Machine Vision-Based Adaptive System and Method for One-Piece Screw Injection Molding of Plastic Cover Plates
The machine vision-based adaptive system for screw-injection molded plastic cover plates utilizes grayscale co-occurrence matrix contrast monitoring to detect texture features. By combining mechanism and data-driven logic, it solves the quality fluctuations caused by mold temperature drift and screw wear, achieves stable control of multi-defect coupling, and improves the quality stability and detection accuracy of injection molding.
Patent Information
- Application Number
- CN202611235237.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-17
AI Technical Summary
Existing machine vision-assisted injection molding solutions cannot identify and counteract long-term process fluctuations such as slow mold temperature drift and progressive screw wear, resulting in poor quality stability in continuous production. Furthermore, parameter adjustments are contradictory in multi-defect coupling scenarios, the detection rate of weak defects in hidden locations is low, the control model lacks physical mechanism constraints, and the reliability of industrial field implementation is insufficient.
An adaptive system for screw injection molding of integrated plastic cover plates using machine vision includes an active vision imaging module, a slow-loop process drift compensation module, and a fast-loop coupled defect control module. By monitoring texture features through gray-level co-occurrence matrix contrast and combining mechanism and data-driven logic, it generates process parameter corrections to achieve compensation for mold temperature drift and screw wear, as well as real-time control of defects.
It significantly extends the quality stability window for continuous production, is suitable for one-piece molded plastic covers with stringent requirements for dimensional consistency and surface finish, effectively decouples multiple defect couplings and safely and reliably corrects parameters, thereby improving production stability and inspection accuracy.
Smart Images

Figure CN123103992A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of injection molding quality control technology, and more specifically, to a machine vision-based adaptive system and method for screw injection molding of integrated plastic cover plates. Background Technology
[0002] One-piece molded plastic covers are widely used in consumer electronics, automotive interiors, and home appliance housings. These products typically integrate complex structures such as snap-fits, reinforcing ribs, undercuts, and thin hinges, and are formed in a single injection molding process. They require high levels of surface finish, dimensional accuracy, and structural strength. During the injection molding process, multiple factors, including mold temperature drift, screw wear, raw material batch fluctuations, and changes in ambient temperature and humidity, can easily lead to surface and structural defects such as shrinkage marks, flash, warping, weld lines, and material shortages. Furthermore, due to the complex structure of the covers, multiple defects often occur simultaneously, further increasing the difficulty of process control. Existing publication number CN114565614B discloses a method and system for analyzing surface defects of injection molded parts based on machine vision. The method includes: acquiring a grayscale image of the injection molded part's surface; extracting features from the grayscale image to obtain defect areas; marking the location of the injection molded part's gate; calculating the gradient value and gradient direction of each pixel within the defect area; obtaining the lines connecting each pixel within the defect area to the location of the injection molded part's gate; obtaining the angle between the gradient direction of each pixel within the defect area and the connecting lines, denoted as the air mark extension angle; statistically analyzing the binary tuple formed by the gradient value and the air mark extension angle of each pixel within the defect area to obtain a defect evaluation value; calculating a defect classification coefficient based on the defect evaluation value; and determining the defect type of the defect area according to the defect classification coefficient and a set range. This method can accurately analyze the causes of air mark defects, enabling workers to specifically handle injection molded parts with air mark defects. In the process of realizing this application, the inventors discovered the following problems with the existing technology: Existing machine vision-assisted injection molding solutions are mainly used for defect detection and defect sorting. Most of the control solutions adopt a single-mold adjustment and a closed-loop structure that adjusts in a timely manner according to problems that occur during production. However, this cannot identify and counteract long-term process fluctuations such as slow mold temperature drift and progressive screw wear, resulting in poor quality stability in continuous production. For multi-defect coupled scenarios, the independent adjustment logic of one parameter corresponding to one defect is used, which cannot locate the common root cause of defects and is prone to contradictory parameter adjustments and the waxing and waning of defects. Visual inspection adopts a static imaging mode with fixed posture and fixed light source, resulting in a low detection rate of weak defects in hidden locations. The control models are mostly pure data-driven black box models, lacking physical mechanism constraints, and have insufficient reliability for industrial field implementation. Therefore, to address the above issues, a machine vision-based adaptive system and method for screw injection molding of integrated plastic cover plates is proposed. Summary of the Invention
[0003] In order to overcome the above-mentioned defects of the prior art, this application provides a machine vision-based adaptive system and method for screw injection molding of one-piece plastic cover plates to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, this application provides the following technical solution: a machine vision-based adaptive system for screw injection molding of integrated plastic cover plates, applied to an injection molding machine with a screw injection unit. The system includes an active vision imaging module, a slow-loop process drift compensation module, a fast-loop coupling defect control module, and a collaborative control execution module. The active vision imaging module, the slow-loop process drift compensation module, and the fast-loop coupling defect control module are all signal-connected to the collaborative control execution module. The active vision imaging module is used to acquire images of the surface of the integrally molded plastic cover plate after the screw injection molding demolding, extract defect classification features and underlying texture features, and provide visual data for molding process drift compensation and defect control. The slow-loop process drift compensation module is used to generate a screw injection molding process parameter trend compensation amount based on the long-term change trend of the underlying texture features, so as to offset the chronic process fluctuations caused by mold temperature drift and screw progressive wear. The fast-loop coupling defect control module is used to decouple the root cause process parameters of multiple defects based on defect classification characteristics, and generate real-time correction values for screw injection molding process parameters; the collaborative control execution module is used to integrate trend compensation values and real-time correction values, output the final molding process parameters and send them to the screw injection molding machine for execution.
