An automatic flash correction system for injection molded parts based on machine vision and ultrasonic combined detection

Abstract

The application discloses an automatic correction system for sprue fins of injection molded parts based on combined detection of machine vision and ultrasonic waves, and relates to the technical field of intelligent manufacturing of injection molded part processing; the system comprises feeding positioning, combined detection, correction execution, re-inspection modules and a central control module which are sequentially connected; the feeding positioning module completes high-precision positioning of the injection molded parts; the combined detection module integrates machine vision and ultrasonic detection units, and synchronously realizes micron-level accurate detection of the surface parameters and internal extension conditions of the sprue fins; the correction execution module completes profile cutting of the sprue fins and collects the chippings according to the detection data; the re-inspection module verifies the unqualified products again and standardizes the disposal of the unqualified products; and the central control module dynamically adjusts parameters through a self-adaptive control algorithm, records traceable data of the whole process and feeds back to the front-end injection molding machine; the system realizes automation and high-precision management and control of the detection and correction of the sprue fins, and improves the product qualification rate and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology for injection molded parts, specifically to an automatic flash correction system for injection molded parts based on combined machine vision and ultrasonic detection. Background Technology

[0002] Currently, in the field of intelligent manufacturing of injection molded parts, flash detection and correction are key aspects that restrict production efficiency and product quality.

[0003] Existing automated systems mostly rely on single machine vision to detect surface defects. They can only acquire external parameters such as the location and size of the flash, but cannot detect the depth and root condition of the flash extending into the workpiece. This results in a single detection dimension and a high rate of missed detections. The actuators mostly use fixed cutting tools or simple grinding structures, which lack the ability to adapt to contouring. Parameters such as cutting pressure and feed rate need to be preset manually and cannot be adaptively adjusted according to the thickness of the flash, which can easily lead to overcutting, undercutting and workpiece damage. The system lacks secondary inspection and closed-loop verification, making it impossible to verify the effectiveness of corrections and hindering the standardized sorting of defective products. Furthermore, the modules exhibit weak collaboration and data is not interconnected, lacking full-process traceability management and data feedback mechanisms, thus failing to optimize the injection molding process from the source. Additionally, the equipment has poor adaptability to injection molded parts of different specifications and irregular shapes, resulting in a low overall level of automation and intelligence, making it difficult to meet the demands of high-precision, large-scale production. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an automatic flash correction system for injection molded parts based on the combined detection of machine vision and ultrasonic waves.

[0005] The objective of this invention can be achieved through the following technical solutions: This application provides an automatic flash correction system for injection molded parts based on machine vision and ultrasonic joint inspection, including a feeding and positioning module, a joint inspection module, a correction execution module, a re-inspection module and a central control module connected in sequence; The feeding and positioning module is used to perform posture correction and precise positioning of the injection molded part, so that the surface to be inspected and the inspection end of the injection molded part maintain a preset posture. Preferably, the feeding and positioning module includes a clamping and positioning unit, a displacement detection unit, and an attitude adjustment unit; The clamping and positioning unit is used to adapt to injection molded parts of different sizes and specifications and provide controllable clamping force; The displacement detection unit is used to collect the position coordinates of the injection molded part in real time. The posture adjustment unit is used to drive the clamping and positioning unit to rotate, translate and lift according to the position coordinates, so that the surface to be inspected of the injection molded part and the detection end of the joint detection module maintain a preset distance and angle.

[0006] Preferably, the joint detection module includes a machine vision detection unit and an ultrasonic detection unit, used to simultaneously scan the injection molded part and output the surface parameters and internal extension detection data of the flash. The machine vision inspection unit and the ultrasonic inspection unit are integrated on the same inspection bracket and move synchronously to perform a full-coverage scan of the injection molded parts. The machine vision detection unit is used to acquire surface images of injection molded parts. It enhances flash features and suppresses background noise by embedding a spatial attention mechanism into a multi-layer feature fusion target detection network. It also uses a distance intersection ratio non-maximum suppression algorithm to identify the position, length, thickness and distribution range of flash. The ultrasonic detection unit is used to emit high-frequency ultrasonic signals, capture the reflected signals corresponding to the flash, generate a two-dimensional image of the inside of the injection molded part through ultrasonic full-focus imaging technology, and quantify the root thickness and internal extension depth of the flash by combining the machine vision detection results.

[0007] Preferably, the joint detection module also includes a multimodal data fusion unit; The multimodal data fusion unit is used to perform pixel-level registration between the flash surface feature map extracted by the machine vision detection unit and the internal extended two-dimensional imaging map generated by the ultrasonic detection unit to construct a three-dimensional digital defect model of the flash. The registration process employs a non-rigid registration algorithm based on mutual information. Using the flash edge in the visual image as a reference, the root thickness and internal extension depth in the ultrasonic image are mapped to the corresponding pixel coordinates to generate unified defect tensor data that includes surface geometry, internal extension, and root connection state. The unified defect tensor data serves as input to the correction execution module, used to plan the cutting depth, boundary, and feed path for contour correction.

[0008] Preferably, the correction execution module is used to perform contour correction on the flash based on the detection data, and simultaneously collect the debris generated during the correction to complete the correction; The correction execution module includes a contour correction unit, a pressure adjustment unit, and a debris collection unit; The contour correction unit is used to perform contour cutting along the flash contour according to the flash position coordinates and size parameters output by the joint detection module, so as to complete the correction of the flash. The pressure regulating unit is used to monitor the cutting pressure in real time and automatically adjust the cutting pressure and feed rate according to the flash thickness; Among them, low-pressure and low-speed cutting is used for thin-walled flash, and high-pressure and high-speed cutting is used for thick flash. The pressure adjustment unit compares the real-time monitored cutting pressure with the preset threshold through closed-loop control, and dynamically fine-tunes the depth of cut and the moving speed. The debris collection unit is used to collect plastic debris generated during the cutting process of the contour correction unit by means of negative pressure adsorption. The adsorption port of the debris collection unit moves in conjunction with the cutting trajectory of the contour correction unit, and discharges the collected debris into the recycling container after the cutting is completed.

[0009] Preferably, the correction execution module uses an intelligent contour milling head, which includes a miniature six-dimensional force sensor and a high-frequency vibration sensor. The miniature six-dimensional force sensor is used to collect the three-dimensional cutting force and torque in real time during the cutting process, and the high-frequency vibration sensor is used to collect the characteristic vibration spectrum generated by the contact between the tool and the flash. The correction execution module has a built-in flash material identification model. By analyzing the characteristic frequency peaks in the vibration spectrum, it automatically determines whether the flash material is brittle or tough, and dynamically adjusts the spindle speed and cutting strategy: for brittle materials, it adopts impact cutting with high speed, low feed and small depth of cut, and for tough materials, it adopts tensile cutting with low speed, high feed and large depth of cut. The data from the six-dimensional force sensor is also used for overcut warning: when the axial cutting force exceeds the flash root fracture threshold, the tool is immediately retracted and the depth of cut is reduced.

