A mold forming optimization system based on intelligent control

Abstract

The application specifically relates to the technical field of intelligent control, and discloses a mold forming optimization system based on intelligent control, which comprises a defective product image acquisition module, a defect intelligent identification module, a defect scoring quantification module, a process parameter synchronous acquisition module, a process risk dynamic analysis module and a process correlation closed-loop optimization module. The process parameter synchronous acquisition module is used for acquiring real-time full-process process parameters of an injection molding process. The process risk dynamic analysis module is used for calculating a single-parameter risk index and a comprehensive risk index. When a parameter deviates from an optimal target value, an abnormality is identified, and an early warning is triggered immediately. A defect-process correlation matrix is used for quickly positioning a core cause parameter. A reverse optimization algorithm is used for simulating the influence of different parameter adjustments on a comprehensive defect score within the constraint range of equipment, generating a reverse optimization scheme. A time stamp is used for forming a full-link traceable data chain, which greatly reduces stop-loss time and production cost, and realizes precise risk evaluation and prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, more particularly, the present application relates to a mold forming optimization system based on intelligent control. BACKGROUND

[0002] The mold determines the geometric shape, dimensional accuracy and surface quality of the injection molding product directly through the internal preset cavity, core and other structures, and the rationality of the mold design and the stability of the performance directly affect the production efficiency and the scrap rate. With the rise of automation technology, automated equipment has been gradually introduced into the mold forming field, effectively improving production efficiency and product quality.

[0003] The existing mold forming technology takes the injection molding machine as the core equipment, and the product forming is completed through the basic cooperation of manual and equipment. The mold quality detection focuses on the defect judgment after the finished product is offline, and the process adjustment depends on the experience accumulation of engineers, and the parameter optimization is realized through trial and error and manual adjustment.

[0004] However, when it is actually used, there are still some shortcomings, first, the quality control is lagging, the existing mold forming technology only detects defects through manual or simple equipment after the finished product is offline, cannot identify parameter abnormalities during the injection molding process, and does not establish real-time correlation between process parameters and finished product defects. After the defect is found, it is necessary to trace back to analyze the production data, it is difficult to quickly stop loss, resulting in that the quality problem is not found in time;

[0005] Second, the process adjustment lacks scientificity, the parameter setting of the existing mold forming technology is more dependent on the personal experience of engineers, different engineers have different parameter settings for the same product, resulting in large quality fluctuations of different batches of finished products, low tuning efficiency, and when the product design is changed, trial molding adjustment is required, resulting in prolonged production cycle and increased production cost;

[0006] Third, the risk assessment is extensive, the existing mold forming technology only compares the real-time parameters with the fixed threshold to determine whether the parameters are qualified, cannot quantify the degree of parameter deviation from the optimal value, ignores the synergistic effect of multiple process parameters, cannot identify defects caused by multiple parameters, and data collection is completed independently, lacks precise correlation in time sequence, cannot trace the root cause of defects through data, and cannot predict the quality risk of future production based on historical data. SUMMARY

[0007] Therefore, embodiments of the present application provide a mold forming optimization system based on intelligent control, which realizes real-time injection molding process full-process process parameter through a process parameter synchronous acquisition module, calculates a single parameter risk index and a comprehensive risk index based on a process risk dynamic analysis module, can identify abnormalities in time when the parameter deviates from the optimal target value, can quickly locate the core cause parameter through a defect-process correlation matrix, and effectively solves the problems of quality control lag, lack of scientific process adjustment and rough risk assessment in the background technology.

