Defect recognition apparatus and method for a pipeline

Abstract

The application discloses a kind of blemish identification device and method for pipeline, the blemish identification device includes a host and several slaves, host and slave all include processing chip and camera, the processing chip of host is greater than the processing chip of slave, the slave is used to obtain the slave local image of detection target and identify the blemish parameter in local image, and local image and blemish parameter are transmitted to the host;The host is used to obtain the host local image of detection target and identify the blemish parameter in local image, and the host carries out data fusion to slave local image, host local image and all blemish parameters to obtain the blemish data of detection target.The application can realize the automatic detection of the blemish of various factory production line production articles, realize the effect of simple installation, low cost, the detection result is more accurate, greatly improves production efficiency and product qualification rate, and simultaneously can also reduce production cost.

Description

Technical Field

[0001] This invention relates to a defect identification device and method for an assembly line. Background Technology

[0002] In the quality control system of industrial manufacturing, defect detection is a crucial link in ensuring product reliability, safety, and pass rate. Traditional detection methods, as well as the emerging automated and intelligent detection technologies, are constantly evolving, but many core problems remain unresolved. Industrial scenarios are basically divided into manual inspection and machine inspection. Using manual visual inspection to detect defects in industrial products is inefficient and prone to causing visual fatigue. Machine inspection solutions are currently complex to install, lack adaptability, require customization for different products, are too costly, and lack versatility. Furthermore, current machine inspection systems largely rely on powerful hardware (high-performance chips + large-target sensors + high-definition lenses + large controllers, etc.), which are subject to market fluctuations leading to unstable costs and supply. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the shortcomings of the existing technology in the detection of defects in production line products, especially the low accuracy of the detection results of defects in fabric woven on looms, the complexity of equipment installation, and the serious impact on production efficiency. The present invention provides a defect identification device and method for production lines that achieves simple installation, low cost, more accurate detection results, significantly improves production efficiency and product qualification rate, and at the same time reduces production costs.

[0004] The present invention solves the above-mentioned technical problems through the following technical solution:

[0005] A defect identification device is characterized in that it includes a host and several slave devices, each of which includes a processing chip and a camera. The processing chip of the host has a higher computing power than the processing chips of the slave devices.

[0006] The slave device is used to acquire a local image of the target being detected and identify defect parameters in the local image, and then transmit the local image and defect parameters to the host device.

[0007] The host is used to acquire a local image of the target and identify defect parameters in the local image. The host performs data fusion on the slave local image, the host local image and all defect parameters to obtain defect data of the target.

[0008] Preferably, the defect identification device includes a fixed rod and a switch, the slave and master units are both fixed on the fixed rod, the detection target is located below the fixed rod, and the signal lines of the master and slave units are fixed along the fixed rod and connected to the switch.

[0009] Preferably, the defect identification device includes a supplementary light, and the host obtains the polarization angle and brightness of the supplementary light according to the material or image analysis data of the preset detection target, and sets the supplementary light to provide supplementary illumination to the detection target according to the polarization angle and brightness.

[0010] Preferably, the defect identification device includes a rectangular detection area, the detection target moves along the width direction of the detection area, the host local image and the slave local image include an overlapping area in the detection area, the entire overlapping area covers the detection area, the host is used to perform weighted fusion on the defect parameters sent by each slave to obtain a comprehensive confidence level, and judge whether the defect parameters sent by the slave are valid based on the threshold of the comprehensive confidence level, the size of the corresponding defect category and the frequency of occurrence, and add the judgment result as data to the defect data of the detection target.

[0011] Preferably, the host is used for:

[0012] A fused image of local images from the slave device and the host device is acquired, and the location of defects in the fused image is identified, wherein the location of the defect is the position of the defect on the target being detected.

[0013] A spatial model of the defect location is established using local images from the slave and master cameras at different times;

[0014] The shadow shape of the spatial model at the target time is predicted based on the spatial model, the polarization angle, and the brightness.