[0010] The re-inspection module uses the same machine vision inspection unit and ultrasonic inspection unit as the joint inspection module to perform a second inspection on the corrected injection molded part. It compares the inspection parameters before and after the correction. If the preset standard is not met, it feeds back to the central control module for a second correction. If the standard is met, the injection molded part is sent to the discharge end. Preferably, the re-inspection module includes a secondary inspection unit, a non-conforming product handling unit, and a self-calibration unit; The secondary detection unit is used to automatically start after receiving the correction completion signal from the correction execution module, and to perform a full-coverage scan of the injection molded part along the preset scanning path, simultaneously acquiring machine vision images of the injection molded part surface and ultrasonic reflection signals inside, and automatically identifying the location, thickness and internal extension depth of the residual flash after correction. The secondary detection unit is also used to compare the identified residual data with the initial detection data output by the joint detection module and the preset qualification standard item by item. If all residual parameters are within the qualification threshold, it is judged as qualified and the injection molded part is sent to the discharge end. If any parameter exceeds the qualification threshold, it is judged as unqualified and the unqualified type and location are fed back to the central control module, which then schedules the correction execution module to perform secondary correction. The non-conforming product processing unit is used to print non-conforming marks on the surface of the injection molded part when the preset standard is still not met after secondary correction, and to sort the non-conforming products into the corresponding collection area according to the defect type. The self-calibration unit is used to periodically and automatically calibrate the detection accuracy of the secondary detection unit.

[0011] Preferably, the self-test calibration unit is used to periodically initiate the self-test process, specifically including: The machine vision inspection unit and ultrasonic inspection unit of the secondary inspection unit are used to verify the accuracy of the standard sample. When the detection deviation exceeds the preset threshold, the camera focal length, the illumination intensity of the ring light component, the scanning angle of the ultrasonic probe and the signal comparison threshold are automatically adjusted until the detection accuracy is restored to the preset range.

[0012] Preferably, the central control module automatically adjusts detection and correction parameters through an adaptive control algorithm, records data throughout the entire process to form a traceable archive, and feeds the data back to the front-end injection molding machine.

[0013] The central control module includes a parameter adaptive adjustment unit, a human-machine interaction unit, and a fault diagnosis unit. The parameter adaptive adjustment unit is used to automatically adjust the scanning frequency and detection accuracy of the joint detection module, as well as the cutting pressure and feed speed of the correction execution module, based on the flash residue data fed back by the re-inspection module, forming a closed-loop control of detection-analysis-adjustment-correction. The human-computer interaction unit is used to display the system's operating status, detection data, correction data, and re-inspection results in real time, and supports manual setting of detection thresholds, correction parameters, and pass standards; The fault diagnosis unit is used to monitor the operating status of each module in real time. When a fault is detected, it automatically identifies the fault type and location, issues an audible and visual alarm, and displays troubleshooting suggestions.

[0014] Preferably, the parameter adaptive adjustment unit specifically includes: When the residual flash data is received from the re-inspection module, the parameter adaptive adjustment unit compares the residual thickness with the preset qualified threshold. If the residual thickness exceeds the qualified threshold, it is determined that the correction is insufficient. Furthermore, based on the ratio of the residual thickness to the original thickness of the flash, the cutting pressure and feed speed of the correction execution module are increased in stages, while the scanning frequency and image acquisition accuracy of the joint detection module are improved during the second inspection. If the residual thickness does not exceed the qualified threshold but scratches or indentations appear on the corrected surface, it is judged as overcutting, and the cutting pressure and feed rate are reduced in stages. At the same time, the signal gain of ultrasonic detection is reduced to reduce false alarms. After each parameter adjustment, the parameter adaptive adjustment unit synchronously updates the adjusted parameter values ​​to each execution module and records the adjustment log to form a traceable parameter change file.

[0015] The beneficial effects of this invention are as follows: This invention achieves full-dimensional, micron-level precision detection of the surface and internal extension of flash on injection molded parts through a combination of machine vision and ultrasonic testing. Combined with multi-axis linkage contour correction and closed-loop control of secondary re-inspection, it solves problems such as missed detection in traditional testing and low accuracy of manual correction, significantly improving the accuracy of flash treatment for injection molded parts and the product qualification rate, and effectively avoiding secondary damage to workpieces. This invention realizes the automated and intelligent management of the entire process of flash detection, correction, and re-inspection of injection molded parts. Each module is coordinated and scheduled by the central control module and the parameters are adaptively adjusted. It can also feed back the data of the entire process to the front-end injection molding machine to optimize the process parameters, reduce the probability of flash generation from the source, reduce manual intervention, improve the continuous operation efficiency of the injection molding production line, and reduce production and maintenance costs. Attached Figure Description

[0016] To better understand and implement this application, the technical solution is described in detail below with reference to the accompanying drawings.

[0017] Figure 1 A schematic diagram of an automatic flash correction system for injection molded parts based on combined machine vision and ultrasonic detection, provided in Embodiment 1 of this application; Figure 2 This is a flowchart illustrating an automatic flash correction system for injection molded parts based on combined machine vision and ultrasonic testing, as provided in Embodiment 1 of this application. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, exemplary embodiments will be described in detail below, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0020] The following detailed description of the specific implementation methods, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided in detail.

[0021] Example 1: See Figure 1-2This embodiment provides an automatic correction system for flash of injection molded parts based on machine vision and ultrasonic joint detection, including a feeding and positioning module, a joint detection module, a correction execution module, a re-inspection module and a central control module connected in sequence; The clamping and positioning unit is used to adapt to injection molded parts of different sizes and specifications and provide controllable clamping force. Furthermore, the clamping and positioning unit includes symmetrically arranged electric grippers and replaceable flexible contouring pads. The electric grippers are controlled in a closed loop by a servo motor and a torque sensor. The clamping force is automatically adjusted according to the material, size and preset clamping force threshold of the injection molded part. The gripper opening stroke is adjustable to accommodate injection molded parts with different shapes and contours. The flexible contouring pads are made of a non-slip polymer material that does not damage the surface of the workpiece.

[0022] The displacement detection unit is used to collect the position coordinates of the injection molded part in real time. Further, the displacement detection unit includes multiple sets of high-precision laser displacement sensors and positioning reference cameras. The laser displacement sensors are arranged in different positions of the injection molded part to obtain the three-dimensional coordinate data of the edge of the injection molded part through high-frequency sampling. The positioning reference camera, together with the ring light source, collects the reference feature image on the injection molded part. The rotation angle and offset of the injection molded part in the plane are calculated by the sub-pixel edge extraction algorithm.

[0023] Specifically, the original image acquired by the positioning reference camera is a grayscale image. The algorithm performs median filtering to denoise the image, grayscale histogram equalization to enhance contrast, and Canny operator to extract pixel-level edge points for preprocessing. The grayscale values ​​are then fitted with a quadratic polynomial or interpolated using the moment method near the gradient direction of the pixel-level edge points to improve the edge positioning accuracy to the sub-pixel level, with a typical positioning error of less than 0.1 pixels. For reference features such as circular positioning holes or rectangular positioning grooves on injection molded parts, the algorithm fits sub-pixel edge points using the least squares method to obtain the center coordinates or corner coordinates of the reference features.

[0024] When the reference feature consists of two positioning circular holes, the algorithm fits the sub-pixel coordinates of the two center points, denoted as P1(x1,y1) and P2(x2,y2). Simultaneously, it obtains the theoretical coordinates of these two references in the standard orientation, P1′(x1′,y1′) and P2′(x2′,y2′), from the CAD model of the injection molded part. The offset and rotation angle of the injection molded part in the plane are calculated. The difference between the measured and theoretical coordinates of one of the reference points is used as the overall translation, i.e., Δx = x1 - x1′, Δy = y1 - y1′. The rotation angle is calculated based on the angle between the measured line connecting the two reference points and the theoretical line connecting them. The formula for calculating the rotation angle θ is: θ = arctan[((y2 - y1)×(x2' - x1') - (x2 - x1)×(y2' - y1'))÷((x2 - x1)×(x2' - x1') + (y2 - y1)×(y2')] - y1'))〕where θ is the rotation angle of the injection molded part about an axis perpendicular to the detection plane.

