Extruder and system based on visual detection

Abstract

The present application relates to the field of machine vision and industrial automation control technology, and more particularly to an extruder and system based on visual inspection, comprising a physical device and a control device, the physical device comprising an image acquisition unit, a protection unit, an illumination unit and a control execution unit. The image acquisition unit uses a main camera and an auxiliary camera to synchronously acquire images of the discharge port area from the top view and side view angles respectively. The control device comprises an image acquisition module, a visual feature extraction module, a discharge state analysis module, a multi-level response decision module and a control instruction issuing module, and through multi-angle image sequence acquisition, moving material area segmentation, multi-dimensional visual feature extraction, dynamic threshold model matching and hierarchical response decision, real-time detection and accurate control of the hidden abnormalities in the discharging process are realized. The present application can effectively identify abnormalities such as flow rate mutation and morphology variation, and avoid product quality defects and unplanned shutdowns.

Description

Technical Field

[0001] This invention relates to the fields of machine vision and industrial automation control technology, and in particular to an extruder and system based on vision inspection. Background Technology

[0002] The extruder feeding monitoring system uses sensors, such as temperature sensors and pressure sensors, to collect thermodynamic and mechanical parameters during the extrusion process in real time. Based on the data fed back by the sensors, the monitoring unit analyzes the correlation between temperature gradient and pressure fluctuation, infers the continuity and consistency of material flow, and then adjusts the screw speed or heating power through closed-loop control to maintain the stability and predictability of the feeding process, thereby supporting the automated operation of the production process.

[0003] Existing sensor-based extruder feeding monitoring technologies suffer from the following technical challenges: Current methods primarily rely on point or line measurement techniques such as proximity switches or flow meters, which can only acquire limited location or overall flow signals, failing to provide comprehensive, visualized morphological analysis of the feeding process. During continuous extrusion of polymer melts or granules, material flow rates may change abruptly due to formulation variations, uneven heating, or screw wear, leading to material agglomeration, bridging, or other morphological variations. These latent anomalies occur inside the bushing or in the outlet area, and existing sensors struggle to directly capture their two-dimensional spatial distribution and dynamic evolution details. For example, when localized adhesion occurs on the inner wall of the bushing, causing asymmetrical material flow, the overall feeding rate may temporarily remain normal, but the material has actually deviated from the ideal flow trajectory, creating a potential for dimensional fluctuations or inherent defects in subsequent products. Furthermore, such potential faults cannot be detected in time, ultimately leading to unplanned shutdowns. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an extruder and system based on vision detection, which solves the technical problem of product quality defects and unplanned downtime caused by the difficulty of capturing latent anomalies such as sudden changes in flow rate and morphological variations of polymer materials during the extruder bushing feeding process in real time by existing sensors.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the present invention provides a vision-based extruder, comprising a physical device and a control device, wherein the control device establishes a communication connection with the physical device; The physical device includes an image acquisition unit, a protection unit, a lighting unit, and a control execution unit; The image acquisition unit includes a main camera and an auxiliary camera. The main camera is mounted at a downward angle directly above the extruder bushing discharge port to acquire global discharge trajectory images. The auxiliary camera is mounted horizontally to the side of the bushing discharge port to acquire local images of the material being discharged along the inner wall. The protective unit uses a high-temperature resistant, dustproof, and sealed structure to encapsulate the main camera and the auxiliary camera. The lighting unit provides uniform fill light for the main camera and the auxiliary camera; The control execution unit includes an audible and visual alarm, a forced feeding mechanism drive interface, and an extruder main control system emergency stop interface; The control device includes: The image acquisition module is used to synchronously trigger the main camera and the auxiliary camera to acquire raw image data streams, perform non-uniformity correction and dynamic range compression on the raw image data streams, and output the processed image sequence; The visual feature extraction module is used to receive the image sequence output by the image acquisition module, segment the moving material region from the image sequence, and extract the material feeding amount feature, material feeding speed feature and material morphology feature of the moving material region to generate a feature vector; The material feeding status analysis module is used to receive the feature vector generated by the visual feature extraction module, match the feature vector with a dynamic threshold model, and output the status judgment result. A multi-level response decision module is used to receive the status judgment result output by the material feeding status analysis module, query a response strategy table based on the status judgment result, and generate control instructions. The control command issuing module is used to receive the control commands generated by the multi-level response decision module, convert the control commands into drive signals, and send the drive signals to the control execution unit.