[0005] Preferably, the active vision imaging module includes an imaging execution unit, a light source control unit, and a visual feature extraction unit; wherein: The imaging execution unit is used to grasp the demolded cover plate with a six-axis robotic arm and adjust the imaging posture, and work with an industrial area scan camera to complete the acquisition of global and local images. The light source control unit is used to switch between three sets of light sources: 0° coaxial light, 45° diffuse light, and 75° grazing light, to match the lighting requirements of different detection scenarios. The visual feature extraction unit is used to extract defect classification features and underlying texture features from the acquired image. The underlying texture features are quantized using gray-level co-occurrence matrix contrast.
[0006] Preferably, the slow-loop process drift compensation module includes a trend statistics unit, a drift determination unit, and a slow-loop compensation generation unit; wherein: The trend statistics unit is used to statistically analyze the changing trend of the underlying texture features with a fixed module window and calculate the feature trend slope. The drift determination unit is used to compare the feature mean with the normal range and combine it with the trend slope threshold to determine whether there is a systematic process drift. The slow loop compensation generation unit is used to generate a small-amplitude cumulative slow loop parameter compensation amount based on the mapping relationship between visual features and process parameters.
[0007] Preferably, the fast-loop coupling defect control module includes a defect quantification unit, a root cause decoupling unit, and a fast-loop correction generation unit; wherein: The defect quantification unit is used to calculate the overall severity of various appearance defects and generate a defect severity vector. The root cause decoupling unit is used to calculate the comprehensive influence of each process parameter based on the defect and process causal weight matrix specific to the integrally molded cover plate, and to sort and determine the root cause parameters; the fast-loop correction generation unit is used to adopt mechanism and data dual-driven logic, first delineate the safe and feasible domain by the mechanism model, and then output the correction amount and limit the amplitude by case reasoning to obtain the final fast-loop correction value.
[0008] A machine vision-based adaptive molding method for one-piece plastic cover plate screw injection molding, applied to an injection molding machine with a screw injection unit, includes the following steps: Step S1: Complete the construction and initial parameter calibration of the screw injection molding system, build a vision inspection platform and establish process parameter mapping relationships, and pre-build a basic molding mechanism model and case library; Step S2: After the screw injection molding demolding, active vision dynamic imaging is performed to extract the defect classification features and underlying texture features of the cover plate; Step S3: Based on the underlying texture features, perform slow-loop process drift trend compensation to generate screw injection molding process parameter trend compensation amount; Step S4: Based on defect classification features, perform fast-loop coupling defect decoupling control to generate real-time correction values for screw injection molding process parameters; Step S5: Integrate the slow-loop trend compensation amount and the fast-loop real-time correction amount to generate the final molding process parameters and send them to the screw injection molding machine for application in the next injection molding cycle; Step S6: Collect visual data after the next demolding cycle for closed-loop feedback verification, and repeatedly execute the molding control process.
[0009] Preferably, step S1 further includes: Step S11: Install an industrial area scan camera, multiple independently controlled strip light sources, and a six-axis material handling robot on the discharge side of the screw injection molding machine, and connect them to the central controller to form a vision inspection platform; Step S12: Establish an industrial communication connection between the central controller and the screw injection molding machine control system, and map the register addresses, ranges and adjustment accuracy of adjustable process parameters such as injection speed, holding pressure, holding pressure switching position, mold temperature, and cooling time. Step S13: Combine the structural features of the cover plate with the PVT characteristics of the material to construct a basic molding process mechanism model, define the safe adjustment boundary of the screw injection parameters and the molding process window, pre-store the historical optimal parameter adjustment cases corresponding to typical defect combinations, and form an initial case library.
[0010] Preferably, step S2 further includes: Step S21: The robotic arm picks up the cover plate after the screw injection molding and demolding and moves it to the standard imaging station. The cover plate is captured with 45° diffuse light to obtain a full-area frontal image and complete the initial overall appearance inspection. Step S22: Retrieve the process parameter deviation and defect detection results of the previous mold, predict the high-risk defect area and defect type of the current mold, control the robot to flip the cover plate and switch the 75° grazing light to perform local high-precision imaging of the high-risk area. Step S23: Extract defect classification features from the image for fast loop control, and simultaneously select a fixed sampling window for the cover plate planar region. Calculate the gray-level co-occurrence matrix contrast CON as the underlying texture feature for slow loop compensation. The calculation formula is as follows: ; in, The number of gray levels in the image. The grayscale index of the two pixels. This is the joint probability element of the gray-level co-occurrence matrix under a specified pixel interval and statistical direction.