[0015] Acquire the fused image at the target time, and compare the shadow shape with the defect location pixels in the fused image to obtain comparison information;

[0016] The overall confidence level is obtained by using comparison information, host defect parameters, and slave defect parameters, wherein the larger the absolute value of the Z-axis of the spatial model, the higher the weight of the comparison information in the overall confidence level.

[0017] Preferably, the slave device is used to encrypt the slave device partial image and defect parameters and then transmit the data to the host device. The host device is used to verify the slave device partial image and defect parameters sent by the slave device and determine whether the verification result is valid. If it is valid, the host device performs data fusion on the slave device partial image, the host device partial image and all defect parameters to obtain the defect data of the detection target.

[0018] Preferably, the defect identification device is used for fabric inspection, the inspection target is fabric, and the host is used to acquire defect data of the inspection target. The defect data includes defect location, defect image, defect type, defect size, inspection time, and corresponding process parameters.

[0019] Preferably, the defect identification device uses defect data and the corresponding process parameters to train an artificial intelligence association model, and uses the association model to generate production process data.

[0020] The present invention also provides a defect identification method, characterized in that the defect identification method utilizes the defect identification device described above for defect detection.

[0021] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of the present invention.

[0022] The positive and progressive effects of this invention are as follows:

[0023] This invention enables automated detection of defects in products produced on assembly lines, especially knitted fabrics. It achieves simple installation, low cost, more accurate detection results, significantly improves production efficiency and product qualification rate, and also reduces production costs. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the defect identification device according to Embodiment 1 of the present invention.

[0025] Figure 2 This is a flowchart of the defect identification method according to Embodiment 1 of the present invention. Detailed Implementation

[0026] The present invention will be further illustrated by way of embodiments below, but the present invention is not limited to the scope of the embodiments described herein.

[0027] Example 1

[0028] See Figure 1 This embodiment provides a defect identification device for detecting product quality. In this embodiment, the detection target is the fabric of a loom, that is, this application uses a loom as an example to illustrate the implementation of the technical solution, but it does not mean that the scope of protection of this application is limited to the detection of fabric on a loom. Any implementation method of detecting product surface defects through a vision solution on a production line, as well as various changes or modifications to the implementation method, all fall within the protection scope of this invention.

[0029] The defect identification device includes a master unit and several slave units. Both the master unit and the slave units include a processing chip and a camera. The processing chip of the master unit has a higher computing power than the processing chip of the slave units.

[0030] The slave device is used to acquire a local image of the target being detected and identify defect parameters in the local image, and then transmit the local image and defect parameters to the host device.

[0031] The host is used to acquire a local image of the target and identify defect parameters in the local image. The host performs data fusion on the slave local image, the host local image and all defect parameters to obtain defect data of the target.

[0032] The defect identification device includes a fixed rod and a switch. The slave and master units are both fixed on the fixed rod. The detection target is located below the fixed rod. The signal lines of the master and slave units are fixed along the fixed rod and connected to the switch.

[0033] The defect identification device includes a supplementary light. The host obtains the polarization angle and brightness of the supplementary light based on the material or image analysis data of the preset detection target, and sets the supplementary light to provide supplementary illumination to the detection target according to the polarization angle and brightness.

[0034] The defect identification device includes a rectangular detection area. The target to be detected moves along the width of the detection area. The host local image and the slave local image include an overlapping area in the detection area. The entire overlapping area covers the detection area. The host is used to perform weighted fusion on the defect parameters sent by each slave to obtain a comprehensive confidence score. The host determines whether the defect parameters sent by the slave are valid based on the threshold of the comprehensive confidence score, and adds the determination result as data to the defect data of the target to be detected.

[0035] Specifically, the host is used for:

[0036] A fused image of local images from the slave device and the host device is acquired, and the location of defects in the fused image is identified, wherein the location of the defect is the position of the defect on the target being detected.

[0037] A spatial model of the defect location is established using local images from the slave and master cameras at different times;

[0038] The shadow shape of the spatial model at the target time is predicted based on the spatial model, the polarization angle, and the brightness.