[0025] When the reference feature is non-circular, the algorithm extracts multiple sub-pixel corner points and estimates the rotation matrix and translation vector through least-squares matching or perspective transformation. Similarly, it calculates the offset and rotation angle. All calculation results, along with the 3D coordinate data acquired by the laser displacement sensor, are packaged and uploaded to the central control module via a real-time communication interface. This drives the attitude adjustment unit for precise compensation. The algorithm can complete the calculation within 10ms per frame, meeting the real-time requirements of the production line.

[0026] The attitude adjustment unit is used to drive the clamping and positioning unit to rotate, translate, and lift according to the position coordinates, so that the surface to be inspected of the injection molded part and the detection end of the joint detection module maintain a preset distance and angle. Further, the attitude adjustment unit includes a multi-axis precision electric platform. The central control module compares the coordinate and angle data fed back by the displacement detection unit with the preset ideal detection attitude and calculates the compensation amount of each axis. It then performs coarse adjustment, fine adjustment, and image stabilization steps in sequence. In the coarse adjustment stage, large deviations are quickly eliminated. In the fine adjustment stage, the distance and angle errors are controlled within the micrometer level and within one-hundredth of a degree by combining the closed-loop feedback of the laser displacement sensor. In the image stabilization stage, the vibration isolation device is activated to suppress external vibration. If the preset attitude cannot be achieved after multiple iterative adjustments, the positioning is judged to have failed and the injection molded part is moved out of the detection channel.

[0027] For example, a certain type of automotive dashboard bracket injection molded part is 200mm long and 120mm wide. The surface to be inspected is the plane where the mold parting line is located. When the feeding and positioning module is working, the clamping and positioning unit automatically adjusts the opening of the electric gripper to 210mm, and the initial clamping force is set to 100N. The three sets of laser sensors of the displacement detection unit measure that the injection molded part is offset by 0.5mm in the X direction and deflected by 0.2° around the Z axis. Based on this, the attitude adjustment unit drives the platform to complete the translation of 0.5mm and the rotation of 0.2° within 1.5 seconds, so that the vertical distance between the surface to be inspected and the inspection end is accurately maintained at 80mm±0.01mm, the parallelism is better than 0.01°, the positioning success rate reaches 99%, and the requirements of subsequent joint inspection are met.

[0028] Specifically, the feeding and positioning module solves the problems of poor adaptability, large posture deviation, insufficient positioning accuracy, and easy damage to the workpiece surface in traditional injection molded parts by combining adaptive clamping, high-precision coordinate acquisition, and multi-axis attitude fine adjustment. It achieves stable clamping and micron-level precise alignment of injection molded parts of different specifications, ensuring that the surface to be inspected and the inspection end maintain a constant distance and angle, providing a stable and reliable positioning foundation for subsequent flash inspection and correction operations, and improving the system's inspection accuracy and operational stability.

[0029] The inspection bracket of the joint inspection module carries a machine vision inspection unit and an ultrasonic inspection unit to perform an S-shaped full-coverage scan along the contour of the injection molded part at a speed of 8 mm / s. The machine vision inspection unit is equipped with 12 high-resolution flash detection cameras with 5-megapixel global shutter, arranged in a circular array on the inspection bracket. With the help of a programmable ring multispectral light source, white, blue and red light are output respectively to enhance the contrast of flash under different plastic materials. The cameras synchronously acquire images of the injection molded part surface at a rate of 30 frames per second. The acquisition range is areas that are prone to flash, such as the parting line, the edge of the ejector pin hole and the root of the buckle. The acquired images are first preprocessed by median filtering for noise reduction and adaptive histogram equalization, and then input into the multi-layer feature fusion target detection network. This network uses ResNet-50 as its backbone and embeds a spatial attention mechanism in each layer of the feature pyramid. An attention weight map is generated through a 1×1 convolutional layer. This weight map is used to represent the importance of each region in the image. Regions with higher weight values ​​correspond to effective features such as fly edges, while regions with lower weight values ​​correspond to noise information such as background texture and surface reflection. The weight map is multiplied element-wise with the original feature map, so that the network focuses on the high-frequency information of fly edges and suppresses irrelevant interference. After the network outputs candidate detection boxes, the distance intersection ratio non-maximum suppression algorithm is used for post-processing. Specifically, this involves adding the ratio of the Euclidean distance between the center points of two candidate boxes to the length of the diagonal of the minimum bounding rectangle to the traditional intersection-union ratio. When the intersection-union ratio of the distance between two candidate boxes is greater than the threshold of 0.25, the box with higher confidence is retained and the redundant box is removed. Finally, the system accurately identifies the position coordinates, length and thickness distribution of 7 flash edges on the mold line of the mobile phone frame. The longest flash edge is 15mm long, the maximum thickness is 0.22mm and the minimum thickness is 0.08mm. The thickness distribution data is output in the form of contour point cloud.

[0030] The ultrasonic detection unit operates synchronously with the machine vision detection unit, employing a 128-element phased array ultrasonic probe with a center frequency of 10MHz and a sampling frequency of 50MHz. The probe maintains a 0.5mm gap with the surface of the injection-molded part to be inspected via water coupling, using deionized water with a flow rate controlled at 10mL / min. After the ultrasonic signal is emitted, a strong reflected echo is generated at the junction of the flash root and the injection-molded part body. The echo signal is processed by a full-focusing algorithm: the detection area is divided into a 0.05mm × 0.05mm grid, and the signal amplitude from all transmit-receive array element pairs is calculated and superimposed at each grid point to form a two-dimensional internal imaging map. In the imaging map, the flash root appears as a high-brightness continuous bright band, while the internal extended area appears as a gradually attenuating secondary bright area. The system combines the surface position of the flash from the machine vision inspection results and extracts the corresponding depth direction amplitude curve from the ultrasonic image. The point where the amplitude drops to three standard deviations of the background noise is taken as the boundary between the flash root and the internal extension, thereby quantifying the root thickness and internal extension depth of the flash. The inspection results show that the root thickness of the flash in the middle section of the mold parting line is 0.18mm, and the internal extension depth into the injection molded part body is 0.35mm; the root thickness of the flash at the edge of the snap fastener is 0.12mm, and the internal extension depth is only 0.05mm. All inspection data, including flash position, length, thickness, root thickness, and internal extension depth, are transmitted to the central control module in real time in the form of structured data packets for the correction execution module to call.

[0031] The entire joint inspection process takes 9 seconds per scan. The test results of 50 mobile phone mid-frame injection molded parts show that the surface size detection error of the flash is less than ±0.01mm and the internal extension depth detection error is less than ±0.02mm. Compared with the traditional single vision inspection method, the false negative rate is reduced from 15% to 1.2%, and the recognition accuracy of internal extension flash reaches 98.5%.

[0032] Furthermore, the joint detection module also includes a multimodal data fusion unit; The multimodal data fusion unit is used to perform pixel-level registration between the surface feature map of the flash extracted by the machine vision detection unit and the internal extension two-dimensional imaging map generated by the ultrasonic detection unit to construct a three-dimensional digital defect model of the flash. Specifically, it includes: performing three-dimensional coordinate transformation between surface feature points and internal depth points according to the spatial mapping relationship obtained by registration; using the flash edge, root section, and extension trajectory as key features, completing three-dimensional spatial reconstruction through voxelization filling; and generating a full-dimensional three-dimensional digital defect model containing surface geometric contours, root section features, and internal extension trajectory.