[0008] To achieve the above object, the present application provides the following technical scheme: a mold forming optimization system based on intelligent control, comprising a defect finished product image acquisition module, a defect intelligent identification module, a defect scoring quantization module, a process parameter synchronous acquisition module, a process risk dynamic analysis module and a process correlation closed-loop optimization module:

[0009] The defect finished product image acquisition module: configure a defect finished product acquisition device, real-time multi-angle acquisition of defect finished product image data of injection molding finished product, and transfer to the defect intelligent identification module;

[0010] The defect intelligent identification module: comprising a defect intelligent identification model, identifying defect correlation data through the defect intelligent identification model, converting the defect correlation data into standard defect text, and transferring to the defect scoring quantization module;

[0011] The defect scoring quantization module: based on the standard defect text, introducing a defect weight coefficient and a defect severity value, calculating a comprehensive defect score, and transferring to the process correlation closed-loop optimization module;

[0012] The process parameter synchronous acquisition module: deploying a core acquisition tool, real-time acquisition of full-process process parameters synchronized with the production cycle of the defect finished product, storing in a time sequence process database, and transferring to the process risk dynamic analysis module;

[0013] The process risk dynamic analysis module: calculating a single parameter risk index based on the full-process process parameters, introducing a parameter weight, calculating a comprehensive risk index, and transferring to the process correlation closed-loop optimization module;

[0014] The process correlation closed-loop optimization module: based on the comprehensive defect score and the risk dynamic analysis result, constructing a defect-process correlation matrix, performing real-time early warning judgment, generating a reverse optimization scheme, and feeding back to the defect scoring quantization module and the process risk dynamic analysis module.

[0015] The technical effects and advantages of the present application are as follows:

[0016] 1. The present application synchronously collects process parameters of the whole process of the real-time injection molding process through a process parameter synchronous acquisition module, calculates single parameter risk indexes and comprehensive risk indexes based on a process risk dynamic analysis module, can identify abnormalities in time when parameters deviate from optimal target values, immediately triggers early warning, quickly locates core cause parameters through a defect-process correlation matrix, and greatly reduces stop-loss time;

[0017] 2. The present application establishes a quantitative mapping model of process parameter risk-defect score through a defect-process correlation matrix, outputs a standardized parameter setting scheme, simulates the influence of different parameter adjustments on comprehensive defect scores within the constraint range of equipment through a reverse optimization algorithm, generates a reverse optimization scheme, reduces the waste rate caused by trial molding, and significantly reduces production cost;

[0018] 3. The present application calculates comprehensive defect scores through a defect score quantification module, quantifies the degree of deviation from optimal target values, quantifies the synergistic effect of multi-parameter risks through a defect-process correlation matrix, synchronizes process parameters, risk indexes, defect texts and comprehensive defect scores through a time stamp, forms a full-link traceable data chain, and realizes precise risk assessment and prediction. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 It is a schematic diagram of the overall structure of the present application.

[0020] Figure 2 It is a schematic diagram of the defect finished product image data acquisition process of the present application.

[0021] Figure 3 It is a schematic diagram of the defect-process correlation matrix construction step of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] As shown in the accompanying drawings Figure 1 A mold forming optimization system based on intelligent control, including a defect finished product image acquisition module, a defect intelligent identification module, a defect score quantification module, a process parameter synchronous acquisition module, a process risk dynamic analysis module and a process correlation closed-loop optimization module.

[0024] The output end of the defective product image acquisition module is connected to the input end of the defect intelligent recognition module, the output end of the defect intelligent recognition module is connected to the input end of the defect scoring quantification module, the output end of the defect scoring quantification module is connected to the input end of the process risk dynamic analysis module, the output end of the process parameter synchronous acquisition module is connected to the input end of the process risk dynamic analysis module, the output end of the defect scoring quantification module and the output end of the process risk dynamic analysis module are connected to the input end of the process correlation closed-loop optimization module, and the output end of the process correlation closed-loop optimization module is connected to the input end of the defect scoring quantification module and the process risk dynamic analysis module and is connected to the injection molding machine controller.

[0025] In a more specific application of the present application, the injection molding machine controller is a special-purpose automatic control unit integrating hardware circuit, control algorithm and human-computer interaction interface, which can real-time regulate and control the key process parameters of pressure, speed, temperature and time in the injection molding process according to the preset program or the instructions output by the process correlation closed-loop optimization module.

[0026] The specific embodiments of the present application include the following contents:

[0027] The defective product image acquisition module: configure a defective product acquisition device to acquire defective product image data of the injection molded product in real time from multiple angles and transmit the data to the defect intelligent recognition module;

[0028] Further, the defective product acquisition device includes a high-resolution industrial camera, a customized light source system, and an image acquisition card and transmission component, and the defective product image data includes core visual data, associated metadata and quality check data.

