission ratio.

[0094] Partition process parameter adjustment amount calculation: die local gap adjustment: the displacement-gap conversion coefficient of the deviation area in the circumferential direction and the corresponding area servo motor is calculated to adjust the stroke of each motor.

[0095] Head partition temperature adjustment: for the area that needs to be thickened, calculate the temperature reduction according to the viscosity-temperature curve of the material.

[0096] Sort the parameters in the wall thickness adjustment parameters according to the impact on directness and response speed, and set the adjustment rate according to the single-step maximum adjustment amount.

[0097] Superimpose the adjustment value on the type process parameter, and adjust it according to the equipment safety and process feasibility range to generate the molding control parameter.

[0098] Convert the molding control parameter into a format that the equipment can recognize to realize automatic adjustment: instruction format conversion: convert the molding control parameter into a digital signal according to the equipment communication protocol:

[0099] Screw rotation speed→ frequency converter frequency instruction. Die gap→ servo motor pulse number. Temperature→ PID controller set value.

[0100] Synchronous issuance and state feedback: issue the instructions in order according to the adjustment priority, and simultaneously collect the actual value of the parameters in real time through sensors (such as encoder feedback actual speed, displacement sensor feedback actual die gap).

[0101] If the deviation between the actual value and the target value exceeds ±2% (such as target speed 40r / s, actual speed 39r / s), trigger the secondary fine adjustment until the deviation is ≤±1%.

[0102] As shown in Figure 3 , the steps of obtaining time sequence thickness data include:

[0103] Based on the moving speed of the blow molding product and the minimum detection accuracy, calculate the sampling frequency, and divide the detection area according to the structure and circumferential direction of the blow molding product.

[0104] Partition according to the structure of the blow molding product: such as dividing the bottle body into mouth section, body section and bottom section, and setting independent detection threshold for each section.

[0105] For the circumferential direction: divide the detection points according to the angle, and record the thickness distribution of the whole circumferential direction.

[0106] According to the sampling frequency and the detection area, collect image data and pre-process to obtain pre-processed thickness data. According to the sampling frequency, bind the collection frequency of the image sensor (such as high-speed industrial camera) with the pulse signal of the main encoder of the production line to ensure that each frame of image corresponds to a fixed displacement of the blow molding product, avoiding sampling misplacement caused by speed fluctuation. For structure partition such as mouth section, body section and bottom section, use multiple sensors for cooperative collection: use macro lens for mouth area, wide-angle lens for body area, and add side-view camera to capture corner details.

[0107] The 36 detection points are divided by 10° intervals in the circumferential direction, each detection point corresponding to an independent image ROI (region of interest), ensuring that there is no dead angle in the full circumferential thickness data.

[0108] In combination with the transmission laser thickness measurement and infrared thermal imaging, the laser thickness image and the temperature distribution image of each detection area are synchronously collected, each frame of image is attached with a time stamp and encoder position information, and the image data is formed.

[0109] The step of noise filtering the image data includes: using adaptive median filtering on the laser image to eliminate salt and pepper noise caused by dust and bubbles; using Gaussian filtering to smooth the temperature fluctuation noise of the infrared image, and retaining the real temperature gradient characteristics. For low-contrast areas such as bottle bottom corners, a limited contrast adaptive histogram equalization algorithm is used to enhance local details, ensuring that the wall thickness boundary is clear and identifiable, and contrast enhancement is performed.

[0110] Based on the preset gray-scale-thickness calibration curve (obtained by calibrating standard thickness blocks), the gray-scale value of the laser image is converted into the actual wall thickness value to obtain preprocessed thickness data.

[0111] Taking the main encoder pulse of the production line as the reference time, the preprocessed thickness data collected by different sensors is calibrated, the data segments are divided according to the molding cycle for intermittent production, and abnormal value elimination and continuity verification are performed to obtain time series thickness data.

[0112] Abnormal value elimination: when the thickness value of a certain point exceeds the theoretical range (such as target wall thickness 1.2 mm, actual measurement 3.0 mm or 0.1 mm), it is marked as invalid value and the abnormal reason (such as sensor shielding, image blur) is recorded.