[0033] The registration process employs a non-rigid registration algorithm based on mutual information. Using the flash edge in the visual image as a reference, the root thickness and internal extension depth in the ultrasonic image are mapped to corresponding pixel coordinates, generating unified defect tensor data containing surface geometry, internal extension, and root connection states. Specifically, this includes: using the machine vision image as a fixed image and the ultrasonic image as a floating image, calculating the grayscale entropy and joint entropy of the two images, and performing iterative optimization with maximizing mutual information as the objective function; constructing a non-rigid deformation field using B-spline free transformation to compensate for sensor installation deviations and local workpiece deformation; using the flash edge as a strong constraint feature to achieve sub-pixel-level alignment between the surface contour and the internal structure; and iterating until mutual information convergence, outputting the spatial mapping relationship to complete pixel-level registration.

[0034] The unified defect tensor data serves as input to the correction execution module, used to plan the cutting depth, boundary, and feed path for contour correction.

[0035] For example, in a scenario involving the inspection of flash on injection-molded automotive dashboard parts, the joint inspection module operates at a speed of 8mm / s. The machine vision inspection unit uses a multispectral blue light source adapted to matte plastic material and scans along the workpiece contour in an S-shape. After median filtering and adaptive histogram equalization preprocessing, the signal is fed into a multi-layer feature fusion target detection network with ResNet-50 as the backbone and embedded with a spatial attention mechanism. The distance intersection ratio non-maximum suppression algorithm accurately identifies the position, surface thickness, and contour distribution of six flash edges at the mold line and corners. The ultrasonic inspection unit simultaneously uses a 10MHz, 128-element phased array ultrasonic probe to collect the reflection signal at the root of the flash edge using deionized water coupling. The internal two-dimensional imaging map is generated by a full-focusing algorithm to quantify the root thickness and internal extension depth. The multimodal data fusion unit performs pixel-level registration between machine vision surface feature maps and ultrasonic internal imaging maps, using the visual image as the fixed image and the ultrasonic image as the floating image. It completes sub-pixel-level alignment based on mutual information non-rigid registration algorithm and B-spline free transformation. According to the spatial mapping relationship, it performs three-dimensional coordinate transformation between surface feature points and internal depth points. Through voxel filling, it constructs a three-dimensional digital defect model containing surface geometric contours, root cross-sectional features, and internal extension trajectories, and generates unified defect tensor data. A single detection takes 10 seconds, with surface detection error less than ±0.01mm, internal depth detection error less than ±0.02mm, and no missed internal extension flash edges.

[0036] Specifically, the joint inspection module uses machine vision and ultrasound to conduct simultaneous joint inspections, combining multimodal data fusion and 3D modeling. This solves the problems of traditional single-vision inspection, such as the inability to identify the internal extension state of flash, easy to miss detection, inaccurate positioning, and insufficient data dimensions. It achieves full-dimensional, micron-level precise quantification of surface and internal defects. At the same time, relying on intelligent algorithms and closed-loop data output, it significantly reduces the missed detection rate and improves the detection accuracy and stability.

[0037] The correction execution module is used to perform contour correction on the flash according to the detection data, and simultaneously collect the debris generated during the correction to complete the correction; the correction execution module includes a contour correction unit, a pressure adjustment unit and a debris collection unit.

[0038] The contour correction unit is used to perform contour cutting along the flash contour based on the flash position coordinates and size parameters output by the joint detection module, thereby correcting the flash. The contour correction unit is equipped with a tool magazine and an automatic tool changer. The tool magazine contains precision cutting heads, scraping heads, and arc trimming heads. The central control module automatically matches the optimal tool head model and issues a tool change command based on the flash thickness, contour curvature, and location type. The tool magazine completes rapid tool head replacement through a rotary indexing and telescopic locking mechanism, realizing automatic switching of cutting methods for different flash shapes. For long strip-shaped parting line flash, the tool automatically switches to a precision cutting head for straight-line cutting; for small clip root flash, the tool automatically switches to a scraping head for micro-scraping; and for arc transition area flash, the tool automatically switches to an arc trimming head for smooth contouring.

[0039] Furthermore, the contour correction unit uses cubic B-spline curve fitting to generate a smooth and continuous three-dimensional toolpath trajectory based on the flash contour point cloud data, and drives the cutting mechanism to perform a contour-following motion along the flash contour. To ensure that the cutter head always fits the root surface of the flash for cutting, the contour correction unit is also equipped with a laser rangefinder to measure the distance between the cutter head and the reference surface of the injection molded part in real time, and dynamically adjusts the Z-axis height of the cutter head in combination with the flash thickness data, so that the movement trajectory of the cutter head is consistent with the actual contour of the flash in three-dimensional space.

[0040] The correction execution module uses an intelligent contour milling head, which includes a miniature six-dimensional force sensor and a high-frequency vibration sensor. The miniature six-dimensional force sensor is rigidly embedded between the tool mounting base and the spindle of the intelligent contour milling head. During the cutting process, the X, Y, and Z three-axis cutting forces and the three-axis torques act on the sensor elastic body, causing the strain gauge to deform and causing the output voltage of the Wheatstone bridge to change. After the voltage signal is amplified by the instrument, low-pass filtered and converted by A / D, the built-in calculation chip calculates and outputs the digital quantities of the three-axis cutting force and torque in real time, realizing the real-time continuous acquisition of cutting force and torque. The high-frequency vibration sensor is fixedly installed on the intelligent contour milling head near the tool. The vibration acceleration signal generated by the tool contacting the flash during cutting is picked up by the sensor in real time. After the environmental and transmission noise is filtered out by anti-aliasing filtering, the time domain signal is sampled at a fixed high-frequency sampling rate. Then, the time domain vibration signal is converted into a frequency domain signal by fast Fourier transform, and finally the characteristic vibration spectrum of the tool and flash cutting contact is extracted.

[0041] The correction execution module has a built-in flash material identification model. The flash material identification model uses the offline calibrated standard feature spectrum library of brittle plastics and tough plastics as the identification benchmark. First, it normalizes the real-time collected vibration spectrum data and extracts three types of feature parameters: main peak frequency, harmonic distribution, and amplitude attenuation rate. Then, it performs feature matching and similarity calculation with the standard samples in the spectrum library. By analyzing the characteristic frequency peaks in the vibration spectrum, the material properties are automatically determined: when the characteristic frequency peaks are concentrated in the high-frequency range of 8kHz–20kHz and the amplitude decay rate is greater than 80% / ms, it is determined to be brittle plastic; when the characteristic frequency peaks are concentrated in the low-frequency range of 1kHz–6kHz, the amplitude is stable and the decay rate is less than 30% / ms, it is determined to be tough plastic. Subsequently, the spindle speed and cutting strategy are dynamically adjusted. The central control module outputs tiered control commands based on the discrimination results: brittle materials are cut using impact cutting with a spindle speed of 20,000 r / min–30,000 r / min, a feed rate of 3–8 mm / s, and a depth of cut of 0.02–0.08 mm; ductile materials are cut using tensile cutting with a spindle speed of 5,000 r / min–12,000 r / min, a feed rate of 15–25 mm / s, and a depth of cut of 0.1–0.3 mm. This achieves adaptive matching between cutting parameters and material properties.

[0042] The data from the six-dimensional force sensor is also used for overcut warning; the central control module reads the axial cutting force value in real time and compares it with the preset root fracture threshold. When the axial cutting force exceeds the threshold, an interrupt signal is immediately output to trigger the tool to retract quickly and simultaneously reduce the cutting depth by 0.01–0.03 mm to avoid damaging the injection molded part.