[0006] Furthermore, in the visual detection-based extruder of the present invention, the image acquisition module synchronously triggers the main camera and the auxiliary camera, including sending synchronization pulse signals to the main camera and the auxiliary camera; The image acquisition module performs non-uniformity correction on the acquired raw image data stream, including calling a pre-stored calibration parameter matrix to perform gain and bias compensation on the pixels of the raw image. The dynamic range compression includes converting the high bit depth raw image data to an 8 bit depth.

[0007] Furthermore, in the visual detection-based extruder of the present invention, the visual feature extraction module segments the moving material region from the image sequence, including using a Gaussian mixture model to establish a pixel color probability distribution model of the background image sequence, and matching the pixel values ​​of the new input image with the background model to segment the moving material region. Extracting the material feeding amount feature includes counting the total number of foreground pixels within the moving material area and multiplying it by the pixel actual area conversion factor; Extracting the feeding speed features includes calculating the median flow velocity of the optical flow field in the moving material region between consecutive frames; Extracting the material morphological features includes calculating the roundness of the outline of the moving material region, the tortuosity of the boundary, and the contrast of the gray-level co-occurrence matrix within the region.

[0008] Furthermore, in the visual detection-based extruder of the present invention, the feeding state analysis module matches the feature vector with a dynamic threshold model, wherein the dynamic threshold model defines a range of positive proportional relationship between the feeding quantity feature and the feeding speed feature. The matching includes checking whether the decrease in the material feeding rate characteristic value is within the range allowed by the positive proportional relationship when the material feeding speed characteristic value decreases.

[0009] Furthermore, in the visual detection-based extruder of the present invention, the multi-level response decision module queries the response strategy table based on the state determination result, including: When the status determination result is a minor abnormality and the duration is less than the set threshold, a level one warning instruction is generated. When the status determination result is a minor abnormality and the duration exceeds the set threshold, or when the result is a serious abnormality, a secondary adjustment instruction including the material supply compensation amount is generated. When the status determination result indicates that the material flow is completely interrupted, a shutdown command is generated.

[0010] Furthermore, in the vision-based extruder of the present invention, the control command issuing module converts control commands into drive signals, including: The first-level warning command is converted into a pulse width modulation signal and sent to the audible and visual alarm. The secondary adjustment command is converted into an analog voltage signal or a Modbus TCP message and sent to the forced feeding mechanism drive interface; The shutdown command is converted into a switching signal that triggers the closing of the relay dry contacts and sent to the emergency stop interface.

[0011] Furthermore, in the visual inspection-based extruder of the present invention, the main camera is mounted at a 45° downward angle with its optical axis aligned with the center of the feed inlet; the auxiliary camera lens is focused on the edge area of ​​the bushing outlet.

[0012] Furthermore, in the visual inspection-based extruder of the present invention, the dynamic threshold model can be updated online according to changes in the production formula; the graded response strategy table allows users to configure and adjust it according to process requirements.

[0013] Furthermore, in the visual inspection-based extruder of the present invention, the dynamic threshold model is updated online according to changes in the production formula, including: Receive formula switching signals, and after formula switching, acquire image sequences of normal production operation under the new formula; Based on the newly acquired image sequence, the corresponding feature vector is extracted through the visual feature extraction module; The normal fluctuation range of each feature parameter and the correlation rules between feature parameters are recalculated using the newly extracted feature vector set to update the dynamic threshold model.