[0029] In this embodiment, it is specifically pointed out that the high-resolution industrial camera adopts an industrial area camera with a resolution of not less than 5 million pixels, supports a high-speed shutter with a shutter speed of ≥1 / 1000s, avoids image blur caused by product transmission movement, and converts optical signals of the product appearance into digital image data; the customized light source system includes a main light source and an auxiliary light source, the main light source can be a ring-shaped LED light source or a strip-shaped LED light source, the auxiliary light source can be a coaxial light source, the brightness and color temperature of the light source are adjustable, the light source angle and brightness are adjusted to eliminate product surface reflection and shadow; the image acquisition card and transmission component adopt an industrial image acquisition card supporting Camera Link or GigE Vision protocol, matched with a shielded gigabit Ethernet transmission line, the acquisition card is responsible for high-speed decoding and caching of the digital image signals output by the industrial camera to avoid data loss, and the transmission component ensures real-time transmission of the image data at a rate of ≥25 frames / second, matching the beat of the injection molding production line.

[0030] As shown in Figure 2 The acquisition process of the defective product image data is as follows:

[0031] S1.1: Set up a collection trigger mechanism. According to the production cycle signal of the injection molding machine, when the injection molding machine completes a molding, send a molding cycle start signal through an optical sensor to start the image collection process, realize the synchronization of the production cycle and the collection cycle, set up a mechanical positioning clamp at the collection station to ensure that the placement posture and position of the finished product are consistent when it enters the collection area each time;

[0032] S1.2: Set 3-5 fixed collection angles for the high-defect area of the injection molded product, covering the front, side, gate position, parting line, and bottom five key areas. Load the defect product collection equipment on the multi-axis motion platform to complete all angle image shooting within 2-3 seconds, collect 1-2 images at each angle, and the collection efficiency matches the production line beat;

[0033] S1.3: After collection is completed, automatically verify the clarity and brightness of each image. If there are unqualified images such as blur, overexposure, and underexposure, trigger re-collection. If collection fails for three consecutive times, send an alarm signal to the injection molding machine controller to remind the operator to check the equipment.

[0034] It should be further explained that the core visual data is for the key appearance of the injection molded product, i.e. the position where defects are prone to occur. According to historical data, collect injection molded product images from different angles, including front images, side images, and gate position images. Defects in the front image include but are not limited to shrinkage, bubbles, and color difference; defects in the side image include but are not limited to warping and flash; the gate position image focuses on the starting point of the melt filling and the defects in the surrounding area, including but not limited to material shortage and depression;

[0035] The associated metadata includes production cycle identification information and collection scene information. The production cycle identification information specifically includes production batch number, production cycle serial number, and injection molding machine number. The collection scene information specifically includes collection angle label and collection timestamp;

[0036] Quality verification data includes clarity parameters, brightness parameters, and lighting parameters. The clarity parameter refers to the clarity index calculated by the contrast of image gray value, for example, contrast ≥ 0.8, which is used to judge whether the image is blurred to avoid missing defects due to image blur. The brightness parameter refers to the average brightness value of the image, for example, brightness value 200-255 cd / m 2 The lighting parameter refers to the color temperature of the light source, for example, 5000K, which is used to confirm whether the image is overexposed or underexposed to ensure that the image truly presents the defect characteristics.

[0037] Defect intelligent recognition module: includes a defect intelligent recognition model, which identifies defect associated data through the defect intelligent recognition model, converts the defect associated data into standard defect text, and transmits it to the defect scoring quantization module;

[0038] Further, the defect-related data includes defect type, defect severity level, and defect position information, the defect severity level includes L1, L2, and L3, and the standard defect text adopts the string format of “defect type-severity level-position information”.