[0011] Preferably, step S3 further includes: Step S31: Using N consecutive moduli as a statistical window, fit the changing trend of the gray-level co-occurrence matrix contrast within the window using the least squares method, and calculate the characteristic trend slope. ; Step S32: When the mean value of the features within the window exceeds the preset normal range and the absolute value of the trend slope is greater than the slope threshold, it is determined that there is a systematic process drift. Step S33: Generate the slow-loop compensation amount based on the linear mapping relationship between the characteristic slope and the process parameters. Taking mold temperature compensation as an example, the single-cycle slow-loop temperature compensation amount is... The calculation formula is: ; in, This is the conversion coefficient from contrast feature slope to mold temperature compensation; the amplitude of a single slow loop compensation does not exceed the total range of the corresponding parameter. The basic forming process parameters are corrected in an cumulative manner.
[0012] Preferably, step S4 further includes: Step S41: For the identified defects, combine the defect area ratio... With contrast level Calculate the first Overall severity of class defects Construct a defect severity vector, the calculation formula of which is: ; Step S42: Call the defect and process causal weight matrix specific to the one-piece molded cover plate to calculate the comprehensive influence of each process parameter. The root cause parameters are determined by sorting them from largest to smallest impact, and the formula for calculating the overall impact is as follows: ; in, This represents the total number of defect types currently identified. For the first Class defects and the first Causal correlation weights for process parameters; Step S43: The corrected safety upper and lower bounds of the root cause parameters are output from the basic molding process mechanism model. Within the safe and feasible region, the historical optimal parameter tuning values are retrieved through case similarity matching. After weighted calculation and amplitude limiting, the fast loop real-time correction amount is obtained.
[0013] Preferably, steps S5 and S6 further include: Step S51: Set the basic molding process parameters Real-time correction amount of fast loop Compensation amount for slow-cycle trend By superimposing and synthesizing, the final molding process parameters are obtained. The calculation formula is as follows: ; After passing the safety verification of the full-parameter molding process window, the data is sent to the screw-type injection molding machine for use in the next injection molding cycle. Step S61: Collect visual data of the cover plate after the next demolding, calculate the control effect coefficient to evaluate the effectiveness of parameter adjustment, and if the defects worsen, trigger a safety rollback to restore the previous stable molding process parameters. Step S62: If the current mold defect deviation is less than the dead zone threshold and there is no obvious process drift trend, then keep the current molding process parameters unchanged and enter the next cycle of detection.
[0014] The technical effects and advantages of this application are as follows: 1. Compared with existing technologies, this machine vision-based adaptive system and method for one-piece injection molded plastic covers introduces gray-level co-occurrence matrix contrast as the underlying texture feature. It continuously monitors the micro-texture roughness of the planar area of the injection molded cover through a fixed window, fits the feature trend slope using the least squares method, and determines the existence of systematic process drift by combining mean deviation and slope threshold. Once drift is confirmed, the system generates trend compensation in small steps using cumulative logic, with a single compensation not exceeding 5% of the total parameter range, thus avoiding process oscillation. This slow-loop compensation method does not require additional temperature or displacement sensors; it can reverse-engineer the gradual direction and amplitude of physical quantities such as mold temperature and screw status using only visual features. It quantifies previously invisible chronic interference into controllable process parameter corrections, significantly extending the quality stability window for continuous production. It is particularly suitable for one-piece injection molded plastic covers with stringent requirements for dimensional consistency and surface finish.
[0015] 2. Compared with existing technologies, this machine vision-based adaptive system and method for screw injection molding of one-piece plastic covers addresses the common problem of multiple defects such as shrinkage marks, flash, and warping in one-piece molded plastic covers. If each defect is addressed with a single parameter, conflicts between parameter settings can easily arise, causing defects to increase or decrease simultaneously. Therefore, this system constructs a defect-process causal weight matrix for the plastic cover. By calculating the comprehensive severity vector of various defects, it weights and synthesizes the comprehensive influence of each process parameter, and sorts and locks the root cause parameters leading to the coupling of multiple defects. Furthermore, when generating correction values, a dual-drive mechanism of mechanistic model and case-based reasoning is employed. First, the basic molding mechanism model defines the upper and lower bounds of safe correction based on the material's PVT characteristics and equipment constraints. Then, within the safe and feasible domain, historical best cases are retrieved, and the initial correction value is output using cosine similarity weighting. Finally, the final fast-loop correction value is obtained after amplitude limiting. Attached Figure Description
[0016] Figure 1 This is an overall flowchart of the adaptive molding method of this application; Figure 2 This is a diagram of the fast and slow loop coordinated control system architecture of this application; Figure 3 This is a schematic diagram of the active vision dynamic imaging structure of this application; Figure 4 This is a block diagram of the mechanism and data-driven decoupled control logic of this application. Detailed Implementation
[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0018] Example 1 As attached Figures 1 to 4 The machine vision-based adaptive system for screw injection molding of integrated plastic cover plates includes an active vision imaging module, a slow-loop process drift compensation module, a fast-loop coupling defect control module, and a collaborative control execution module. The active vision imaging module, the slow-loop process drift compensation module, and the fast-loop coupling defect control module are all signal-connected to the collaborative control execution module.