[0043] The pressure regulating unit is used to monitor the cutting pressure in real time and automatically adjust the cutting pressure and feed rate according to the flash thickness. The pressure regulating unit has a built-in high-precision pressure sensor and PID closed-loop calculation module to collect the contact pressure between the cutter head and the surface of the injection molded part in real time and convert it into an electrical signal. The central control module performs digital comparison and calculation between the real-time pressure signal and the preset pressure threshold. First, it calculates the difference between the real-time pressure value and the preset threshold. When the real-time pressure is greater than the preset threshold, it is determined that the cutting load is too high. When the real-time pressure is less than the preset threshold, it is determined that the cutting load is insufficient. The corresponding adjustment amount is output according to the size of the difference. Among them, a flash thickness ≤ 0.1mm is defined as a thin-walled flash, with a corresponding preset pressure threshold of 0.05~0.1MPa and a feed speed of 5~10mm / s; a flash thickness > 0.1mm is defined as a thick flash, with a corresponding preset pressure threshold of 0.15~0.3MPa and a feed speed of 15~25mm / s.

[0044] The pressure regulation unit uses closed-loop control to compare the real-time monitored cutting pressure with preset thresholds item by item, and dynamically fine-tunes the depth of cut and feed rate in stages based on the comparison results and flash thickness parameters. When the absolute value of the difference between the real-time pressure and the preset threshold is less than ±5%, no adjustment is made; when the difference is between 5% and 15%, the depth of cut is finely adjusted by ±0.01mm and the feed rate is finely adjusted by ±5%; when the difference is greater than 15%, the depth of cut is finely adjusted by ±0.02mm and the feed rate is finely adjusted by ±10%.

[0045] The pressure regulation unit superimposes the fine adjustment of the cutting depth to compensate the Z-axis depth command of the original toolpath trajectory of the contour correction unit. It adopts a collaborative control method based on pressure closed-loop feedback priority constraint and trajectory command to keep the actual cutting pressure stable within the preset reasonable range.

[0046] The debris collection unit is used to collect plastic debris generated during the cutting process performed by the contour correction unit through negative pressure adsorption. The adsorption port of the debris collection unit is arranged in a ring around the cutter head, maintaining a 5mm distance from the cutter head, with the adsorption direction pointing towards the cutting point, ensuring that the debris is sucked in by negative pressure before splashing. The adsorption port moves in conjunction with the cutting trajectory of the contour correction unit, so that the adsorption port always covers the cutting debris area, and the adsorption and collection are completed at the moment the debris is generated. After the cutting is completed, the collected debris is discharged into the recycling container.

[0047] For example, when performing flash correction on a mobile phone frame injection molded part, the correction execution module receives flash data output by the joint detection module. The contour correction unit automatically switches the cutting head according to the flash shape. For long strip-shaped parting line flash, a precision cutting cutting head is used; for small flash at the root of the buckle, a scraping cutting head is used; and for flash in the arc transition area, an arc trimming cutting head is used. Based on the flash contour point cloud, a three-dimensional toolpath trajectory is generated using a cubic B-spline curve. In conjunction with a laser rangefinder, the Z-axis height of the cutting head is adjusted in real time to ensure that the cutting head moves precisely to fit the root of the flash. The miniature six-dimensional force sensor of the intelligent contour milling head collects three-dimensional cutting force and torque in real time, and the high-frequency vibration sensor simultaneously picks up cutting vibration. The signal is converted into a characteristic vibration spectrum. The flash material identification model quickly identifies the workpiece as tough plastic based on the spectrum characteristics, and then adaptively adjusts to a stretching cutting strategy with low spindle speed, high feed rate, and large depth of cut. During the cutting process, the pressure adjustment unit adjusts the cutting pressure and feed rate according to the thickness of the flash, and stabilizes the cutting pressure within the preset range through PID closed-loop control. At the same time, the six-dimensional force sensor provides overcut warning in real time, and immediately triggers tool retraction if the cutting force exceeds the limit. The annular adsorption port of the chip collection unit moves synchronously with the cutter head and instantly adsorbs cutting chips in a negative pressure manner. After correction, the residual thickness of the flash is less than 0.01mm, and the workpiece surface is free of scratches and pressure marks.

[0048] Specifically, the correction execution module integrates five functions: automatic tool changing and contour cutting, multi-dimensional force and vibration sensing, material adaptive cutting, pressure closed-loop adjustment, and synchronous chip collection. It adaptively matches cutting strategies and operating parameters based on the flash position, contour, thickness, and material characteristics to accurately remove the root of the flash. At the same time, relying on the force sensor for real-time overcut warning and dynamic pressure adjustment, it avoids overcutting, undercutting, workpiece scratches, and root residue. Combined with synchronous chip collection, it maintains a clean processing environment, achieving high precision, high stability, high adaptability, and automated operation for flash correction of injection molded parts.

[0049] The re-inspection module uses the same machine vision inspection unit and ultrasonic inspection unit as the joint inspection module to perform a second inspection on the corrected injection molded part. It compares the inspection parameters before and after the correction. If the preset standard is not met, it feeds back to the central control module for a second correction. If the standard is met, the injection molded part is sent to the discharge end. The re-inspection module includes a secondary inspection unit, a non-conforming product handling unit, and a self-calibration unit.

[0050] The secondary detection unit is used to automatically start after receiving the correction completion signal from the correction execution module, and to perform a full-coverage scan of the injection molded part along the preset scanning path, simultaneously acquiring machine vision images of the injection molded part surface and ultrasonic reflection signals inside, and automatically identifying the location, thickness and internal extension depth of the residual flash after correction. The post-processing of candidate bounding boxes with burrs employs a non-maximum suppression algorithm based on the intersection-union ratio (IURR). Specifically, this involves: first, sorting all candidate bounding boxes output by the network according to their confidence level from highest to lowest; selecting the box with the highest confidence level as the current baseline box; calculating the IURR between this baseline box and all other candidate boxes; and further incorporating the ratio of the Euclidean distance between the center points of two boxes to the length of the diagonal of the minimum bounding rectangle into the traditional IURR, which more accurately reflects the overlap and positional differences between boxes. A IURR threshold of 0.25 is set. If the IURR between two boxes is greater than this threshold, it is considered a redundant box and is removed; if it is less than or equal to the threshold, it is retained. This process of selecting the baseline box, calculating the IURR, and removing redundant boxes is repeated until all candidate boxes are processed. Finally, the detection box with the highest confidence level and no overlap or redundancy is retained, achieving accurate localization of the burr residue position and contour.

[0051] The secondary detection unit is also used to compare the identified residual data with the initial detection data output by the joint detection module and the preset qualification standards item by item. If all residual parameters are within the qualification threshold, the part is judged to be qualified and the injection molded part is sent to the discharge end. If any parameter exceeds the qualification threshold, the part is judged to be unqualified and the unqualified type and location are automatically marked. This automatic marking process does not require manual intervention. The system automatically generates defect coordinates, defect type and unqualified level information based on the detection results and synchronously feeds it back to the central control module. The central control module schedules the correction execution module to perform secondary correction on the unqualified location.

[0052] Furthermore, the non-conforming product handling unit is used to initiate the non-conforming product handling process when the preset standard is still not met after secondary correction. Specifically, the central control module automatically triggers the handling command based on the secondary inspection results. First, it drives the laser marking mechanism to print a permanent mark containing the defect type, non-conforming location, and inspection time on the non-functional surface of the injection molded part, completing the automatic marking. Then, according to the defect type, the non-conforming products are divided into flash residue, internal extension exceeding the standard, and surface defect categories, and the automatic sorting mechanism is driven to push the workpieces to the corresponding collection area. At the same time, the system stores and archives the non-conforming information, inspection data, and handling results to form a quality traceability record, realizing the standardized and automated handling of non-conforming products.