[0039] In the embodiment, it needs to be particularly pointed out that the construction of the defect recognition type needs to select a deep convolutional neural network ResNet-50 / ResNet-101, extract the subtle features of the defects in the injection product image according to the residual connection structure of the network, and add a multi-label classification output layer after the ResNet base network in view of the characteristics that multiple types of defects coexist in the injection product, collect a large number of labeled images covering common defect types of the injection product, and manually label the defect type, defect severity level, and defect position information of each picture to form a training data set, use the binary cross-entropy loss function to adapt to the multi-label classification task, select the AdamW optimizer, suppress overfitting through weight decay, introduce the learning rate cosine annealing strategy to improve the model convergence speed and recognition accuracy, and ensure that the defect type recognition accuracy is ≥95% and the position positioning error is ≤0.5 mm.

[0040] It needs to be further pointed out that the defect type recognition refers to extracting the pixel-level features in the defect product image data through the defect intelligent recognition model, matching the feature templates in the training data set, and outputting the 1-3 defect types with the highest probability; the defect severity level is the quantitative features of the defect intelligent recognition model, including area, length, and depth, which are compared with the preset threshold to judge the corresponding defect severity level, which is divided into area dimension, influence range, and morphological characteristics, for example, the length of the flash is less than 1 mm for L1, 1-3 mm for L2, and greater than 3 mm for L3, the influence range is, for example, the shrinkage area accounts for less than 5% of the surface area of the product for L1, 5%-15% for L2, and greater than 15% for L3, and the morphological characteristics are, for example, the maximum gap between the product and the reference surface caused by warping is less than 0.1 mm for L1, 0.1-0.3 mm for L2, and greater than 0.3 mm for L3; the defect position information is the specific coordinate range of the defect in the image framed by the target detection algorithm of the defect intelligent recognition model, which is converted into the actual physical position description of the product.

[0041] The defect scoring quantification module: based on the standard defect text, the defect weight coefficient and the defect severity value are introduced to calculate the comprehensive defect score, which is transmitted to the process-related closed-loop optimization module;

[0042] Further, the acquisition of the comprehensive defect score is based on the defect weight coefficient Wi and the defect severity value Li of a single defect to calculate the single defect score, and the weighted sum of all single defect scores of a single injection product in a production cycle is calculated through the formula The comprehensive defect score DS is calculated, and n is the number of defects of the injection molded product.

[0043] In this embodiment, it needs to be specifically pointed out that the defect weight coefficient is determined according to expert experience and historical loss cost. The expert experience refers to that the injection molding industry process experts or quality engineers score according to the influence degree of various defects on product function, appearance and assembly performance. The historical loss cost refers to that the economic loss caused by various defects in the past 1-3 years of production data is adjusted according to the loss ratio. The defect severity value corresponds to the defect severity level, L1=1, L2=2, L3=3. The defect types include material shortage, shrinkage, flash, warpage and weld line, and the corresponding weights are 0.8, 0.6, 0.2, 0.7 and 0.3 respectively.

[0044] It needs to be further pointed out that the higher the comprehensive defect score DS is, the worse the quality of the injection molded product is. DS belongs to 0-1 as a high-quality product without obvious defects. DS belongs to 1-3 as a qualified product with moderate defects but without affecting the core function. DS is greater than 3 as an unqualified product with serious defects, which needs to be scrapped or reworked.

[0045] The process parameter synchronous collection module: deploy the core collection tool, collect the whole-process process parameters synchronized with the production cycle of the defective product in real time, store them into the time sequence process database, and transfer them to the process risk dynamic analysis module;

[0046] Further, the core collection tool includes an injection molding machine controller data interface, a high-frequency sensor and a time sequence process database. The core collection tool collects the whole-process process parameters of the injection molded product production based on a synchronous trigger signal mechanism. The whole-process process parameters include pressure parameters, speed parameters, temperature parameters and time parameters.

[0047] In this embodiment, it needs to be specifically pointed out that the injection molding machine controller data interface adopts an OPC UA industrial communication protocol interface, directly reads the real-time process parameters stored in the injection molding machine controller, does not need to make large-scale modification to the injection molding machine, and is suitable for the injection molding machine controllers of mainstream brands.