[0019] The active vision imaging module is used to acquire images of the surface of the integrally molded plastic cover plate after screw injection molding and demolding, extracting defect classification features and underlying texture features to provide visual data for molding process drift compensation and defect control. Specifically, the active vision imaging module includes an imaging execution unit, a light source control unit, and a visual feature extraction unit. The imaging execution unit includes a six-axis manipulator and an industrial area array camera. The manipulator can grasp the demolded cover plate and adjust the imaging angle and position in space, working with the camera to complete full-area frontal imaging and local high-precision imaging. The light source control unit can independently control the switching of three sets of strip light sources: 0° coaxial light, 45° diffuse light, and 75° grazing light. The 45° diffuse light corresponds to full-area conventional defect detection, the 75° grazing light corresponds to weak defect enhancement imaging, and the 0° coaxial light corresponds to supplementary lighting for high-reflectivity areas. The visual feature extraction unit classifies the scanned defects such as shrinkage marks, flash, warping, and weld lines, and then provides these defects to the fast loop module. The obtained gray-scale co-occurrence matrix contrast underlying texture features are provided to the slow loop module. The slow-loop process drift compensation module is used to generate trend compensation amounts for screw injection molding process parameters based on the long-term variation trend of underlying texture features, offsetting chronic process fluctuations such as mold temperature drift, screw progressive wear, and gradual changes in raw material batches. Specifically, the slow-loop process drift compensation module includes a trend statistics unit, a drift judgment unit, and a slow-loop compensation generation unit. The trend statistics unit uses a fixed mold window to statistically analyze the variation trend of underlying texture features, employs the least squares method to fit the feature values as a function of each mold, and calculates the feature trend slope. The drift judgment unit compares the feature mean within the statistical window with a preset normal process feature range, and, combined with a trend slope threshold, distinguishes between normal random fluctuations and systematic process drift. The slow-loop compensation generation unit, based on the physical correspondence mechanism between visual features and process parameters, generates small-amplitude, cumulative slow-loop parameter compensation amounts, with each compensation amount not exceeding a certain percentage of the total range of the corresponding parameter. To avoid causing process oscillations; The fast-loop coupling defect control module is used to decouple the root cause process parameters of multiple defects based on defect classification characteristics, generate real-time correction values for screw injection molding process parameters, and quickly respond to sudden appearance defects in a single mold. Specifically, the fast-loop coupling defect control module includes a defect quantification unit, a root cause decoupling unit, and a fast-loop correction generation unit. Among them, the defect quantification unit calculates the comprehensive severity of a single type of defect by combining the defect area ratio and contrast level, and constructs a defect severity vector; the root cause decoupling unit calls the defect and process causal weight matrix specific to the one-piece molded cover plate, calculates the comprehensive influence of each process parameter, sorts them from largest to smallest influence, and determines the root cause parameters that lead to multiple defect coupling; the fast-loop correction generation unit adopts a dual-driven logic of mechanism and data. First, the basic molding process mechanism model outputs the safe upper and lower bounds for parameter correction, and then the case reasoning module matches the historical best parameter tuning value within the safe and feasible domain, and obtains the final fast-loop correction value after amplitude limiting. The collaborative control execution module is used to integrate trend compensation and real-time correction, output the final molding process parameters and send them to the screw injection molding machine for execution, while completing global safety verification and closed-loop feedback processing.
[0020] Example 2 This embodiment provides a machine vision-based adaptive molding method for integral plastic cover plate screw injection molding, applied to an injection molding machine with a screw injection unit, specifically including the following steps: Step S1: Complete the construction and initial parameter calibration of the screw injection molding system, build a vision inspection platform and establish process parameter mapping relationships, and pre-build a basic molding mechanism model and case library.
[0021] Specifically, step S1 includes: Step S11: An industrial area array camera, three independently controllable strip light sources, and a six-axis material handling robot are installed on the discharge side of the screw injection molding machine. The camera, light source, and robot are all connected to the central controller to realize the linkage control of imaging posture, lighting mode, and detection logic. Step S12: Establish an industrial communication connection between the central controller and the screw injection molding machine control system, clarify the register addresses, ranges and adjustment accuracy of the adjustable process parameters such as injection speed, holding pressure, holding pressure switching position, mold temperature and cooling time, and establish a mapping relationship between visual control commands and injection molding machine execution parameters to ensure that control commands can be directly issued and executed. Step S13: Based on the structural characteristics of the target one-piece molded plastic cover and the PVT characteristics of the plastic material used, construct a basic molding process mechanism model, define the safety adjustment boundaries and molding process windows of each process parameter; at the same time, pre-store historical optimal parameter adjustment cases corresponding to typical defect combinations of the cover to form an initial case library, providing a benchmark for control decisions.
[0022] Step S2: After the screw injection molding and demolding, active vision dynamic imaging is performed to extract the defect classification features and underlying texture features of the cover plate. This step uses two imaging methods: a global initial inspection and a local re-inspection. The imaging structure is as follows: Figure 3 As shown, it specifically includes: Step S21: The robotic arm picks up the cover plate after the screw injection molding and demolding and places it at the standard imaging station. A full-area frontal image of the cover plate is acquired using 45° diffused light for initial overall appearance inspection and global texture feature extraction. Step S22: Retrieve the process parameter deviation and defect detection results from the previous cycle, and predict the high-risk defect areas and defect types for the current cycle. For example, if the holding pressure was too low in the previous cycle, it is predicted that micro-shrinkage marks are likely to appear in the reinforcing rib area. Control the robot arm to flip the cover plate to the corresponding observation angle, switch to 75° grazing light to perform local high-precision imaging of the high-risk area, and enhance the visual characteristics of weak defects through the shadow effect of grazing light to improve the detection rate. Step S23: Extract defect classification features from the acquired image to identify defect types and distributions such as shrinkage marks, flash, warping, and weld lines; at the same time, select a fixed sampling window in the cover plate plane area and calculate the gray-level co-occurrence matrix contrast CON as the underlying texture feature for slow loop compensation.