[0039] Acquire the fused image at the target time, and compare the shadow shape with the defect location pixels in the fused image to obtain comparison information;

[0040] The overall confidence level is obtained by using comparison information, host defect parameters, and slave defect parameters, wherein the larger the absolute value of the Z-axis of the spatial model, the higher the weight of the comparison information in the overall confidence level.

[0041] The slave device is used to encrypt local images and defect parameters before transmitting data to the host device.

[0042] The host is used to verify the slave local image and defect parameters sent by the slave and to determine whether the verification result is valid. If valid, the host performs data fusion on the slave local image, the host local image and all defect parameters to obtain the defect data of the detection target.

[0043] The defect identification device is used for fabric inspection, and the inspection target is fabric.

[0044] The host computer is used to acquire defect data of the target to be detected. The defect data includes defect location, defect image, defect type, defect size, detection time, and corresponding process parameters.

[0045] The defect identification device uses defect data and the corresponding process parameters to train an artificial intelligence association model, and uses the association model to generate production process data.

[0046] This embodiment uses:

[0047] Central power supply and communication hub: It adopts a multi-port PoE switch, which is powered by a single power adapter and provides data exchange and power distribution functions for the system. It bids farewell to the traditional adapter power supply mode. Low computing power means low power consumption. Even multiple machines can achieve PoE power supply and network port communication through network cables. It is easy to install, aesthetically pleasing and efficient.

[0048] Computing power collaborative processing system:

[0049] One high-performance main camera: It uses a domestically produced Rockchip high-performance chip as the main control unit of the system, and integrates various sensors and image acquisition chips;

[0050] Seven or more low-computing-power slave cameras: using domestically produced Rockchip low-computing-power chips, responsible for image acquisition and preliminary processing;

[0051] It forms a "1+n" heterogeneous computing power architecture to achieve optimized allocation of computing resources; at the same time, it enables large-scale defect detection.

[0052] Intelligent supplemental lighting system:

[0053] Integrated brightness-adjustable fill light assembly;

[0054] It supports dynamic angle adjustment and 256 levels of brightness adjustment; for defects on different types of objects, the angle and brightness can be adjusted by the camera's exposure GAIN value, increasing brightness while reducing reflections to adapt to the sensor and achieve optimal shooting conditions. Applicable to various types of objects to be detected, the LED light panel integrates various colors and wavelengths of LEDs. It can intelligently identify and adjust the light color and wavelength for different materials to achieve higher detection efficiency; avoiding the cumbersome work of traditional single-function devices that require reinstalling corresponding equipment for different items.

[0055] Intelligent detection system:

[0056] The spatiotemporal dual-mode detection algorithm specifically includes slave-machine motion defect detection and master-machine fusion algorithm; a process adaptive learning engine to establish a knowledge base linking defects and process parameters; and intelligent image stitching, which can freely match the number of cameras according to the length of the object being detected.

[0057] Human-computer interaction system:

[0058] External LCM touchscreen with configurable function buttons and 4G / WiFi dual-network communication module.

[0059] Communication encryption system:

[0060] Domestically produced encryption chips prevent software and hardware theft, de-jitter and encrypt communication between devices, resulting in more stable and reliable operation.

[0061] This application is applicable to various application scenarios and multi-device linkage solutions. It can be flexibly configured according to the size of the object being detected. For large object detection, the number of slave devices can be increased appropriately, while for small object detection, the number of slave devices can be reduced. The host can intelligently stitch together the collected image data into a complete image, which is used for algorithm recognition. Multiple multi-functional universal input and output ports are reserved, and users can customize the specific functions required by the factory according to their own needs.

[0062] See Figure 2 Using the aforementioned defect identification device, this embodiment also provides a defect identification method, including:

[0063] Step 100: Power on the system and each hardware module performs a self-test.

[0064] Step 101: Device security authentication, which involves encrypted authentication between the host and slave devices to ensure that the system consists of authorized devices and prevent tampering.

[0065] Step 102: Parameter initialization, loading initial configurations such as fabric material, detection parameters, and algorithm model.