[0048] The high-frequency sensor is a high-frequency industrial sensor deployed to supplement the key parameters that cannot be directly collected by the injection molding machine controller, including a cavity pressure sensor, a melt temperature sensor and a multi-axis speed sensor. The cavity pressure sensor is installed inside the mold cavity to directly collect the pressure change of the melt in the cavity, with an accuracy of ±0.1 MPa and a sampling frequency of ≥100 Hz. The melt temperature sensor is installed at the nozzle of the injection molding machine to collect the temperature of the melt before it comes out of the mold in real time, with an accuracy of ±1℃ and a sampling frequency of 50 Hz. The multi-axis speed sensor is installed at the screw drive mechanism of the injection molding machine to collect the multi-stage injection speed with an accuracy of ±1 mm / s and a sampling frequency of 100 Hz.

[0049] The time series process database refers to a deployed industrial-grade time series database, such as InfluxDB, used to store process parameter data with timestamps, support high-concurrency writing (≥1000 / s) and fast time series query, and adapt to the continuous production rhythm of the injection molding production line.

[0050] It should be further explained that the synchronization trigger signal mechanism specifically refers to taking the molding cycle start signal of the injection molding machine as the reference trigger point. When the injection molding machine completes demolding and a new molding cycle starts, the injection molding machine controller sends a molding cycle start signal to the process parameter synchronization acquisition module through the OPC UA interface, and attaches a unique production cycle serial number. The process parameter synchronization acquisition module triggers the injection molding machine controller data interface and sensors, adopts differentiated acquisition frequencies according to the time lengths of each stage of the injection molding process, uses high-frequency acquisition 50-100Hz in the injection and holding stages to ensure capturing transient fluctuations in pressure and speed, and uses low-frequency acquisition 1-5Hz in the cooling stage to reduce data redundancy and acquire full-process process parameters.

[0051] The pressure parameters in the full-process process parameters include screw head pressure, cavity pressure, and holding pressure. The cavity pressure is the preferred acquisition item, directly reflecting the filling state of the melt in the cavity. The holding pressure needs to record the real-time change value in the holding stage. The speed parameters include filling stage speed, holding stage speed, and end-of-injection speed. According to the speed stages set by the injection molding machine, 3-5 data points are acquired in each stage to form a complete speed curve. The temperature parameters include barrel feeding section temperature, melting section temperature, metering section temperature, mold temperature, and melt temperature. The barrel temperature needs to acquire the actual temperature of each heating area, and the mold temperature needs to acquire the temperature on both sides of the moving mold and the fixed mold. The time parameters include holding time, holding switch position, injection time, cooling time, and production cycle time. The holding switch position records the specific travel unit of the screw from the injection stage to the holding stage, which is mm. The cycle time is the total time consumed for a single molding.

[0052] Process risk dynamic analysis module: based on the full-process process parameters, calculate the single-parameter risk index, introduce the parameter weight, calculate the comprehensive risk index, and transfer to the process correlation closed-loop optimization module;

[0053] Further, the calculation of the single-parameter risk index needs to obtain the real-time value X j of each parameter in the full-process process parameters, the optimal target value XT j corresponding to each parameter in the process standard database, the maximum allowable running value USL j , and the minimum allowable running value LSL j , through the formula:

[0054] ,

[0055] The single-parameter risk index R of each parameter is calculated j , j represents the jth parameter in the whole-process parameter, the single-parameter risk index of each parameter and the parameter weight V j The comprehensive risk index R is calculated by the formula:

[0056] ,

[0057] The comprehensive risk index R is calculated by the formula: j The sum of all V j is 1, the initial parameter weight is obtained by principal component analysis, and a parameter dynamic adjustment mechanism is constructed to adjust the parameter weight.

[0058] In this embodiment, it is particularly pointed out that the real-time value of the pressure parameter is the average value in the cycle, the real-time value of the speed parameter is the stage average of the multi-stage injection speed curve, the real-time value of the temperature parameter is the cycle average of the barrel feeding section temperature, the melting section temperature, the metering section temperature, the mold temperature and the melt temperature, and the real-time value of the time parameter is the actual length of the holding pressure time, the injection time and the cooling time; The process standard database is set by the material supplier and the process engineer, the optimal target value is the ideal value of the parameter operation, the allowed maximum value and the allowed minimum value are the range of parameter fluctuation, and the parameter is out of control if it exceeds the range, and the range of single-parameter risk index is 0-1.