[0023] To eliminate the interference of structural features such as reinforcing ribs, snaps, undercuts, and thin hinges on the texture feature statistics of the one-piece molded plastic cover plate, the ROI selection only applies to the planar area of the cover plate, excluding the areas containing the aforementioned structural features. Within the planar area of the cover plate, four square sampling windows of fixed size are selected, each window being 100×100 pixels. The positions of the four windows are respectively the upper left, upper right, lower left, and lower right regions of the geometric center of the cover plate, with the center of each window 30-50mm away from the geometric center of the cover plate, the specific distance being calibrated according to the actual size of the cover plate. For cover plates of the same product model, the positions of the four windows are fixed during the system calibration stage and remain unchanged during subsequent production. For each module, the gray-level co-occurrence matrix contrast value of the four windows is calculated, and the arithmetic mean of the four windows is taken as the final CON feature value of that module.
[0024] The formula for calculating the contrast ratio CON of the gray-level co-occurrence matrix is: ; in, The grayscale co-occurrence matrix contrast is used to quantify the micro-texture roughness of the surface of a one-piece molded plastic cover. The larger the value, the rougher the surface texture. The gray level of the image is fixed at 256 in this scheme, which corresponds to the total number of gray levels of an 8-bit grayscale image. The grayscale indexes of the two pixels are given, with a value range of 1. ; For a specified distance and direction, the grayscale value is... The joint probability of a pixel with grayscale value j appearing in pairs. The co-occurrence pixel spacing is uniformly taken for the cover plate planar region. Statistics This ensures consistency in feature statistics and eliminates the interference of directional differences on feature values. Because the surface micro-roughness of the one-piece molded plastic cover plate is directly related to the cooling rate of the melt, when the mold temperature drops slowly, the cooling rate of the melt surface accelerates, the surface roughness of the product increases, and the corresponding gray-scale co-occurrence matrix contrast index increases monotonically. This feature can indirectly reflect the slow process fluctuations of mold temperature drift and screw wear without identifying specific defects, providing a quantifiable visual basis for slow loop compensation. Step S3: Based on the underlying texture features, perform slow-loop process drift trend compensation to generate screw injection molding process parameter trend compensation amount.
[0025] Step S3 also includes: Step S31: Using N consecutive modulo operations as the statistical window, N=20 by default in this scenario, the contrast feature value corresponding to the i-th modulo operation is... The least squares method is used to fit the trend of eigenvalues with modulus, and the slope of the trend is calculated. Positive values indicate a continuous increase in surface roughness, corresponding to a drift trend such as a slow decrease in mold temperature; negative values indicate a continuous decrease in surface roughness. Step S32: The preset normal fluctuation range of the feature is... The slope threshold is When simultaneously satisfying the conditions that the mean of the features within the window exceeds the normal range, and the absolute value of the trend slope is... Two conditions are considered when a systematic process drift is determined to exist. The slope threshold is one of these conditions. use Quantile calibration: During the process stabilization period, CON eigenvalues are continuously collected for 200 cycles. The slope of each 20-cycle window is calculated, and the absolute value of the slope of these 200 cycles is taken. quantiles as The typical value range is 0.02-0.08 eigenvalues / modulus; Step S33: Generate the slow-loop compensation amount based on the linear mapping relationship between visual features and process parameters. Taking mold temperature compensation as an example, the single-cycle slow-loop temperature compensation amount... The calculation formula is: ; in, The conversion coefficient from contrast characteristic slope to mold temperature compensation is expressed in °C·mold number / characteristic value, and is determined experimentally based on the PVT characteristics of the cover plate material and the thermal conductivity of the mold. The calibration test procedure is as follows: Under different mold temperature settings, 50 mold runs are produced at 50℃, 55℃, 60℃, 65℃, and 70℃ respectively, and the contrast ratio (CON) value of the gray-level co-occurrence matrix is recorded for each run; the trend slope at each temperature is calculated, and a linear relationship between the trend slope and the mold temperature deviation is established; the result is obtained through least squares fitting. The numerical value was obtained; a verification experiment was conducted at the intermediate temperature point to confirm the fitting accuracy. The slow loop compensation adopts a small step size and cumulative correction logic. The single slow loop compensation amplitude does not exceed 5% of the total range of the corresponding parameter. The compensation amount is gradually corrected in an cumulative manner to correct the basic molding process parameters. It corresponds to slowly changing process factors such as mold scaling and screw wear. By inferring the process drift direction and amplitude through the characteristic trend of continuous multi-mold cycles, it achieves online compensation for process drift without additional sensors and avoids process oscillation. Step S4: Based on defect classification features, perform fast-loop coupling defect decoupling control to generate real-time correction values for screw injection molding process parameters. The control logic is as follows: Figure 4 As shown: Step S4 further includes: Step S41: For each type of identified defect, calculate the comprehensive severity of the single-type defect and construct a defect severity vector. For appearance defects of the integrally molded cover plate, such as shrinkage marks, flash, warping, and weld lines, calculate the comprehensive severity of the single-type defect and construct a defect severity vector. Class defects are defined to quantify their severity. The calculation formula is: ; in, For the first The area ratio of the defect type, with a value range of [0,1], is obtained by dividing the number of pixels in the defect area by the total number of pixels in the corresponding detection area, reflecting the distribution range of the defect; For the first The contrast level of the defect type, with a value range of [0,1], is calculated by normalizing the grayscale difference between the defect area and the normal area, reflecting the degree of obviousness of the defect. The value ranges from [0,1], with larger values indicating more severe defects.