[0113] Continuity verification: if the thickness change rate of 5 consecutive points exceeds 50% (such as from 1.2 mm to 0.5 mm), trigger sensor state self-check (such as laser power, camera focal length).

[0114] Determine whether the time series thickness data reaches the preset target thickness threshold, if yes, maintain the current molding control parameters, otherwise optimize the wall thickness adjustment parameters.

[0115] Obtain wall thickness deviation features by feature extraction on the time series thickness data, analyze the linkage relationship between the wall thickness deviation features and the type process parameters according to the sequence pattern mining method to obtain the associated correction parameters, and optimize the wall thickness adjustment parameters to obtain the wall thickness correction parameters.

[0116] The step of extracting wall thickness deviation features includes:

[0117] According to the preset target wall thickness threshold as the reference, the absolute deviation and the relative deviation of each sampling point are calculated to define the deviation, and the time series thickness data is sliced to form regionalized sub-data sets according to the detection area.

[0118] The dynamic change pattern is obtained by analyzing the trend, fluctuation and sudden characteristics of wall thickness deviation from the perspective of time series.

[0119] The steps to obtain the dynamic change pattern include:

[0120] Calculate the average and cumulative deviations within a preset time period, and obtain the trend characteristics by fitting the deviation change curve over time through linear regression.

[0121] The formula for calculating the average error is expressed as follows:

[0122]

[0123] In the formula, It is the average error. It is the number of data points. It is the first Error value for each data point It is the independent variable.

[0124] The formula for calculating the cumulative deviation is expressed as follows:

[0125]

[0126] In the formula, It is the cumulative deviation. It is the first The absolute value of the error for each data point.

[0127] The formula for fitting the curve of deviation over time using linear regression is expressed as follows:

[0128]

[0129] In the formula, It controls the deviation of the input. It's the slope. It is time. It is the intercept.

[0130] The wave characteristics are obtained by calculating the standard deviation and range of the regionalized subset datasets and performing Fourier transform to extract the wave frequency and corresponding amplitude.

[0131] The formula for calculating the standard deviation is expressed as:

[0132]

[0133] In the formula, It is the standard deviation.

[0134] The formula for calculating the range is expressed as follows:

[0135]

[0136] wherein, is the maximum value in the range, is is the minimum value in the range, is is the minimum value in the range.

[0137] The deviation peak value and duration exceeding the preset target wall thickness threshold are extracted as the deviation time, and the deviation change amount of adjacent data points is calculated as the burstiness feature.

[0138] The formula for calculating the deviation change amount is:

[0139]

[0140] wherein, is the deviation change amount.

[0141] The trendiness feature, the volatility feature, and the burstiness feature are integrated to obtain the dynamic change rule.

[0142] The position characteristics of the wall thickness deviation are analyzed from the perspective of the spatial distribution of the blow molded product, and local defects and regional differences are identified.

[0143] Regional deviation distribution features:

[0144] Regional average deviation: the average deviation of each detection region is calculated to compare the deviation degrees of different regions.

[0145] Deviation proportion: the proportion of the number of out-of-tolerance points in a certain region to the total number of points is calculated (for example, the out-of-tolerance proportion of the corner region is 30%, indicating that the molding of this region is unstable).

[0146] Circumferential direction deviation distribution: for the same axial position, a circumferential angle (0°-360°) and deviation value relationship curve is drawn, the angle corresponding to the maximum deviation is extracted, and the axial position of the deviation mutation is identified (for example, the gradient at the connection between the bottle mouth and the bottle body is large, which may be due to unreasonable transition design of the mold mouth).

[0147] Spatial correlation feature: a clustering algorithm (such as K-means) is used to cluster the full-space deviation points, and continuous out-of-tolerance regions (such as “100-150 mm axial range on the right side of the bottle body is continuously thick”) are identified. The area, center coordinates, and other features of the region are extracted to determine whether it is a systematic defect.

[0148] The dynamic change rule and the local defects and regional differences are integrated to form the wall thickness deviation feature.

[0149] As shown in Figure 4 , the steps of obtaining the correlation correction parameter include:

[0150] The process parameters and wall thickness deviation characteristics are classified into discrete states according to the normal range, and the adjustment actions and sudden characteristics are sequence-encoded to form associated sequence pairs.

[0151] The PrefixSpan algorithm is used to mine frequent patterns in associated sequence pairs.