[0053] The self-calibration unit is used to periodically and automatically calibrate the detection accuracy of the secondary detection unit. Its periodically initiated self-calibration process specifically includes: using a standard sample with a standard size and standard depth burr structure as a calibration benchmark, the machine vision detection unit and ultrasonic detection unit of the secondary detection unit are sequentially checked for accuracy. The machine vision detection unit checks the image detection deviation using the known contour size of the standard sample, and the ultrasonic detection unit checks the internal depth detection deviation using the known depth of the standard sample. When the detection deviation exceeds a preset threshold, the camera focal length, the illumination intensity of the ring light component, the scanning angle of the ultrasonic probe, and the signal comparison threshold are automatically adjusted. Through multiple iterative calibrations, the detection accuracy is restored to the preset range, ensuring the long-term stable operation of the secondary detection unit.

[0054] For example, after the mobile phone mid-frame injection molding part re-inspection module is started, the secondary inspection unit uses the same equipment and parameter standards as the initial inspection to perform a full-coverage scan of the corrected injection molding part; the machine vision unit accurately identifies the residual flash, and the ultrasonic inspection unit simultaneously checks whether there are any uncleaned flash marks inside, comparing the inspection data with the initial inspection data and the pass standards item by item; if the residual flash thickness of a product exceeds the pass threshold of 0.01mm, the system automatically marks the location and feeds it back to the central control module, scheduling the correction execution module to perform a secondary micro-correction; if it still does not meet the standard after the secondary correction, the non-conforming product handling unit is started, the product is sorted and stored separately, and the defect type is recorded for subsequent quality traceability; the self-inspection calibration unit starts every 8 hours, using standard calibration samples to verify the detection accuracy, without the need for manual intervention, so as to achieve smooth output of qualified products and standardized disposal of non-conforming products.

[0055] Specifically, the re-inspection module utilizes a secondary inspection mechanism with the same standards, parameters, and algorithms as the initial joint inspection to achieve a comprehensive review of residual flash and internal defects in the corrected injection molded parts. It relies on a distance intersection ratio (DIP) non-maximum suppression algorithm to ensure accurate and redundant positioning of residual flash. The entire process of inspection, comparison, marking, and feedback is automated, requiring no manual intervention and preventing defective products from flowing into subsequent processes. Through closed-loop scheduling of secondary corrections, the first-pass yield is significantly improved, reducing workpiece waste and rework costs. The classification, disposal, and automatic marking and archiving of defective products enable full traceability of product quality, facilitating subsequent defect analysis and process optimization. A periodic self-calibration mechanism dynamically maintains inspection accuracy, eliminating equipment deviations caused by long-term operation and ensuring the consistency and reliability of inspection results.

[0056] The central control module automatically adjusts detection and correction parameters through an adaptive control algorithm, records data throughout the entire process to form a traceable archive, and feeds the data back to the front-end injection molding machine; The central control module includes a parameter adaptive adjustment unit, a human-machine interaction unit, and a fault diagnosis unit.

[0057] After receiving the residual flash data from the re-inspection module, the parameter adaptive adjustment unit compares the residual thickness with a preset qualified threshold. If the residual thickness exceeds the qualified threshold, it is determined that the correction is insufficient. Then, based on the ratio of the residual thickness to the original flash thickness, the cutting pressure and feed rate of the correction execution module are increased in stages. For example, when the ratio is less than 0.3, the cutting pressure increases by 5% and the feed rate increases by 8%; when the ratio is between 0.3 and 0.6, the cutting pressure increases by 12% and the feed rate increases by 15%; when the ratio is greater than 0.6, the cutting pressure increases by 20% and the feed rate increases by 25%. At the same time, the scanning frequency of the joint detection module during secondary detection is increased from 50Hz to 120Hz, and the image acquisition accuracy is improved from 200dpi to 400dpi.

[0058] If the residual thickness does not exceed the acceptable threshold but scratches or indentations appear on the corrected surface, it is judged as overcutting. In this case, the cutting pressure and feed rate are reduced in stages according to the scratch depth. For example, when the scratch depth is less than 0.02mm, the cutting pressure is reduced by 10% and the feed rate is reduced by 8%; when the scratch depth is greater than or equal to 0.02mm, the cutting pressure is reduced by 20% and the feed rate is reduced by 18%. At the same time, the signal gain of the ultrasonic detection is reduced from 60dB to 45dB to reduce false alarms. After each parameter adjustment, the parameter adaptive adjustment unit synchronously updates the adjusted cutting pressure value, feed rate value, scanning frequency, image acquisition accuracy, and ultrasonic gain to the joint detection module and the correction execution module, and records the adjustment log, including the adjustment time, parameters before adjustment, parameters after adjustment, residual thickness value, and judgment reason, forming a traceable parameter change file.

[0059] The human-machine interface unit displays the system's operating status, detection data such as the original and residual thickness of the flash, correction data such as the current cutting pressure and feed rate, and re-inspection results such as the pass rate and number of overcuts in real time. It also supports operators to manually set detection thresholds, correction parameters such as the maximum cutting pressure limit, and pass standards such as the residual thickness not exceeding 0.05mm and without scratches via the touch screen.

[0060] The fault diagnosis unit monitors the operating status of each module every 50 milliseconds. When it detects faults such as communication interruption of the joint detection module, motor overload of the correction execution module, or sensor failure of the re-inspection module, it automatically identifies the fault type and location. For example, it locates the specific hardware address by comparing the module's heartbeat signal and current feedback, and then issues an audible and visual alarm and displays troubleshooting suggestions on the human-machine interface unit to check the network cable connection of the joint detection module or reset the overheat protection switch of the correction execution module, thereby ensuring the reliability and traceability of the system's closed-loop control.

[0061] For example, in the production of small plastic injection molded parts, the central control module automatically adjusts detection and correction parameters through an adaptive control algorithm, records the entire process data to form a traceable archive, and feeds it back to the front-end injection molding machine. After receiving the flash residue data from the re-inspection module, the parameter adaptive adjustment unit compares the residual thickness of 0.12mm with the preset qualified threshold of 0.05mm, and determines that the correction is insufficient. The ratio of the residual thickness to the original flash thickness of 0.18mm is calculated to be 66.7%, which is greater than 0.6. According to the rules, the cutting pressure is increased by 20%, the feed rate is increased by 25%, and the scanning frequency of the joint detection module is increased from 50Hz to 120Hz, and the image acquisition accuracy is increased from 200dpi to 400dpi. After adjustment, the parameters of each module are updated synchronously, and a log containing the adjustment time, parameters before and after adjustment, residual thickness, and the reason for the judgment is recorded. The human-machine interaction unit displays the original flash thickness, residual thickness, current cutting pressure, feed rate, and re-inspection pass rate of the injection molded part in real time. The operator sets the detection threshold to 0.05mm, the maximum cutting pressure limit, and the pass standard through the touch screen. The fault diagnosis unit monitors each module every 50 milliseconds. When it detects a communication interruption in the joint detection module, it automatically identifies the fault type and location, issues an audible and visual alarm, and displays troubleshooting suggestions on the human-machine interface unit, such as checking the network cable connection of the joint detection module.

[0062] Specifically, the central control module solves the problems of inaccurate parameter adjustment, lack of traceability of full-process data, untimely monitoring of the operating status of each module, low efficiency of fault diagnosis, and inconvenient human-machine interaction during the flash detection and correction of injection molded parts. Through adaptive control algorithms, it automatically and accurately adjusts detection and correction parameters, records full-process data to form a traceable archive and feeds it back to the front-end injection molding machine. With the help of various functional units, it realizes adaptive parameter adjustment, real-time data display and manual setting, automatic fault identification alarm and troubleshooting guidance, ensuring the accuracy and efficiency of flash detection and correction of injection molded parts and the reliability of closed-loop control of the system, avoiding problems of insufficient or excessive correction.