[0026] Step S42: Call the specific defect and process causal weight matrix for the one-piece molded cover to decouple the root cause parameters of multiple defects. Construct a specific defect and process causal weight matrix tailored to the molding characteristics of the one-piece molded plastic cover. Matrix elements Indicates the first Class defects and the first The causal correlation strength of the process parameters, with a value range of [value range missing]. The higher the value, the higher the correlation. This solution has built-in exclusive weighting rules for the cover plate: the correlation weight between shrinkage marks and holding pressure, and the holding pressure switching position is the highest; the correlation weight between flash and clamping force, and the holding pressure switching position is the highest; when shrinkage marks and flash occur simultaneously, the overall weight of the holding pressure switching position is higher than that of the other parameters. The severity of all current defects is weighted and combined to calculate the overall impact of each process parameter. The calculation formula is: ; in, This represents the total number of defect types identified in the current module, for a single type of defect. When multiple defects are coupled All process parameters are set according to The parameters with the highest overall impact, ranked from largest to smallest, are the root causes of the current multi-defect coupling and should be adjusted first. When shrinkage marks and flash coexist in the snap-fit area, it is first determined that the pressure holding transition point is too late, and the pressure holding switching position and pressure holding pressure are adjusted first. When warping and weld lines coexist in the reinforcing rib area, it is first determined that the melt front temperature is insufficient, and the injection temperature and injection speed are adjusted first.
[0027] Step S43: Calculate the fast-loop real-time correction amount using a dual-driven logic approach based on both mechanism and data. The first step involves the basic molding process mechanism model outputting a safety upper bound for parameter correction based on the root cause parameter types obtained through reasoning. and the lower world The adjustment range is limited to not exceeding the material forming process window and the equipment safety range, thus constituting a safe and feasible domain.
[0028] The second step is to retrieve the most similar vector to the current defect vector from the historical case database. Based on historical cases, this solution is the default. Case similarity The cosine similarity of the defect feature vectors is used for calculation: ; in, This represents the defect severity vector under the current operating conditions. For the first A vector of the severity of defects in each historical case; The dot product of two vectors. This represents the magnitude of the corresponding vector. The similarity value range is... The closer the value is to 1, the more similar the defect characteristics of the two working conditions are. Using similarity as the weight, the parameter tuning values of historical cases are weighted and summed to obtain the initial correction amount: ; in, For the first In the first historical case The actual optimal correction amount of the process parameters The third step is to substitute the initial correction amount into the mechanistic safety and feasibility region for amplitude limiting to obtain the final fast-loop correction amount. : ; in This is a limiting function. If the input value is within the upper or lower limit range, it outputs directly; if it exceeds the upper limit, it outputs the upper limit value; and if it is below the lower limit, it outputs the lower limit value. The mechanistic model is responsible for a safety net, preventing the data-driven model from outputting dangerous parameters that exceed the process window; the case-based reasoning is responsible for outputting refined optimal values within the safe range, balancing the safety and control precision of the industrial site. Step S5: Integrate the slow-loop trend compensation amount and the fast-loop real-time correction amount to generate the final molding process parameters and issue them for execution.
[0029] Step S51: Combine the basic molding process parameters, the fast-loop instantaneous correction amount, and the slow-loop trend compensation amount, and finally apply them to the next mold. Process parameters The calculation formula is: ; in, The first one adopted for current production The basic process parameter values, i.e. the reference parameters used in the previous module, are cumulatively corrected by slow loop compensation; The first output of the fast loop The parameter is adjusted in real time, representing a single adjustment value for the current module defect; The first output of the slow loop The parameter trend compensation amount is the cumulative correction value for long-term process drift. It is not updated for every module, but only when the drift determination is triggered.
[0030] The final parameters after synthesis need to undergo a full-parameter process window global security check again to meet the following requirements: ; in, , For the first The total allowable minimum and maximum values for each process parameter are jointly determined by the equipment's range and the material molding process window. If the range is exceeded, the parameters are automatically limited to the boundary values to ensure production safety. After successful verification, the central controller sends the final parameter instructions to the screw-type injection molding machine control system for application in the next injection molding cycle. Step S6: Collect visual data after the next demolding cycle for closed-loop feedback verification, and repeatedly execute the molding control process.
[0031] Step S61: Collect visual data of the next module cover plate and calculate the control effect coefficient. To assess the effectiveness of a single intervention: ; in, For the first time before the adjustment (previous module) The severity of the defects is used as a benchmark for effectiveness evaluation; For the adjusted (current module) number The severity of the defect is used as a comparison value; The total number of defect types participating in the assessment is consistent with the number of defect types identified before regulation.