[0014] Secondly, the vision-based extruder system provided by the present invention is applied to the vision-based extruder as described above, comprising: The image acquisition unit includes a main camera mounted at a downward angle directly above the extruder bushing discharge port and an auxiliary camera mounted horizontally to the side of the bushing discharge port. The control execution unit includes an audible and visual alarm, a forced feeding mechanism drive interface, and an extruder main control system emergency stop interface. A control device, communicatively connected to the image acquisition unit and the control execution unit, comprising: The image acquisition module is used to synchronously trigger the main camera and the auxiliary camera to acquire raw image data streams, perform non-uniformity correction and dynamic range compression on the raw image data streams, and output the processed image sequence; The visual feature extraction module is used to receive the image sequence output by the image acquisition module, segment the moving material region from the image sequence, and extract the material feeding amount feature, material feeding speed feature and material morphology feature of the moving material region to generate a feature vector; The material feeding status analysis module is used to receive the feature vector generated by the visual feature extraction module, match the feature vector with a dynamic threshold model, and output the status judgment result. A multi-level response decision module is used to receive the status judgment result output by the material feeding status analysis module, query a response strategy table based on the status judgment result, and generate control instructions. The control command issuing module is used to receive the control commands generated by the multi-level response decision module, convert the control commands into drive signals, and send the drive signals to the control execution unit.

[0015] Beneficial effects of this invention; This invention overcomes the limitations of existing point- or line-based measurement methods by acquiring images from both a main camera and an auxiliary camera, enabling comprehensive and visualized morphological analysis of the material feeding process. The visual feature extraction module converts image information into multi-dimensional quantitative data through parallel computation of material feeding quantity, speed, and morphology features, providing a data foundation for identifying latent anomalies. The material feeding status analysis module employs a dynamic threshold model for multi-feature collaborative analysis, capable of identifying potential faults such as flow asymmetry that cannot be reflected by a single feature exceeding limits. The multi-level response decision module triggers graded responses based on the anomaly level and duration, achieving progressive and precise control from early warning and adjustment to shutdown. Through the collaborative operation of the image acquisition module, visual feature extraction module, material feeding status analysis module, multi-level response decision module, and control command issuance module, this system forms a closed-loop management system from visual perception to execution control, effectively solving product quality defects and unplanned downtime caused by latent anomalies such as sudden changes in flow rate and morphological variations. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the extruder system based on vision inspection provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] Please see Figure 1 In a first aspect, the visual inspection-based extruder provided by the present invention includes a physical device and a control device, wherein the control device establishes a communication connection with the physical device; The physical device includes an image acquisition unit, a protection unit, a lighting unit, and a control execution unit; The image acquisition unit includes a main camera and an auxiliary camera. The main camera is mounted at a downward angle directly above the extruder bushing discharge port to acquire global discharge trajectory images. The auxiliary camera is mounted horizontally to the side of the bushing discharge port to acquire local images of the material being discharged along the inner wall. The protective unit uses a high-temperature resistant, dustproof, and sealed structure to encapsulate the main camera and the auxiliary camera. The lighting unit provides uniform fill light for the main camera and the auxiliary camera; The control execution unit includes an audible and visual alarm, a forced feeding mechanism drive interface, and an extruder main control system emergency stop interface; The control device includes: The image acquisition module is used to synchronously trigger the main camera and the auxiliary camera to acquire raw image data streams, perform non-uniformity correction and dynamic range compression on the raw image data streams, and output the processed image sequence; The visual feature extraction module is used to receive the image sequence output by the image acquisition module, segment the moving material region from the image sequence, and extract the material feeding amount feature, material feeding speed feature and material morphology feature of the moving material region to generate a feature vector; The material feeding status analysis module is used to receive the feature vector generated by the visual feature extraction module, match the feature vector with a dynamic threshold model, and output the status judgment result. A multi-level response decision module is used to receive the status judgment result output by the material feeding status analysis module, query a response strategy table based on the status judgment result, and generate control instructions. The control command issuing module is used to receive the control commands generated by the multi-level response decision module, convert the control commands into drive signals, and send the drive signals to the control execution unit.