[0152] Frequent patterns are transformed into process parameter correction rules, which include trigger conditions, associated process parameters, correction actions, and confidence levels.

[0153] Based on the statistical relationship between the changes in process parameters and the changes in deviations in frequent data processing modes, the correction coefficient is calculated, and the formula is expressed as follows:

[0154]

[0155] In the formula, It is the first Correction coefficients for each process parameter, It is the first Parameters when the secondary association rule is triggered The normalized change It corresponds to the first The amount of wall thickness deviation change triggered by the second time. It is a parameter Pearson correlation coefficient with bias.

[0156] From which we obtain The formula is expressed as:

[0157]

[0158] In the formula, It is the first Parameters when the secondary association rule is triggered The change Process parameters The arithmetic mean of the changes in all association rule-triggered events. Process parameters The sample mean difference of the amount of change across all association rule-triggered events.

[0159] When multiple process parameter correction rules are triggered simultaneously, the process parameter correction rule with the largest correction coefficient is selected as the associated correction parameter.

[0160] The steps to obtain frequent patterns include:

[0161] Iterate through all associated sequence pairs, extract all item prefix patterns, calculate their support, and extract item prefix patterns that exceed a preset minimum support to form an initial frequent sequence. Item prefix patterns can be expressed as: This indicates the status of process parameters. Corresponding bias feature The formula for calculating the support is:

[0162]

[0163] In the formula, is the support, is the number of sequence pairs containing the prefix pattern of the item, is the total number of sequence pairs.

[0164] For each initial frequent sequence, a projection database is generated, and all suffix items of the current initial frequent sequence are combined to form candidate item patterns. The projection database only retains the associated sequence pairs containing the prefix pattern of the item, and truncates all items before the prefix pattern of the item in the sequence, retaining only the subsequence (called "suffix sequence") after the prefix pattern of the item, for mining longer patterns.

[0165] For each candidate item pattern, the number of occurrences in the projection database is counted, and frequent item patterns are selected. The selection method is to retain candidate item patterns with support greater than or equal to the preset minimum support as frequent item patterns.

[0166] For each frequent item pattern, the projection database is continued to be constructed, and the prefix pattern is recursively expanded to generate long sequence frequent patterns. For each frequent item pattern, the new projection database (only retaining the suffix sequence after the pattern) is repeatedly constructed.

[0167] The suffix item is extracted from the new projection database and combined with the current frequent pattern to form a new item candidate pattern, the support is calculated and selected, and the frequent item sequence set is formed.

[0168] When it is impossible to generate longer patterns with support ≥ minimum support, the recursion is stopped.

[0169] Short patterns that do not meet the preset length in long sequence frequent patterns are removed, and are sorted by length to form frequent patterns from short to long. The frequent pattern can be "high-speed production causes the thickness to be too thick, and the die gap is restored to normal after increasing the die gap", with a support of 3% and a confidence of 85%.

[0170] The steps of optimizing the wall thickness correction parameter include:

[0171] As shown in Figure 5 , a correction mapping table is constructed according to the mapping relationship between the associated correction parameter and the wall thickness adjustment parameter, and is sorted according to the influence weight of the associated correction parameter on the wall thickness.

[0172] The sorting can be divided into: first level (die gap correction), second level (extrusion amount / rotation speed correction), and third level (temperature / pressure correction).

[0173] The multi-stage adjustment parameter is obtained by adjusting each stage adjustment amount according to the associated correction parameter and the correction coefficient. The multi-stage adjustment parameter includes a local parameter, a whole parameter and an auxiliary parameter.

[0174] The local parameter is calculated according to the associated correction parameter and the correction coefficient.

[0175] The whole parameter is calculated based on the screw speed correction amount, the screw displacement and the flow efficiency coefficient, and the formula is expressed as:

[0176]

[0177] In the formula, is the screw speed correction amount, is the displacement per revolution of the screw, is the melt flow efficiency coefficient.

[0178] The third-stage adjustment (auxiliary parameter) is calculated according to the head temperature correction amount to calculate the viscosity compensation coefficient to correct the extrusion stability. The stretching correction amount is adjusted according to the blow pressure correction amount ΔP.