[0032] The range is Positive values indicate that the defects are reduced after adjustment, and the larger the value, the better the effect; negative values indicate that the defects are aggravated after adjustment. A safe rollback buffer is set: when... When, only trigger and record the warning, without immediately reverting; when When this occurs, a safety rollback mechanism is triggered, restoring the process parameters to the previous stable state. This condition and its results are then stored in the exception case library for subsequent model iteration and optimization. Frequent accidental rollbacks within the buffer zone are prevented from impacting production efficiency. Step S62: Set an action dead zone to avoid frequent parameter adjustments caused by minor quality fluctuations. Let the severity of the overall defect in the previous control cycle be... The current cycle is The preset dead zone threshold is When both conditions are met and Under two conditions, namely, when the defect deviation is less than the dead zone threshold and there is no obvious process drift trend, the current molding process parameters remain unchanged, no new control commands are generated, the frequency of action of the injection molding machine actuator is reduced, and the service life of the equipment is extended.
[0033] Among them, the defect dead zone threshold use Principle calibration: Take 50 batches of qualified products that have been manually confirmed to have no visible defects, and calculate the average of their overall defect severity. with standard deviation ,Pick The value ranges from 0.05 to 0.15.
[0034] After completing the adjustment of a single control cycle, the system automatically returns to step S2 and enters the next control cycle, continuing closed-loop operation.
Claims
1. A machine vision-based adaptive system for screw injection molding of one-piece plastic cover plates, characterized in that: The system is applied to an injection molding machine with a screw injection unit. The system includes an active vision imaging module, a slow-loop process drift compensation module, a fast-loop coupling defect control module, and a collaborative control execution module. The active vision imaging module, the slow-loop process drift compensation module, and the fast-loop coupling defect control module are all signal-connected to the collaborative control execution module. The active vision imaging module is used to acquire images of the surface of the integrally molded plastic cover plate after the screw injection molding demolding, extract defect classification features and underlying texture features, and provide visual data for molding process drift compensation and defect control. The slow-loop process drift compensation module is used to generate a screw injection molding process parameter trend compensation amount based on the long-term change trend of the underlying texture features, so as to offset the chronic process fluctuations caused by mold temperature drift and screw progressive wear. The fast-loop coupling defect control module is used to decouple the root cause process parameters of multiple defects based on defect classification characteristics, and generate real-time correction values for screw injection molding process parameters; the collaborative control execution module is used to integrate trend compensation values and real-time correction values, output the final molding process parameters and send them to the screw injection molding machine for execution.
2. The machine vision-based adaptive system for integral injection molding of plastic cover plates using screws, as described in claim 1, is characterized in that: The active vision imaging module includes an imaging execution unit, a light source control unit, and a visual feature extraction unit; wherein: The imaging execution unit is used to grasp the demolded cover plate with a six-axis robotic arm and adjust the imaging posture, and work with an industrial area scan camera to complete the acquisition of global and local images. The light source control unit is used to switch between three sets of light sources: 0° coaxial light, 45° diffuse light, and 75° grazing light, to match the lighting requirements of different detection scenarios. The visual feature extraction unit is used to extract defect classification features and underlying texture features from the acquired image. The underlying texture features are quantized using gray-level co-occurrence matrix contrast.
3. The machine vision-based adaptive system for integral injection molding of plastic cover plates using screws, as described in claim 1, is characterized in that: The slow-loop process drift compensation module includes a trend statistics unit, a drift determination unit, and a slow-loop compensation generation unit; wherein: The trend statistics unit is used to statistically analyze the changing trend of the underlying texture features with a fixed module window and calculate the feature trend slope. The drift determination unit is used to compare the feature mean with the normal range and combine it with the trend slope threshold to determine whether there is a systematic process drift. The slow loop compensation generation unit is used to generate a small-amplitude cumulative slow loop parameter compensation amount based on the mapping relationship between visual features and process parameters.
4. The machine vision-based adaptive system for integral injection molding of plastic cover plates using screws, as described in claim 1, is characterized in that: The fast-loop coupling defect control module includes a defect quantification unit, a root cause decoupling unit, and a fast-loop correction generation unit; wherein: The defect quantification unit is used to calculate the overall severity of various appearance defects and generate a defect severity vector. The root cause decoupling unit is used to calculate the comprehensive influence of each process parameter based on the defect and process causal weight matrix specific to the integrally molded cover plate, and to sort and determine the root cause parameters; the fast-loop correction generation unit is used to adopt mechanism and data dual-driven logic, first delineate the safe and feasible domain by the mechanism model, and then output the correction amount and limit the amplitude by case reasoning to obtain the final fast-loop correction value.
5. A machine vision-based adaptive injection molding method for integrally molded plastic cover screws, characterized in that: The method of the machine vision-based screw injection molding integrated plastic cover adaptive system according to any one of claims 1-4 includes the following steps: Step S1: Complete the construction and initial parameter calibration of the screw injection molding system, build a vision inspection platform and establish process parameter mapping relationships, and pre-build a basic molding mechanism model and case library; Step S2: After the screw injection molding demolding, active vision dynamic imaging is performed to extract the defect classification features and underlying texture features of the cover plate; Step S3: Based on the underlying texture features, perform slow-loop process drift trend compensation to generate screw injection molding process parameter trend compensation amount; Step S4: Based on defect classification features, perform fast-loop coupling defect decoupling control to generate real-time correction values for screw injection molding process parameters; Step S5: Integrate the slow-loop trend compensation amount and the fast-loop real-time correction amount to generate the final molding process parameters and send them to the screw injection molding machine for application in the next injection molding cycle; Step S6: Collect visual data after the next demolding cycle for closed-loop feedback verification, and repeatedly execute the molding control process.