[0020] The image acquisition module drives the main and auxiliary cameras to acquire images synchronously. The module sends synchronization trigger signals to both cameras, ensuring strict alignment of their exposure cycle start points. The module receives raw image data streams from both cameras via a high-speed data interface. Subsequently, the module calls a pre-stored calibration parameter matrix to perform gain and bias compensation on each pixel of the raw image to eliminate lens vignetting and sensor pixel response differences, completing non-uniformity correction. The corrected image data undergoes dynamic range compression, mapping the high bit depth of the raw data to a standard bit depth while preserving key brightness and darkness details of material edges and surface textures. After this processing, the module outputs a time-aligned and photometrically normalized dual-view image sequence.

[0021] The visual feature extraction module receives a dual-view image sequence output by the image acquisition module. First, the module uses a Gaussian mixture model to learn from the historical background image sequence, establishing a color probability distribution model for each pixel as a background reference. For each new input image frame, the module calculates the matching probability of each pixel value with the background model, identifying pixels with probabilities below a set threshold as foreground moving pixels, thus accurately segmenting the moving material region. Next, the module performs parallel analysis on the segmented moving regions. By counting the total number of foreground pixels within the moving region and multiplying it by the pixel actual area conversion factor obtained from camera calibration, the material feeding cross-sectional area is calculated as the feeding quantity feature. By calculating the optical flow field of the moving region between consecutive frames, the median velocity of the optical flow vector is taken and multiplied by the inter-frame time interval to obtain the average flow velocity of the material, which is used as the feeding speed feature. By analyzing the roundness of the moving region's contour, boundary tortuosity, and the contrast of the gray-level co-occurrence matrix within the region, the morphological characteristics of the material are quantified as the material morphology feature. Finally, the module combines the feeding quantity feature, feeding speed feature, and material morphology feature to generate a multi-dimensional feature vector.

[0022] The material feeding status analysis module receives multi-dimensional feature vectors generated by the visual feature extraction module. The module incorporates a dynamic threshold model, trained on historical normal production data. This model defines the normal fluctuation range of each feature parameter and the association rules between them, such as the expected positive proportionality between the material feeding quantity and material feeding speed features. The module matches the real-time acquired feature vectors with the dynamic threshold model. The matching process involves two levels: first, a threshold comparison of a single feature is performed to check if each feature value exceeds its independent normal fluctuation range; second, a multi-feature association rule compliance check is performed to analyze whether the coordinated changes between features violate predefined association rules. For example, when the material feeding speed feature value decreases, the module checks whether the decrease in the material feeding quantity feature value is within the allowed range of the positive proportionality relationship. Based on the combined comparison and check results, the module outputs a comprehensive status judgment result that integrates the anomaly level and type identifier.

[0023] The multi-level response decision module receives the comprehensive status judgment result output by the material feeding status analysis module. The module has a pre-stored hierarchical response strategy table, which defines the response actions corresponding to different anomaly levels, types, and durations. The module parses the anomaly level and duration information included in the status judgment result and queries the hierarchical response strategy table for mapping. When the status judgment result is a minor anomaly and the duration is less than a set threshold, the module generates a level-one warning instruction. When the status judgment result is a minor anomaly but the duration exceeds the set threshold, or when the result is a severe anomaly, the module generates a level-two adjustment instruction, which includes a specific material supply compensation amount calculated based on the deviation between the material feeding quantity characteristics and the material feeding speed characteristics. When the status judgment result indicates a complete interruption of material flow or an immediate risk of equipment failure, the module generates a highest-level shutdown instruction.