[0059] The initial parameter weight is set by principal component analysis, data analysis is performed on the single-parameter risk index and the comprehensive defect score of the history of 3-6 months, the first N parameters with the greatest impact in the comprehensive defect score are extracted, for example, the holding pressure, the cavity pressure and the melt temperature, and higher initial weights are given, for example, 0.3, 0.25 and 0.2, and lower weights are given to secondary parameters, for example, the initial weight of the cooling time is 0.05;

[0060] The parameter dynamic adjustment mechanism updates the weight coefficient based on the parameter adjustment effect data fed back by the subsequent process correlation closed-loop optimization module, for example, if the holding pressure is adjusted for several times and the decrease amplitude of the comprehensive defect score is significantly higher than that of adjusting other parameters, the weight of the holding pressure is increased, for example, from 0.3 to 0.35; if the injection speed adjustment has little effect on the comprehensive defect score, the weight is reduced, for example, from 0.15 to 0.1, to ensure that the weight matches the actual impact of the parameter.

[0061] The process correlation closed-loop optimization module: based on the comprehensive defect score and the risk dynamic analysis result, a defect-process correlation matrix is constructed, real-time early warning judgment is performed, a reverse optimization scheme is generated, and feedback is fed back to the defect score quantification module and the process risk dynamic analysis module.

[0062] Further, as shown in Figure 3 , the construction steps of the defect-process correlation matrix are as follows:

[0063] S2.1: Obtain the comprehensive defect score in the defect score quantification module and the standardized defect text, obtain the single-parameter risk index and the comprehensive risk index in the process risk dynamic analysis module, associate the two types of data through the time stamp, and form a sample data set;

[0064] In this embodiment, it needs to be specifically pointed out that the number of samples in the sample data set needs to be accumulated to more than 100,000, covering different defect types and process parameter fluctuation scenarios, and the number of samples corresponding to each common defect text, such as shrinkage-L2-gate, lack of material-L3-cavity, is not less than 500.

[0065] S2.2: Select a machine learning algorithm, divide the sample data set into a training set and a test set according to a ratio of 8:2, take the single-parameter risk index as an input variable and the comprehensive defect score as an output variable, and train a defect-process quantification prediction model;

[0066] In this embodiment, it needs to be specifically pointed out that the single-parameter risk index is a core process variable affecting defects, the comprehensive defect score is a quantitative index reflecting quality results, the machine learning algorithm includes random forest, gradient boosting tree and neural network, among which the random forest is suitable for processing nonlinear relationship and can output the contribution of each R j to the DS, the gradient boosting tree is good at processing high-dimensional data and improving the prediction accuracy in the small sample scenario, and the neural network is suitable for the scenario with a large amount of data and can capture more complex parameter interaction; the verification of the defect-process quantification prediction model refers to calculating the deviation rate of the predicted value DS1 and the actual value DS to evaluate the model accuracy, and the deviation rate is required to be less than 5%; if the accuracy is not up to standard, the model is optimized by increasing the sample size and adjusting the algorithm parameters.

[0067] S2.3: Extract the number of standard defect texts as the row of the association matrix, extract the number of process parameters as the column of the association matrix, obtain the contribution of the single-parameter risk index to the defect text as the numerical value of the matrix cell based on the defect-process quantification prediction model, and verify and dynamically update the defect-process association matrix.

[0068] In this embodiment, it needs to be specifically pointed out that the number of standard defect texts needs to remove the number of repeated defect texts as the row of the defect-process correlation matrix, and the value of the matrix cell is, for example, the feature importance of the random forest model output R_holding pressure to shrink-L2-close-to-gate is 0.8, then the corresponding cell in the matrix is filled with 0.8, which represents that the contribution of the holding pressure risk to the shrink defect reaches 80%; verifying and dynamically updating the defect-process correlation matrix specifically means selecting 10% of the validation set samples, substituting them into the correlation matrix to predict the defect text, comparing with the actual defect text, verifying the accuracy of the matrix correlation, and requiring the matching rate to be ≥90%, with the continuous accumulation of new production samples, repeating the process of S2.1-S2.3, retraining the model and updating the correlation value in the matrix, to ensure that the matrix can adapt to the changes of process, material and mold.