[0179] Whether the wall thickness after adjustment by the multi-stage adjustment parameter reaches the preset target thickness threshold is determined. If yes, the current multi-stage adjustment parameter is taken as the wall thickness correction parameter, otherwise the correction coefficient is recalculated.

[0180] The wall thickness correction parameter is executed step by step according to the priority: first, the die local adjustment (single step ≤0.05mm, interval 3 cycles), then the extrusion amount / speed adjustment (single step ≤±1r / s), and finally the auxiliary parameter adjustment (single step ≤±3℃ or ±0.05MPa).

[0181] Dynamic feedback: after each step of adjustment, the real-time wall thickness is collected, the actual deviation is calculated, and if the deviation is >0.05mm, the correction coefficient is iteratively adjusted by re-optimization.

[0182] The wall thickness correction parameter set (including the whole parameter, the local parameter and the auxiliary parameter) is split according to the equipment control dimension, the parsed instructions are converted into digital signals recognizable by the equipment according to the communication protocol of the blow molding machine control system, and the instruction transmission error is ensured.

[0183] Step-by-step execution according to priority: first stage (0-2 seconds): execute the die local gap correction, control the servo motor to adjust to the target position at a rate of single step 0.02mm, and synchronously collect the feedback signal of the die position sensor;

[0184] Second stage (2-5 seconds): execute the screw speed / extrusion amount correction, adjust the speed at a rate of single step ±0.5r / s, and monitor the melt pressure sensor data in real time;

[0185] Third stage (5-10 seconds): Perform temperature / pressure correction, adjust temperature at a rate of ±1°C / s and adjust pressure at a rate of ±0.02 MPa / s to avoid fluctuations in the melt state caused by sudden changes in parameters.

[0186] Synchronously collect the following data to verify the execution effect: image data: take real-time photos of the product during molding through an industrial camera to calculate the current wall thickness value; equipment state data: actual values of screw speed, die gap, head temperature, and blow molding pressure; environmental data: workshop temperature and humidity (fluctuations should be ≤±2°C and ±5%).

[0187] After executing the correction parameters for 10 seconds, extract new time-series thickness data and calculate the corrected wall thickness deviation: if the overall deviation is ≤±0.03 mm and the local deviation is ≤±0.02 mm, determine that the correction is effective and enter the stable production stage.

[0188] If the deviation does not meet the standard (e.g., the local thickness is still 0.04 mm), trigger the fine-tuning mechanism: recalculate the correction coefficient based on the new deviation to generate secondary correction parameters.

[0189] If parameter coupling conflicts occur (e.g., increasing the speed leads to pressure exceeding the limit), start the collaborative correction logic: reduce the speed adjustment amplitude or make compensatory temperature adjustments to ensure that the wall thickness meets the standard under the collaborative action of multiple parameters.

[0190] After the correction parameters pass the verification, lock the current molding control parameters and start the batch production mode: randomly sample and detect every 100 products to confirm the wall thickness stability (standard deviation ≤0.02 mm); automatically record the key parameters of each batch (such as raw material batch, environmental temperature, and average wall thickness) to form a traceable production file.

[0191] If sudden abnormalities occur during production (e.g., raw material supply interruption, sensor failure): the system automatically suspends production and saves the current parameter state.

[0192] After troubleshooting, based on the wall thickness data and parameter state before the interruption, perform 3-piece trial production verification, and resume batch production after meeting the standard.

[0193] Comprehensively detect the offline products: dimensional accuracy: bottle mouth diameter, bottle body perpendicularity;

[0194] Wall thickness uniformity: use an ultrasonic thickness gauge to detect key areas (such as bottle mouth sealing surface and bottle bottom corner);

[0195] Mechanical properties: pressure resistance test, drop test (meet product standards).

[0196] If the finished product qualified rate is greater than or equal to 99.5%, the wall thickness correction parameter, forming control parameter and corresponding process condition (such as raw material model, environmental parameter) of this production are stored in the process knowledge base; if the qualified rate is less than 95%, the deviation reason is traced back and analyzed, the frequent rules and correction coefficients mined by the sequence pattern are updated, and the optimization basis is provided for subsequent production.