[0063] This embodiment takes the central control module as the core and integrates feeding positioning, machine vision and ultrasonic joint detection, contour correction and re-inspection modules. Through micron-level positioning, full-dimensional synchronous detection of surface and internal flash, adaptive pressure contour cutting, secondary re-inspection closed-loop verification and parameter adaptive adjustment, it realizes the full-process automation and intelligent management of flash detection, correction and re-inspection of injection molded parts. It effectively solves the problems of missed detection, low correction accuracy and lack of closed-loop verification in traditional detection, improves product qualification rate and production efficiency, and the data is traceable and feeds back to optimize the front-end injection molding process.

[0064] Example 2: This example targets large, irregularly shaped injection molded parts for household appliance casings, measuring 600mm long × 400mm wide × 150mm high. These parts have rounded corners and concave-convex snap-fit ​​structures on the sides, making them prone to 0.1-0.5mm thick flash at the parting line and snap-fit ​​connections. Some of this flash extends 2-3mm into the injection molded part body. An automatic flash correction system based on machine vision and ultrasonic joint detection is applied. The adjustable clamping mechanism of the feeding and positioning module adjusts the clamping arm spacing to 620mm×420mm via an electric push rod, adapting to the clamping requirements of large-sized appliance shells. The flexible anti-slip pad uses nitrile rubber with a Shore hardness of 70, and the clamping force is adjusted to 250N to ensure stable clamping of large parts. The displacement sensor group uses a combination of 8 laser displacement sensors and 2 visual positioning sensors to collect the three-dimensional coordinates of the injection molded part. The attitude adjustment motor drives the clamping mechanism to complete translation and rotation fine-tuning, so that the surface of the appliance shell to be inspected is kept at a 90mm vertical distance and a parallelism of ±0.02° with the detection end of the joint inspection module. The entire positioning process takes 8 seconds. The inspection bracket slide of the joint inspection module moves at 10mm / s. The system performs a full-coverage scan along the outline of the appliance casing at high speed. The detection end finely adjusts the angle to 25° for the rounded corners of the casing and to 30° for the buckle structure. Twelve high-definition area array cameras are deployed in a multi-view industrial camera system to achieve image acquisition of the casing without blind spots. The ring-shaped supplementary lighting component adjusts the light intensity to 4000 lux according to the high-gloss plastic material of the casing. The image preprocessing component accurately identifies the surface position and size of eight flashes at the mold line and four flashes at the buckle connection through a multi-layer feature fusion target detection network and DIoU algorithm. The 5-10MHz phased array ultrasonic probe of the ultrasonic detection unit synchronously emits signals, penetrates the thick wall area of ​​the appliance casing, accurately detects the internal extension depth of six flashes, and distinguishes between flashes at two buckle locations and internal shrinkage defects. After the detection data is fused, a full-dimensional flash detection report is generated. A single detection takes 10 seconds. Furthermore, the DIoU algorithm specifically includes: First, the algorithm receives a dataset of candidate detection boxes for appliance casing burrs, output by the image preprocessing component after multi-layer feature fusion and spatial attention enhancement. This dataset contains the coordinates, dimensions, and confidence scores of all suspected burrs at the mold line and buckle locations. Due to the complex structure of the appliance casing buckle, there are many overlaps, redundancies, and positioning deviations in the candidate boxes. Second, the algorithm sorts all candidate detection boxes from high to low confidence scores and selects the burr candidate box with the highest confidence score as the initial reference box. This reference box corresponds to a 5mm long and 0.3mm thick explicit burr at the mold line of the appliance casing. The algorithm extracts the core parameters of the reference box, such as the center coordinates, length, width, and boundary coordinates. Then, the algorithm sequentially calculates the DIoU between all other candidate boxes and the reference box. The algorithm calculates the overlap area ratio of the two frames, simultaneously quantifies the Euclidean distance between their centers, and measures the minimum bounding box size to comprehensively determine the matching degree between the frames. For candidate frames with a small burr gap of ≤0.5mm at the buckle, the DIoU threshold is set to 0.2. When the DIoU value between a candidate frame and the reference frame is greater than this threshold, it is judged as a duplicate detection frame or a redundant frame and is directly removed. If it is less than the threshold, it is retained as a valid candidate frame. Subsequently, among the valid candidate frames after removing redundant frames, the frame with the highest confidence is selected again as the new reference frame. This reference frame corresponds to a small burr with a thickness of 0.1mm at the buckle. The above steps of DIoU value calculation and redundant frame removal are repeated iteratively until all candidate frames have been traversed and judged. Finally, the algorithm outputs a unique and accurate set of burr detection frames after filtering, accurately matching the actual position, contour boundary, and size range of each burr at the appliance shell mold line and buckle. This avoids the problem of traditional algorithms relying solely on overlapping area, which leads to the accidental deletion or repeated detection of dense, small, and adjacent flash. Specifically, the algorithm accurately distinguishes and independently detects three adjacent small flashes at the buckle, providing a precise image recognition basis for the subsequent correction execution module to select scraping heads and set low-pressure, low-speed cutting parameters for small flashes at the buckle, thus ensuring the accuracy and completeness of flash detection in complex parts of large, irregularly shaped home appliance shells. Based on the detection data, the six-axis servo-linked robotic arm of the correction execution module selects a precision cutting head for the long strip of flash at the mold parting line and a scraping head for the small flash at the buckle. The pressure adjustment component adjusts the cutting pressure of the flash at the mold parting line to 0.3MPa and the feed speed to 20mm / s, and the cutting pressure of the flash at the buckle to 0.08MPa and the feed speed to 8mm / s, so as to avoid damage to the concave and convex buckles and rounded corners of the outer shell during the cutting process. The dual suction port negative pressure structure of the debris collection unit moves synchronously with the robotic arm, close to the cutting area to realize the collection of plastic debris as it is produced. A single correction takes 15 seconds. The re-inspection module uses the same detection parameters and scanning path as the joint detection module to perform a second verification on the corrected appliance casing. It compares the flash data before and after correction. For a casing with 0.03mm flash remaining at a buckle, the feedback is sent to the central control module to schedule the correction execution module to perform a second micro-cut. After the second correction, the re-inspection meets the standard. For a defective product that still has the potential for internal flash extension after two corrections, the defect marking unit laser prints the defect number, type and detection time on the non-working surface of the back of the casing and transmits it to the defective product collection area simultaneously. The central control module records the positioning parameters, detection data, two corrected cutting parameters, and re-inspection results of the appliance shell throughout the entire process, forming a traceable file. At the same time, it feeds back the data analysis results of the high incidence of flash on the mold closing line to the front-end injection molding machine, reducing the mold closing pressure of the injection molding machine by 5MPa and the melt temperature by 3℃. In subsequent production, the probability of flash generation of this type of appliance shell is reduced by 60%. The entire system realizes the automated, high-precision detection and correction of flash in injection molded parts of large irregularly shaped appliance shells, and is adapted to the continuous production needs of the appliance injection molding production line.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An automatic flash correction system for injection molded parts based on combined machine vision and ultrasonic detection, characterized in that, It includes a feeding and positioning module, a joint detection module, a correction and execution module, a re-inspection module, and a central control module connected in sequence; The feeding and positioning module is used to perform posture correction and precise positioning of the injection molded part, so that the surface to be inspected and the inspection end of the injection molded part maintain a preset posture. The joint detection module includes a machine vision detection unit and an ultrasonic detection unit, which are used to simultaneously scan the injection molded part and output the surface parameters and internal extension detection data of the flash. The correction execution module is used to perform contour correction on the flash based on the detection data, and simultaneously collect the debris generated during the correction to complete the correction. The re-inspection module uses the same machine vision inspection unit and ultrasonic inspection unit as the joint inspection module to perform a second inspection on the corrected injection molded part. It compares the inspection parameters before and after the correction. If the preset standard is not met, it feeds back to the central control module for a second correction. If the standard is met, the injection molded part is sent to the discharge end. The central control module automatically adjusts detection and correction parameters through an adaptive control algorithm, records data throughout the entire process to form a traceable archive, and feeds the data back to the front-end injection molding machine.