[0069] It needs to be further pointed out that real-time early warning judgment specifically means that based on the current production cycle comprehensive risk index and single parameter risk index, the early warning threshold set according to the enterprise quality standard and production experience, for example, R≥0.6 is high-risk early warning, 0.3≤R<0.6 is medium-risk early warning, the threshold can be dynamically adjusted through historical data statistics, for example, when R≥0.6, the defect rate is more than 15%; when the real-time R exceeds the set threshold, the system immediately triggers the early warning, and pushes the early warning information through the workshop display screen and the operator terminal, including the current R value, risk level and production cycle serial number, reminding the relevant personnel to pay attention.

[0070] Further, the generation steps of the reverse optimization scheme are as follows:

[0071] S3.1: determine the process parameters that make the comprehensive risk index R drop below the early warning threshold, while ensuring that the predicted comprehensive defect score DS1<1, and set the constraint condition;

[0072] S3.2: select optimization algorithm to simulate the influence of different process parameter adjustment combinations on R and DS1, preferentially adjust the core process parameters with high contribution in the defect-process correlation matrix, and output the optimization scheme;

[0073] S3.3: the optimization scheme is automatically issued to the injection molding machine controller through the OPC UA interface, and the injection molding machine controller receives the instruction and executes the new parameters immediately after the current production cycle ends.

[0074] In this embodiment, it needs to be specifically pointed out that the early warning threshold can be 0.3, and the constraint condition specifically refers to that the parameter adjustment needs to meet the physical limit of the injection molding machine equipment, for example, the maximum adjustable pressure holding pressure is 100MPa, the minimum is 70MPa, the material property requirement, for example, the melt temperature cannot exceed the material decomposition temperature, and the production efficiency constraint, for example, after the cooling time adjustment, the production cycle cannot be extended by more than 10%; the optimization algorithm can be a constrained nonlinear programming algorithm, for example, a sequential quadratic programming method SQP, or a greedy optimization algorithm based on a correlation matrix; the form of the optimization scheme is parameter adjustment suggestion + expected effect, for example, parameter adjustment suggestion: the pressure holding pressure is increased from 80MPa to 85MPa; the mold temperature is reduced from 65℃ to 62℃; the pressure holding time is extended from 5s to 5.2s; the expected effect: the comprehensive risk R is reduced from 0.7 to 0.25, the shrinkage-L2-close-to-gate defect probability is reduced from 85% to 5%, and DS1≤0.8.

[0075] It needs to be further explained that the parameter adjustment effect data is collected after the execution of the optimization scheme, including quality end data and process end data, the quality end data includes new cycle product images, standardized defect text and comprehensive defect score, the process end data includes new cycle process parameter real-time value, single parameter risk index and comprehensive risk index; the comprehensive risk index deviation and the comprehensive defect score deviation are calculated, the deviation reasons are traced back through the defect-process correlation matrix, and are respectively transmitted to the defect score quantification module and the process risk dynamic analysis module, and the correlation matrix, the single parameter weight V of risk calculation and the defect score weight parameter W are iterated.