[0197] The image recognition-based blow molding product wall thickness online control method provided in the embodiment improves the sampling accuracy and synchronization of time sequence thickness data significantly through real-time image recognition and dynamic partition calibration, and improves the wall thickness control accuracy and stability in combination with regional deviation feature extraction. The associated correction parameters based on sequence pattern mining can automatically generate parameter adjustment rules without manual experience, and can adapt to the changes of raw materials and equipment states through recursive expansion of frequent patterns, realize closed-loop adaptive control of "deviation-correction", and realize intelligent optimization and dynamic adaptation of process parameters. Through the mapping relationship between the die gap distribution adjustment scheme and the local wall thickness deviation, differential correction strategies can be formulated for different regions (such as bottle mouth sealing surface and bottle body bearing area), avoiding the interference of overall adjustment on other regions, and enhancing the directional correction ability of local deviation. The high-precision and intelligent online control of blow molding product wall thickness is realized, and the production efficiency and cost control level are improved.

[0198] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions or the essential part of the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method described in each embodiment or some part of the embodiment.

[0199] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An image recognition based on-line control method of blow molded article wall thickness, characterized by, The method comprises the following steps: Collecting raw material characteristic data, formulating type process parameters according to the shape of the target blow molding product, and calculating wall thickness adjustment parameters according to the wall thickness control method by adjusting the melt extrusion amount and the die gap shape; The step of formulating the type process parameters comprises: Determining performance indicators according to the use of the product, obtaining shape parameters of the target blow molding product, and obtaining the raw material characteristic data by analyzing the basic performance and processing adaptability of the raw material; Calculating the theoretical extrusion amount according to the single weight and production speed of the target product, calculating the initial screw rotation speed in combination with the screw diameter and pitch, and setting the head temperature of different functional segments according to the use function; Setting the blow molding pressure and cooling water temperature according to the raw material characteristic data, calculating the initial gap of the die according to the target wall thickness and blow-up ratio of the product, and correcting the type process parameters by combining the historical optimal parameters of similar products; The step of calculating the wall thickness adjustment parameters comprises: Calculating the melt volume extrusion amount in the current production according to the initial screw rotation speed; Collecting the average die gap, die gap distribution, and die outlet diameter as key parameters of the die; Calculating the target demand extrusion amount and the target demand gap according to the preset target thickness threshold, which is invariant in die shape; Extruding the raw material according to the target demand extrusion amount, and correcting the local wall thickness deviation area according to the target demand gap to obtain a die gap distribution adjustment scheme; Detecting the head pressure data in the blow molding process, calculating the real-time melt viscosity in combination with the initial screw rotation speed, and adjusting the target demand extrusion amount and the target demand gap to obtain the wall thickness adjustment parameters when the viscosity exceeds the preset reference viscosity; Obtaining molding control parameters by adjusting the type process parameters according to the wall thickness adjustment parameters, collecting image data of the target blow molding product in the molding process, and processing the image data using a dynamic partition calibration method to obtain time-series thickness data; Determining whether the time-series thickness data reaches the preset target thickness threshold, maintaining the current molding control parameters if yes, and otherwise optimizing the wall thickness adjustment parameters; Obtaining wall thickness deviation features by feature extraction on the time-series thickness data, analyzing the linkage relationship between the wall thickness deviation features and the type process parameters according to a sequence pattern mining method to obtain associated correction parameters, and optimizing the wall thickness adjustment parameters to obtain wall thickness correction parameters.

2. The image recognition based blown plastic article wall thickness online control method according to claim 1, characterized in that, The step of adjusting the molding control parameters comprises: Constructing a parameter mapping table according to the corresponding association between the wall thickness adjustment parameters and the type process parameters, determining the influence amplitude of the unit change of the type process parameters on the wall thickness, and forming a sensitivity coefficient table; Converting the wall thickness adjustment parameters into adjustment values of the process parameters based on the parameter mapping table and the sensitivity coefficient table; Sorting each parameter in the wall thickness adjustment parameters according to the directness of influence and response speed, and setting the adjustment rate according to the maximum adjustment amount of a single step; Superimposing the adjustment values on the type process parameters, and adjusting according to the equipment safety and process feasibility range to generate the molding control parameters.