[0024] The control command issuing module receives control commands generated by the multi-level response decision module. The module converts abstract control commands into drive signals that can be recognized by specific execution units. For first-level warning commands, the module generates a pulse-width modulation signal of a specific frequency through a protocol converter and drives the audible and visual alarm to issue a warning. For second-level adjustment commands, the module converts the material compensation amount into an analog voltage signal or a message conforming to the Modbus TCP protocol and sends it to the driver of the forced feeding mechanism through the corresponding interface. For the highest-level shutdown command, the module triggers the closure of a relay dry contact, sending a passive switch signal to the emergency stop interface of the extruder main control system, triggering the emergency stop circuit. The module ensures that control commands are accurately and reliably delivered to the control execution units such as the audible and visual alarm, the forced feeding mechanism drive interface, or the emergency stop interface.

[0025] The image acquisition module of this invention sends synchronization pulse signals to the main camera and the auxiliary camera to achieve exposure cycle alignment. It eliminates lens vignetting and sensor pixel response differences by calling the pre-stored calibration parameter matrix to perform gain and offset compensation on the original image pixels. It also converts the high bit depth original image data to 8 bit depth to preserve the light and dark details of material edges and surface textures.

[0026] The visual feature extraction module of this invention uses a Gaussian mixture model to establish a pixel color probability distribution model of the background image sequence. By matching the pixel values ​​of the new input image with the background model, the moving material region is segmented. The total number of foreground pixels in the moving material region is counted and multiplied by the pixel actual area conversion factor to obtain the material feeding feature. The median flow velocity of the optical flow field in the moving material region between consecutive frames is calculated to obtain the material feeding speed feature. The circularity of the outline of the moving material region, the tortuosity of the boundary, and the contrast of the gray-level co-occurrence matrix inside the region are analyzed to obtain the material morphology feature.

[0027] The material feeding status analysis module of this invention matches feature vectors with a dynamic threshold model. The dynamic threshold model defines a range of positive proportional relationships between material feeding quantity features and material feeding speed features. The matching process includes checking whether the decrease in the material feeding quantity feature value is within the range allowed by the positive proportional relationship when the material feeding speed feature value decreases, so as to analyze the multi-feature collaborative change pattern.

[0028] The multi-level response decision module of this invention queries the response strategy table based on the status judgment result. When the status judgment result is slightly abnormal and the duration is less than the set threshold, a first-level warning instruction is generated. When the status judgment result is slightly abnormal and the duration exceeds the set threshold or the result is seriously abnormal, a second-level adjustment instruction including the material supply compensation amount is generated. When the status judgment result indicates that the material flow is completely interrupted, a shutdown instruction is generated.

[0029] The control command issuing module of this invention converts the first-level early warning command into a pulse width modulation signal and sends it to the audible and visual alarm; converts the second-level adjustment command into an analog voltage signal or a Modbus TCP message and sends it to the forced feeding mechanism drive interface; and converts the stop command into a switching signal that triggers the closing of the relay dry contact and sends it to the emergency stop interface.

[0030] The main camera of this invention is installed at a 45-degree downward angle with its optical axis aligned with the center of the discharge port to capture global discharge trajectory images. The auxiliary camera lens focuses on the edge area of ​​the bushing outlet to capture local images of the material being discharged along the inner wall.

[0031] The dynamic threshold model of this invention can be updated online according to changes in production formula, and the hierarchical response strategy table allows users to configure and adjust it according to process requirements to adapt to different production conditions.

[0032] The online update process of the dynamic threshold model of the present invention includes receiving a formula switching signal and collecting image sequences of normal production operation under the new formula; extracting corresponding feature vectors through a visual feature extraction module based on the newly collected image sequences; and recalculating the normal fluctuation range of each feature parameter and the correlation rules between feature parameters using the newly extracted feature vector set to update the dynamic threshold model.