[0066] Step 103: The host obtains the polarization angle and brightness of the supplementary light according to the material or image analysis data of the preset detection target, and sets the supplementary light to supplement the detection target according to the polarization angle and brightness.

[0067] The system automatically adjusts the polarization angle and brightness intensity of the fill light based on the preset fabric material or preliminary image analysis from the main camera device to obtain the best image quality and reduce reflection interference.

[0068] Step 104: The slave device acquires a local image of the target and identifies defect parameters in the local image, and transmits the local image and defect parameters to the host device.

[0069] All cameras simultaneously capture images of the fabric.

[0070] Slave processing (spatiotemporal domain - motion detection):

[0071] The low-computing-power slave camera runs a lightweight algorithm to analyze defects in the fabric during movement in real time. It features fast computation speed and emphasizes real-time performance.

[0072] Host processing (spatiotemporal domain - static recognition):

[0073] High-performance mainframe cameras run complex deep learning models to perform high-precision image analysis, identifying subtle or complex static defects such as dirt and damage. They prioritize high computational accuracy.

[0074] Step 105: The host acquires a local image of the target and identifies defect parameters in the local image;

[0075] Step 106: The host machine performs data fusion on the slave machine partial image, the host machine partial image, and all defect parameters to obtain defect data of the detection target.

[0076] Step 107: Use the defect data and the corresponding process parameters to train an artificial intelligence association model, and use the association model to generate production process data.

[0077] The defect data identified by the device is compared with the standard defect model in the database. At the same time, a defect similarity ratio requirement is predetermined. Defects above this ratio are considered to be of the same quality and are assigned the same or similar process parameters. The association model is used to continuously generate new production process data and continuously superimpose the algorithm database to achieve higher production process requirements.

[0078] The defect recognition device includes a rectangular detection area. The target to be detected moves along the width of the detection area. The host partial image and the slave partial image include an overlapping area in the detection area, and the entire overlapping area covers the detection area. Step 106 includes:

[0079] The host performs weighted fusion of the defect parameters sent by each slave to obtain a comprehensive confidence level;

[0080] The validity of the defect parameters sent by the slave device is determined based on the threshold of the comprehensive confidence level, and the determination result is added as data to the defect data of the detection target.

[0081] The master receives preliminary detection results from all slaves and its own high-precision results.

[0082] The fusion algorithm performs weighted fusion of multiple detection results for each suspected defect (e.g., Score_M comes from motion detection, and Score_S comes from static recognition) to calculate the final comprehensive confidence score.

[0083] Based on a preset threshold, it determines whether a defect is genuine. This mechanism effectively reduces the false alarm rate.

[0084] In cases where the error is not a false alarm, the system immediately triggers a stop signal, notifying the fabric inspection machine to cease operation. Simultaneously, it records all detailed information about the defect (image, type, location, time, and the process parameters at the time).

[0085] It's a false alarm: the system doesn't trigger any action and continues normal operation and testing.

[0086] Specifically, the defect identification method:

[0087] A fused image of local images from the slave device and the host device is acquired, and the location of defects in the fused image is identified, wherein the location of the defect is the position of the defect on the target being detected.

[0088] A spatial model of the defect location is established using local images from the slave and master cameras at different times;

[0089] The shadow shape of the spatial model at the target time is predicted based on the spatial model, the polarization angle, and the brightness.

[0090] Acquire the fused image at the target time, and compare the shadow shape with the defect location pixels in the fused image to obtain comparison information;

[0091] The overall confidence level is obtained by using comparison information, host defect parameters, and slave defect parameters, wherein the larger the absolute value of the Z-axis of the spatial model, the higher the weight of the comparison information in the overall confidence level.

[0092] The slave device is used to encrypt local images and defect parameters before transmitting data to the host device. The defect identification method includes:

[0093] Verify the slave local image and defect parameters sent by the slave device and determine whether the verification result is valid. If valid, perform data fusion on the slave local image, the master local image and all defect parameters to obtain the defect data of the detection target.