[0076] Secondly: in the drawings of the disclosed embodiments of the present application, only the structures related to the disclosed embodiments of the present application are involved, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0077] Finally: the above only describes the preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A mold forming optimization system based on intelligent control, characterized in that, It includes a defective finished product image acquisition module, a defect intelligent identification module, a defect scoring and quantification module, a process parameter synchronous acquisition module, a process risk dynamic analysis module, and a process correlation closed-loop optimization module. Defective Finished Product Image Acquisition Module: Configured with defective finished product acquisition equipment, it acquires defective finished product image data of injection molded products from multiple angles in real time and transmits it to the defect intelligent recognition module; The intelligent defect identification module includes an intelligent defect identification model, which identifies defect-related data, converts the defect-related data into standard defect text, and transmits it to the defect scoring and quantification module. Defect scoring quantification module: Based on standard defect text, introduce defect weight coefficient and defect severity value, calculate comprehensive defect score, and pass it to process-related closed-loop optimization module; Process parameter synchronous acquisition module: Deploy core acquisition tools to collect full-process process parameters in real time, synchronized with the production cycle of defective finished products, store them in the time-series process database, and transmit them to the process risk dynamic analysis module; The core acquisition tool includes an injection molding machine controller data interface, a high-frequency sensor, and a time-series process database. The core acquisition tool collects the process parameters of the entire injection molding process based on a synchronous trigger signal mechanism. The process parameters include pressure parameters, speed parameters, temperature parameters, and time parameters. The process risk dynamic analysis module calculates the single-parameter risk index based on the process parameters of the entire process, introduces parameter weights, calculates the comprehensive risk index, and transmits it to the process correlation closed-loop optimization module. The calculation of the single-parameter risk index requires obtaining the real-time value X of each parameter in the entire process parameters. j Obtain the optimal target value XT for each parameter in the process standard database. j Maximum allowed running value USL j and allow running minimum LSL values j Through the formula: ， The single-parameter risk index R for each parameter is calculated. j Let j represent the j-th parameter in the overall process parameters, based on the single-parameter risk index and parameter weight V for each parameter. j Through the formula: ， The comprehensive risk index R is calculated, where all V j The sum is 1. The initial parameter weights are obtained through principal component analysis, and a dynamic parameter adjustment mechanism is constructed to adjust the parameter weights. Process-related closed-loop optimization module: Based on the comprehensive defect score and risk dynamic analysis results, a defect-process correlation matrix is ​​constructed, real-time early warning judgment is performed, reverse optimization scheme is generated, and feedback is given to the defect score quantification module and the process risk dynamic analysis module; The steps for constructing the defect-process correlation matrix are as follows: S2.1: Obtain the comprehensive defect score and standardized defect text from the defect scoring quantification module, and obtain the single-parameter risk index and comprehensive risk index from the process risk dynamic analysis module. Link the two types of data through timestamps to form a sample dataset. S2.2: Using the single-parameter risk index as the input variable and the comprehensive defect score as the output variable, a machine learning algorithm is selected, and the sample dataset is divided into a training set and a test set in an 8:2 ratio to train the defect-process quantitative prediction model. S2.3: Extract the number of standard defect texts as rows of the association matrix, extract the number of process parameters as columns of the association matrix, obtain the contribution of the single-parameter risk index to the defect text based on the defect-process quantitative prediction model as the value of the matrix cell, verify and dynamically update the defect-process association matrix. The steps for generating the reverse optimization scheme are as follows: S3.1: Determine the process parameters that reduce the comprehensive risk index R to below the warning threshold, while ensuring that the predicted comprehensive defect score DS1 < 1, and set constraints. S3.2: Select an optimization algorithm to simulate the impact of different combinations of process parameter adjustments on R and DS1, prioritize adjusting the core process parameters with high contribution in the defect-process correlation matrix, and output the optimization scheme; S3.3: The optimization plan is automatically sent to the injection molding machine controller via the OPC UA interface. After receiving the instruction, the injection molding machine controller immediately executes the new parameters after the current production cycle ends.

2. The mold forming optimization system based on intelligent control according to claim 1, characterized in that: The defective product acquisition equipment includes a high-resolution industrial camera, a customized light source system, and an image acquisition card and transmission components. The defective product image data includes core visual data, associated metadata, and quality verification data.

3. The mold forming optimization system based on intelligent control according to claim 1, characterized in that: The defect-related data includes defect type, defect severity level, and defect location information. Defect severity levels include L1, L2, and L3. The standard defect text adopts the string format of "defect type-severity level-location information".

4. The mold forming optimization system based on intelligent control according to claim 1, characterized in that: The comprehensive defect score is obtained by calculating the individual defect score based on the defect weight coefficient Wi and the defect severity value Li of each individual defect. The scores of all individual defects for a single injection-molded product within a production cycle are then weighted and summed using the formula... The comprehensive defect score DS is calculated, where n is the number of defects in the injection molded product.

Images