3. The image recognition based blown plastic article wall thickness on-line control method according to claim 1, characterized in that, The step of obtaining the time-series thickness data comprises: The sampling frequency is calculated based on the moving speed of the blow molded product and the minimum detection accuracy, and the detection area is divided according to the structure and the circumferential direction of the blow molded product; The image data is collected according to the sampling frequency and the detection area, and pre-processing is performed to obtain pre-processing thickness data; The pre-processing thickness data collected by different sensors is calibrated based on the pulse of the main encoder of the production line as the reference time, the data segment is divided according to the forming cycle for intermittent production, and the abnormal value is removed and the continuity is checked to obtain the time series thickness data.

4. The image recognition based blown parison wall thickness online control method of claim 3, wherein, The steps of extracting the wall thickness deviation feature include: According to the preset target wall thickness threshold as the reference, the absolute deviation and the relative deviation of each sampling point are calculated to define the deviation, and the time series thickness data is sliced to form a regionalized sub-data set according to the detection area; From the perspective of time sequence, the trend feature, the fluctuation feature and the burst feature of the wall thickness deviation are analyzed to obtain the dynamic change rule; From the perspective of the spatial distribution of the blow molded product, the position characteristics of the wall thickness deviation are analyzed to identify local defects and regional differences; The dynamic change rule and the local defects and regional differences are integrated to form the wall thickness deviation feature.

5. The image recognition based blown parison wall thickness online control method of claim 4, wherein, The steps of obtaining the dynamic change rule include: The average deviation and the cumulative deviation in the preset period are calculated, and the change curve of the deviation with time is fitted by linear regression to obtain the trend feature; The standard deviation and the range of the regionalized sub-data set are calculated, and the fluctuation frequency and the corresponding amplitude are extracted by Fourier transform to obtain the fluctuation feature; The deviation peak value and the duration that exceed the preset target wall thickness threshold are extracted as the deviation time, and the deviation change amount of adjacent data points is calculated as the burst feature; The trend feature, the fluctuation feature and the burst feature are integrated to obtain the dynamic change rule.

6. The image recognition based blown parison wall thickness online control method of claim 5, wherein, The steps of obtaining the correlation correction parameter include: The type process parameters and the wall thickness deviation feature are divided into discrete states according to the normal range, and the adjustment action and the burst feature are sequentially coded to form a correlation sequence pair; The PrefixSpan algorithm is used to mine the frequent patterns in the correlation sequence pair; The frequent patterns are converted into process parameter correction rules, including trigger conditions, related process parameters, correction actions and confidence levels; Based on the statistical relationship between the process parameter change amount and the deviation change amount in the frequent patterns, a correction coefficient is calculated; When multiple process parameter correction rules are triggered at the same time, the process parameter correction rule with the maximum correction coefficient is selected as the correlation correction parameter.

7. The image recognition based blown parison wall thickness online control method of claim 6, wherein, The steps of obtaining the frequent patterns include: All the prefix patterns of the items are extracted by traversing all the correlation sequence pairs to calculate the support, and the item prefix patterns exceeding the preset minimum support are formed to form an initial frequent sequence; A projection database is generated for each initial frequent sequence, and all suffix items of the current initial frequent sequence are combined to form candidate item patterns; For each candidate item pattern, the number of occurrences in the projection database is counted, and the frequent item patterns are screened out; For each frequent item pattern, the projection database is continuously constructed, the prefix pattern is recursively expanded, and the long sequence frequent pattern is generated; The short patterns not satisfying the preset length in the long sequence frequent patterns are pruned, and the frequent patterns are sorted by length to form the frequent patterns from short to long.

8. The image recognition based blown parison wall thickness online control method of claim 6, wherein, The step of optimizing the wall thickness correction parameter comprises: A correction mapping table is constructed according to the mapping relationship between the correlation correction parameter and the wall thickness adjustment parameter, and a plurality of adjustment amounts are sorted according to the influence weight of the correlation correction parameter on the wall thickness; Each adjustment amount is adjusted according to the correlation correction parameter and the correction coefficient to obtain a plurality of adjustment parameters; It is judged whether the wall thickness adjusted by the plurality of adjustment parameters reaches the preset target thickness threshold, if yes, the current plurality of adjustment parameters are taken as the wall thickness correction parameter, otherwise, the correction coefficient is recalculated.

Images