[0033] This invention addresses the challenge of detecting hidden anomalies during the feeding process of polymer materials by constructing a control system that integrates multi-angle visual perception and intelligent decision-making. The system employs a main camera and an auxiliary camera to simultaneously acquire image sequences of the feeding area from top-down and side-view angles, respectively. Optical distortion and environmental interference are eliminated through non-uniformity correction and dynamic range compression, resulting in clear and stable dual-view image data. The visual feature extraction module uses background modeling technology to segment the moving material area and calculates three dimensions of parameters in parallel: feeding quantity, feeding speed, and material morphology. The feeding quantity feature obtains cross-sectional area data through pixel statistics and calibration conversion; the feeding speed feature calculates the material motion vector using optical flow; and the material morphology feature assesses the aggregation state through contour analysis and texture quantification, thereby achieving a comprehensive digital representation of the feeding process.

[0034] The material feeding status analysis module performs multi-level matching between real-time feature vectors and a dynamic threshold model. This model, trained on historical normal data, not only sets independent threshold ranges for each feature parameter but also establishes correlation rules between features. When a sudden drop in the material feeding speed is detected, the module simultaneously checks whether the change in the material feeding quantity violates the proportional relationship rule and, combined with material morphology characteristics, determines whether agglomeration occurs. This multi-feature collaborative analysis mechanism can identify latent anomalies such as flow asymmetry that existing single-point sensors cannot detect. The multi-level response decision module maps anomaly levels and durations to tiered strategies. For short-term, minor anomalies, it triggers warning commands; for persistent or severe anomalies, it generates adjustment commands including compensation amounts; and for material interruption risks, it triggers shutdown commands, forming a progressive response system. The control command issuance module converts abstract commands into specific equipment drive signals. Through protocol conversion, it drives audible and visual alarms, forced feeding mechanisms, or emergency stop interfaces to execute corresponding actions, ultimately achieving closed-loop management from visual perception to execution control, effectively preventing product quality defects and unplanned downtime caused by latent anomalies.

Claims

1. A vision-based extruder, characterized in that, It includes a physical device and a control device, wherein the control device establishes a communication connection with the physical device, characterized in that: The physical device includes an image acquisition unit, a protection unit, a lighting unit, and a control execution unit; The image acquisition unit includes a main camera and an auxiliary camera. The main camera is mounted at a 45-degree downward angle directly above the extruder bushing discharge port to acquire global discharge trajectory images. The auxiliary camera is mounted horizontally to the side of the bushing discharge port to acquire local images of the material being discharged along the inner wall. The protective unit uses a high-temperature resistant, dustproof, and sealed structure to encapsulate the main camera and the auxiliary camera. The lighting unit provides uniform fill light for the main camera and the auxiliary camera; The control execution unit includes an audible and visual alarm, a forced feeding mechanism drive interface, and an extruder main control system emergency stop interface; The control device includes an image acquisition module, a visual feature extraction module, a material feeding status analysis module, a multi-level response decision module, and a control command issuance module. The visual feature extraction module uses a Gaussian mixture model to establish a pixel color probability distribution model for the background image sequence. By matching the pixel values ​​of the new input image with the background model, the moving material region is segmented. The total number of foreground pixels in the moving material region is counted and multiplied by the actual pixel area conversion factor to obtain the material feeding feature. The median flow velocity of the optical flow field in the moving material region between consecutive frames is calculated to obtain the material feeding speed feature. The circularity of the outline of the moving material region, the tortuosity of the boundary, and the contrast of the gray-level co-occurrence matrix inside the region are analyzed to obtain the material morphology feature. The material feeding status analysis module matches the feature vector with the dynamic threshold model. The dynamic threshold model defines a range of positive proportional relationship between the material feeding quantity feature and the material feeding speed feature. The matching process includes: when the material feeding speed feature value decreases, checking whether the decrease in the material feeding quantity feature value is within the range allowed by the positive proportional relationship, so as to identify material bridging or blockage abnormalities. The multi-level response decision module generates corresponding response instructions based on the status determination results; The control command issuing module converts the first-level early warning command into a pulse width modulation signal and sends it to the audible and visual alarm; converts the second-level adjustment command into an analog voltage signal or a Modbus TCP message and sends it to the forced feeding mechanism drive interface to drive the forced feeding mechanism to generate a feeding compensation amount or perform a dispersing intervention; and converts the stop command into a switching signal that triggers the closing of the relay dry contact and sends it to the emergency stop interface.