[0094] The defect identification device is used for fabric inspection, the inspection target is fabric, and the defect data includes defect location, defect image, defect type, defect size, inspection time, and corresponding process parameters.

[0095] This application will establish and update a knowledge base. Regardless of whether any defects are detected, the system will associate the current production process parameters with the test results and store them in the database.

[0096] The process adaptive learning engine periodically analyzes historical data to establish a correlation model between "process parameters" and "defect features".

[0097] When the system detects that a certain process parameter setting is prone to causing specific defects, it will automatically generate process optimization suggestions and feed them back to the production process to reduce the generation of defects from the source and achieve the goal of long-term optimization of the detection success rate.

[0098] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of the present invention is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but all such changes and modifications fall within the scope of protection of the present invention.

Claims

1. A blemish recognition apparatus for a pipeline, characterized by, The flaw recognition device includes a master and a plurality of slaves, and the master and the slaves each include a processing chip and a camera, and the processing chip of the master has greater computing power than the processing chip of the slave, The slave is configured to acquire a slave local image of a detection target and identify a flaw parameter in the local image, and transmit the local image and the flaw parameter to the master. The master is configured to acquire a master local image of the detection target and identify a flaw parameter in the local image, and perform data fusion on the slave local image, the master local image, and all the flaw parameters to obtain flaw data of the detection target. The flaw recognition device includes a fixed rod and a switch, and the master and the slaves are fixed on the fixed rod, the detection target is arranged below the fixed rod, and signal lines of the master and the slaves are fixed along the fixed rod and connected with the switch. The flaw recognition device includes a light supplement lamp, the master acquires a polarization angle and brightness of the light supplement lamp according to a material or image analysis data of a preset detection target, sets the light supplement lamp to supplement light for the detection target according to the polarization angle and the brightness. The flaw recognition device includes a rectangular detection area, the detection target moves along a width direction of the detection area, the master local image and the slave local image include an overlapping area in the detection area, and all the overlapping areas cover the detection area, the master is configured to perform weighted fusion on the flaw parameter sent by each slave to obtain a comprehensive confidence, judge whether the flaw parameter sent by the slave is valid according to a threshold of the comprehensive confidence, a size of a corresponding flaw category, and an occurrence frequency, and add a judgment result as data to the flaw data of the detection target.

2. The apparatus for defect identification for a pipeline of claim 1, wherein, The master is configured to: acquire a fusion image of the slave local image and the master local image, identify a flaw position in the fusion image, and the flaw position is a position of the flaw on the detection target; establish a spatial model of the flaw position by using the slave local image and the master local image at different moments; predict a shadow shape of the spatial model at a target moment according to the spatial model, the polarization angle, and the brightness; acquire a fusion image at the target moment, compare the shadow shape with a flaw position pixel in the fusion image to obtain comparison information; acquire the comprehensive confidence by using the comparison information, the flaw parameter of the master, and the flaw parameter of the slave, and the greater the absolute value of the Z-axis of the spatial model is, the higher the weight of the comparison information in the comprehensive confidence is.

3. The apparatus for pipelined defect identification of claim 1, wherein, The slave is configured to transmit data of the slave local image and the flaw parameter to the master after encryption, and the master is configured to verify the slave local image and the flaw parameter sent by the slave, judge whether a verification result is legal, and perform data fusion on the slave local image, the master local image, and all the flaw parameters to obtain the flaw data of the detection target if the verification result is legal.

4. The apparatus for pipelined defect identification of claim 1, wherein, The flaw recognition device is configured to detect cloth, the detection target is cloth, the master is configured to acquire flaw data of the detection target, and the flaw data includes a flaw position, a flaw image, a flaw type, a flaw size, a detection time, and corresponding process parameters.

5. The apparatus for pipelined defect identification of claim 4, wherein, The flaw identification device trains an artificial intelligence correlation model by using flaw data and process parameters corresponding to the flaw data, and generates production process data by using the correlation model.

6. A method for defect identification for a pipeline, the method comprising: The flaw identification method performs flaw detection by using the flaw identification device according to any one of claims 1 to 5.

Images