6. The machine vision-based adaptive injection molding method for integrally molded plastic cover plate screws according to claim 5, characterized in that: Step S1 further includes: Step S11: Install an industrial area scan camera, multiple independently controlled strip light sources, and a six-axis material handling robot on the discharge side of the screw injection molding machine, and connect them to the central controller to form a vision inspection platform; Step S12: Establish an industrial communication connection between the central controller and the screw injection molding machine control system, and map the register addresses, ranges and adjustment accuracy of adjustable process parameters such as injection speed, holding pressure, holding pressure switching position, mold temperature, and cooling time. Step S13: Combine the structural features of the cover plate with the PVT characteristics of the material to construct a basic molding process mechanism model, define the safe adjustment boundary of the screw injection parameters and the molding process window, pre-store the historical optimal parameter adjustment cases corresponding to typical defect combinations, and form an initial case library.
7. The machine vision-based adaptive injection molding method for integrally molded plastic cover plate screws according to claim 5, characterized in that: Step S2 further includes: Step S21: The robotic arm picks up the cover plate after the screw injection molding and demolding and moves it to the standard imaging station. The cover plate is captured with 45° diffuse light to obtain a full-area frontal image and complete the initial overall appearance inspection. Step S22: Retrieve the process parameter deviation and defect detection results of the previous mold, predict the high-risk defect area and defect type of the current mold, control the robot to flip the cover plate and switch the 75° grazing light to perform local high-precision imaging of the high-risk area. Step S23: Extract defect classification features from the image for fast loop control, and simultaneously select a fixed sampling window in the cover plate plane region to calculate the gray-level co-occurrence matrix contrast CON as the underlying texture feature for slow loop compensation. The calculation formula is as follows: ; in, The number of gray levels in the image. The grayscale index of the two pixels. This is the joint probability element of the gray-level co-occurrence matrix under a specified pixel interval and statistical direction.
8. The machine vision-based adaptive injection molding method for integrally molded plastic cover plate screws according to claim 5, characterized in that: Step S3 further includes: Step S31: Using N consecutive moduli as a statistical window, fit the changing trend of the gray-level co-occurrence matrix contrast within the window using the least squares method, and calculate the characteristic trend slope. ; Step S32: When the mean value of the features within the window exceeds the preset normal range and the absolute value of the trend slope is greater than the slope threshold, it is determined that there is a systematic process drift. Step S33: Generate the slow-loop compensation amount based on the linear mapping relationship between the characteristic slope and the process parameters. Taking mold temperature compensation as an example, the single-cycle slow-loop temperature compensation amount is... The calculation formula is: ; in, This is the conversion coefficient from contrast feature slope to mold temperature compensation; the amplitude of a single slow loop compensation does not exceed the total range of the corresponding parameter. The basic forming process parameters are corrected in an cumulative manner.
9. The machine vision-based adaptive injection molding method for integrally molded plastic cover plate screws according to claim 5, characterized in that: Step S4 further includes: Step S41: For the identified defects, combine the defect area ratio... With contrast level Calculate the first Overall severity of class defects Construct a defect severity vector, the calculation formula of which is: ; Step S42: Call the defect and process causal weight matrix specific to the one-piece molded cover plate to calculate the comprehensive influence of each process parameter. The root cause parameters are determined by sorting them from largest to smallest impact, and the formula for calculating the overall impact is as follows: ; in, This represents the total number of defect types currently identified. For the first Class defects and the first Causal correlation weights for process parameters; Step S43: The corrected safety upper and lower bounds of the root cause parameters are output from the basic molding process mechanism model. Within the safe and feasible region, the historical optimal parameter tuning values are retrieved through case similarity matching. After weighted calculation and amplitude limiting, the fast loop real-time correction amount is obtained.
10. The machine vision-based adaptive injection molding method for integrally molded plastic cover screws according to claim 5, characterized in that: Steps S5 and S6 further include: Step S51: Set the basic molding process parameters Real-time correction amount of fast loop Compensation amount for slow-cycle trend By superimposing and synthesizing, the final molding process parameters are obtained. The calculation formula is as follows: ; After passing the safety verification of the full-parameter molding process window, the data is sent to the screw-type injection molding machine for use in the next injection molding cycle. Step S61: Collect visual data of the cover plate after the next demolding, calculate the control effect coefficient to evaluate the effectiveness of parameter adjustment, and if the defects worsen, trigger a safety rollback to restore the previous stable molding process parameters. Step S62: If the current mold defect deviation is less than the dead zone threshold and there is no obvious process drift trend, then keep the current molding process parameters unchanged and enter the next cycle of detection.
Citation Information
Patent Citations
Machine Vision-Based Method and System for Surface Defect Analysis of Injection Molded Parts
CN114565614B