2. The system according to claim 1, characterized in that, The feeding and positioning module includes a clamping and positioning unit, a displacement detection unit, and an attitude adjustment unit. The clamping and positioning unit is used to adapt to injection molded parts of different sizes and specifications and provide controllable clamping force; The displacement detection unit is used to collect the position coordinates of the injection molded part in real time. The posture adjustment unit is used to drive the clamping and positioning unit to rotate, translate and lift according to the position coordinates, so that the surface to be inspected of the injection molded part and the detection end of the joint detection module maintain a preset distance and angle.

3. The system according to claim 1, characterized in that, The machine vision inspection unit and the ultrasonic inspection unit are integrated on the same inspection bracket and move synchronously to perform a full-coverage scan of the injection molded parts. The machine vision detection unit is used to acquire surface images of injection molded parts. It enhances flash features and suppresses background noise by embedding a spatial attention mechanism into a multi-layer feature fusion target detection network. It also uses a distance intersection ratio non-maximum suppression algorithm to identify the position, length, thickness and distribution range of flash. The ultrasonic detection unit is used to emit high-frequency ultrasonic signals, capture the reflected signals corresponding to the flash, generate a two-dimensional image of the inside of the injection molded part through ultrasonic full-focus imaging technology, and quantify the root thickness and internal extension depth of the flash by combining the machine vision detection results.

4. The system according to claim 1, characterized in that, The correction execution module includes a contour correction unit, a pressure adjustment unit, and a debris collection unit; The contour correction unit is used to perform contour cutting along the flash contour according to the flash position coordinates and size parameters output by the joint detection module, so as to complete the correction of the flash. The pressure regulating unit is used to monitor the cutting pressure in real time and automatically adjust the cutting pressure and feed rate according to the flash thickness; Among them, low-pressure and low-speed cutting is used for thin-walled flash, and high-pressure and high-speed cutting is used for thick flash. The pressure adjustment unit compares the real-time monitored cutting pressure with the preset threshold through closed-loop control, and dynamically fine-tunes the depth of cut and the moving speed. The debris collection unit is used to collect plastic debris generated during the cutting process of the contour correction unit by means of negative pressure adsorption. The adsorption port of the debris collection unit moves in conjunction with the cutting trajectory of the contour correction unit, and discharges the collected debris into the recycling container after the cutting is completed.

5. The system according to claim 1, characterized in that, The re-inspection module includes a secondary inspection unit, a non-conforming product handling unit, and a self-calibration unit. The secondary detection unit is used to automatically start after receiving the correction completion signal from the correction execution module, and to perform a full-coverage scan of the injection molded part along the preset scanning path, simultaneously acquiring machine vision images of the injection molded part surface and ultrasonic reflection signals inside, and automatically identifying the location, thickness and internal extension depth of the residual flash after correction. The secondary detection unit is also used to compare the identified residual data with the initial detection data output by the joint detection module and the preset qualification standard item by item. If all residual parameters are within the qualification threshold, it is judged as qualified and the injection molded part is sent to the discharge end. If any parameter exceeds the qualification threshold, it is judged as unqualified and the unqualified type and location are fed back to the central control module, which then schedules the correction execution module to perform secondary correction. The non-conforming product processing unit is used to print non-conforming marks on the surface of the injection molded part when the preset standard is still not met after secondary correction, and to sort the non-conforming products into the corresponding collection area according to the defect type. The self-calibration unit is used to periodically and automatically calibrate the detection accuracy of the secondary detection unit.

6. The system according to claim 5, characterized in that, The self-test calibration unit is used to periodically initiate the self-test process, specifically including: The machine vision inspection unit and ultrasonic inspection unit of the secondary inspection unit are used to verify the accuracy of the standard sample. When the detection deviation exceeds the preset threshold, the camera focal length, the illumination intensity of the ring light component, the scanning angle of the ultrasonic probe and the signal comparison threshold are automatically adjusted until the detection accuracy is restored to the preset range.

7. The system according to claim 1, characterized in that, The central control module includes a parameter adaptive adjustment unit, a human-machine interaction unit, and a fault diagnosis unit. The parameter adaptive adjustment unit is used to automatically adjust the scanning frequency and detection accuracy of the joint detection module, as well as the cutting pressure and feed speed of the correction execution module, based on the flash residue data fed back by the re-inspection module, forming a closed-loop control of detection-analysis-adjustment-correction. The human-computer interaction unit is used to display the system's operating status, detection data, correction data, and re-inspection results in real time, and supports manual setting of detection thresholds, correction parameters, and pass standards; The fault diagnosis unit is used to monitor the operating status of each module in real time. When a fault is detected, it automatically identifies the fault type and location, issues an audible and visual alarm, and displays troubleshooting suggestions.

8. The system according to claim 7, characterized in that, The parameter adaptive adjustment unit specifically includes: When the residual flash data is received from the re-inspection module, the parameter adaptive adjustment unit compares the residual thickness with the preset qualified threshold. If the residual thickness exceeds the qualified threshold, it is determined that the correction is insufficient. Furthermore, based on the ratio of the residual thickness to the original thickness of the flash, the cutting pressure and feed speed of the correction execution module are increased in stages, while the scanning frequency and image acquisition accuracy of the joint detection module are improved during the second inspection. If the residual thickness does not exceed the qualified threshold but scratches or indentations appear on the corrected surface, it is judged as overcutting, and the cutting pressure and feed rate are reduced in stages. At the same time, the signal gain of ultrasonic detection is reduced to reduce false alarms. After each parameter adjustment, the parameter adaptive adjustment unit synchronously updates the adjusted parameter values ​​to each execution module and records the adjustment log to form a traceable parameter change file.

9. The system according to claim 1, characterized in that, The joint detection module also includes a multimodal data fusion unit; The multimodal data fusion unit is used to perform pixel-level registration between the flash surface feature map extracted by the machine vision detection unit and the internal extended two-dimensional imaging map generated by the ultrasonic detection unit to construct a three-dimensional digital defect model of the flash. The registration process employs a non-rigid registration algorithm based on mutual information. Using the flash edge in the visual image as a reference, the root thickness and internal extension depth in the ultrasonic image are mapped to the corresponding pixel coordinates to generate unified defect tensor data that includes surface geometry, internal extension, and root connection state. The unified defect tensor data serves as input to the correction execution module, used to plan the cutting depth, boundary, and feed path for contour correction.

10. The system according to claim 1, characterized in that, The correction execution module uses an intelligent contour milling head, which includes a miniature six-dimensional force sensor and a high-frequency vibration sensor. The miniature six-dimensional force sensor is used to collect the three-dimensional cutting force and torque in real time during the cutting process, and the high-frequency vibration sensor is used to collect the characteristic vibration spectrum generated by the contact between the tool and the flash. The correction execution module has a built-in flash material identification model. By analyzing the characteristic frequency peaks in the vibration spectrum, it automatically determines whether the flash material is brittle or tough, and dynamically adjusts the spindle speed and cutting strategy: for brittle materials, it adopts impact cutting with high speed, low feed and small depth of cut, and for tough materials, it adopts tensile cutting with low speed, high feed and large depth of cut. The data from the six-dimensional force sensor is also used for overcut warning: when the axial cutting force exceeds the flash root fracture threshold, the tool is immediately retracted and the depth of cut is reduced.

Images