2. The extruder based on vision inspection according to claim 1, characterized in that, The image acquisition module synchronously triggers the main camera and the auxiliary camera, including sending synchronization pulse signals to the main camera and the auxiliary camera; The image acquisition module performs non-uniformity correction on the acquired raw image data stream, including calling a pre-stored calibration parameter matrix to perform gain and bias compensation on the pixels of the raw image. The compensation process involves converting the original image data with a high bit depth to an 8-bit depth.

3. The extruder based on vision inspection according to claim 2, characterized in that, The visual feature extraction module segments the moving material region from the image sequence, including using a Gaussian mixture model to establish a pixel color probability distribution model of the background image sequence, and matching the pixel values ​​of the new input image with the background model to segment the moving material region. Extracting the material feeding amount feature includes counting the total number of foreground pixels within the moving material area and multiplying it by the pixel actual area conversion factor; Extracting the feeding speed features includes calculating the median flow velocity of the optical flow field in the moving material region between consecutive frames; Extracting the material morphological features includes calculating the roundness of the outline of the moving material region, the tortuosity of the boundary, and the contrast of the gray-level co-occurrence matrix within the region.

4. The extruder based on vision inspection according to claim 3, characterized in that, The multi-level response decision module queries the response strategy table based on the state determination result, including: When the status determination result is a minor abnormality and the duration is less than the set threshold, a level one warning instruction is generated. When the status determination result is a minor abnormality and the duration exceeds the set threshold, or when the result is a serious abnormality, a secondary adjustment instruction including the material supply compensation amount is generated. When the status determination result indicates that the material flow is completely interrupted, a shutdown command is generated.

5. The extruder based on vision inspection according to claim 4, characterized in that, The dynamic threshold model can be updated online according to changes in the production formula; the dynamic threshold model allows users to configure and adjust it according to process requirements.

6. The extruder based on vision inspection according to claim 5, characterized in that, The dynamic threshold model is updated online based on changes in the production formula, including: Receive formula switching signals, and after formula switching, acquire image sequences of normal production operation under the new formula; Based on the newly acquired image sequence, the corresponding feature vector is extracted through the visual feature extraction module; The normal fluctuation range of each feature parameter and the correlation rules between feature parameters are recalculated using the newly extracted feature vector set to update the dynamic threshold model.

7. The extruder based on vision inspection according to claim 6, characterized in that, include: The image acquisition unit includes a main camera mounted at a downward angle directly above the extruder bushing discharge port and an auxiliary camera mounted horizontally to the side of the bushing discharge port. The control execution unit includes an audible and visual alarm, a forced feeding mechanism drive interface, and an extruder main control system emergency stop interface. A control device, communicatively connected to the image acquisition unit and the control execution unit, comprising: The image acquisition module is used to synchronously trigger the main camera and the auxiliary camera to acquire raw image data streams, perform non-uniformity correction and dynamic range compression on the raw image data streams, and output the processed image sequence; The visual feature extraction module is used to receive the image sequence output by the image acquisition module, segment the moving material region from the image sequence, and extract the material feeding amount feature, material feeding speed feature and material morphology feature of the moving material region to generate a feature vector; The material feeding status analysis module is used to receive the feature vector generated by the visual feature extraction module, match the feature vector with a dynamic threshold model, and output the status judgment result. A multi-level response decision module is used to receive the status judgment result output by the material feeding status analysis module, query a response strategy table based on the status judgment result, and generate control instructions. The control command issuing module is used to receive the control commands generated by the multi-level response decision module, convert the control commands into drive signals, and send the drive signals to the control execution unit.

Images