Steel surface defect on-line detection method, device, system and storage medium

Abstract

The application provides a steel surface defect online detection method, device, system and storage medium, and relates to the technical field of defect detection. The method comprises the following steps: acquiring a preset reference spot map; wherein the reference spot map is obtained by establishing a coordinate system through a first aluminum nitride spot image of a surface of a defect-free steel sample under a scanning electron microscope; after a target steel leaves an annealing furnace, convex aluminum nitride spots are generated on the surface of the target steel through a pulse nitriding process by using the afterheat of the target steel; a second aluminum nitride spot image of the surface of the target steel is collected, the second aluminum nitride spot image is converted into the coordinate system, and an online spot map is generated; the reference spot map is used as prior knowledge of a pre-trained defect recognition model, the online spot map is input into the defect recognition model, and defects of the target steel are recognized. The application can overcome the interference of the surface condition and the image quality of the steel, and improve the online detection accuracy and efficiency of the steel.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology, and in particular to an online detection method, apparatus, system and storage medium for steel surface defects. Background Technology

[0002] The surface quality of steel directly determines its mechanical properties and service life. This is especially true for special steels such as electrical steel, where surface defects can severely impact their core performance. With the increasing speed and continuity of steel production, efficient and accurate online surface defect detection technology has become a core requirement for quality control.

[0003] Currently, the detection of surface defects in steel mainly relies on manual visual inspection or traditional machine vision inspection. Manual inspection is highly dependent on experience and judgment, while traditional machine vision inspection is based on basic defect morphology recognition algorithms and is widely used in electrical steel inspection due to its non-contact characteristics.

[0004] However, manual inspection is inefficient and has a high rate of false negatives and false negatives, making it unsuitable for high-speed production lines. Traditional machine vision inspection is easily affected by surface conditions such as oil film and the quality of acquired images, resulting in insufficient accuracy in identifying minute defects. At the same time, it relies on a large number of samples, making it difficult to achieve stable and reliable identification when defect samples are scarce.

[0005] Therefore, there is an urgent need for an online detection method for steel surface defects that can overcome interference from steel surface condition and image quality, so as to improve the accuracy and efficiency of online steel detection. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide an online detection method, apparatus, system and storage medium for steel surface defects, so as to improve the accuracy and efficiency of online detection of steel surface defects.

[0007] In a first aspect, embodiments of the present invention provide an online detection method for surface defects in steel, comprising:

[0008] A preset reference spot map is obtained; wherein, the reference spot map is obtained by establishing a coordinate system based on the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope;

[0009] After the target steel leaves the annealing furnace, the residual heat of the target steel is used to generate raised aluminum nitride spots on the surface of the target steel through a pulse nitriding process.

[0010] Acquire images of second aluminum nitride spots on the surface of the target steel, transform the second aluminum nitride spot images into the coordinate system, and generate an online spot map;

[0011] The baseline speckle map is used as prior knowledge for the pre-trained defect recognition model. The online speckle map is then input into the defect recognition model to identify defects in the target steel.

[0012] In one possible implementation, acquiring the second aluminum nitride spot image on the surface of the target steel includes:

[0013] The surface of the target steel is illuminated by a coaxial white light source, and the surface of the target steel is scanned by a line scan camera system;

[0014] Whenever the target steel moves a certain distance, the linear array camera system is controlled to simultaneously acquire images of the target steel surface using blue light and near-infrared light. The blue light image and the near-infrared light image are differentially calculated, and the aluminum nitride spot signal is retained to obtain the second aluminum nitride spot image.

[0015] In one possible implementation, the step of transforming the second aluminum nitride spot image to the coordinate system and generating an online spot map includes:

[0016] Obtain an image of the third aluminum nitride spot on the surface of the first batch of offline steel materials on the same day in the coordinate system using a scanning electron microscope;

[0017] Based on the third aluminum nitride spot image, the second aluminum nitride spot image is transformed into the coordinate system;

[0018] The position coordinates and size parameters of aluminum nitride spots in the second aluminum nitride spot image under the coordinate system are detected to obtain the online spot map.

[0019] In one possible implementation, transforming the second aluminum nitride spot image to the coordinate system based on the third aluminum nitride spot image includes:

[0020] Based on the grain boundary network of the steel, the grain boundary contrast feature points are enhanced and extracted in the third aluminum nitride spot image and the second aluminum nitride spot image, respectively.

[0021] Based on the grain boundary contrast feature point matching relationship between the third aluminum nitride spot image and the second aluminum nitride spot image, the second aluminum nitride spot image is transformed into the coordinate system.

[0022] In one possible implementation, obtaining the preset baseline spot map includes:

[0023] The coordinate system is established with the top left corner of the first aluminum nitride spot image as the origin and pixels as the unit length.

[0024] Based on a preset grayscale threshold, the position coordinates and size parameters of each aluminum nitride spot in the first aluminum nitride spot image are extracted to obtain the preset reference spot map.

[0025] In one possible implementation, during the training of the defect identification model, the penalty weight for incorrect defect identification in the baseline blob map identification area is set to be greater than 1 in the model loss function.

[0026] In one possible implementation, the generation of raised aluminum nitride spots on the surface of the target steel via a pulsed nitriding process includes:

[0027] High-purity ammonia gas is pulsed and injected onto the target steel using a solenoid valve to generate active nitrogen atoms.

[0028] The active nitrogen atoms combine with the aluminum element in the target steel to generate raised aluminum nitride spots on the surface of the target steel that are firmly attached to the target steel matrix.

[0029] Secondly, embodiments of the present invention provide an online detection device for steel surface defects, comprising:

[0030] The benchmark map acquisition module is used to acquire a preset benchmark spot map; wherein, the benchmark spot map is obtained by establishing a coordinate system from the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope;

[0031] The surface spot marking module is used to generate raised aluminum nitride spots on the surface of the target steel by means of the residual heat of the target steel after it leaves the annealing furnace through a pulse nitriding process.

[0032] An online map generation module is used to acquire images of second aluminum nitride spots on the surface of the target steel, convert the second aluminum nitride spot images to the coordinate system, and generate an online spot map;

[0033] The defect detection module is used to use the benchmark spot map as prior knowledge for the pre-trained defect recognition model, input the online spot map into the defect recognition model, and identify the defects of the target steel.

[0034] Thirdly, embodiments of the present invention provide an online detection system for steel surface defects, including a memory and a processor. The memory stores a computer program, and the processor executes the steps of the method as described in the first aspect or any implementation thereof.

[0035] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect or any implementation thereof.

[0036] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0037] In this embodiment of the invention, a reference spot map is obtained by establishing a coordinate system from the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope. This provides a standard reference for the spots on the defect-free steel, clarifying the reference position and distribution characteristics of the aluminum nitride spots. Raised aluminum nitride spots are generated on the surface of the target steel through a pulsed nitriding process. This forms raised and visually identifiable physical markers without damaging the steel's properties, avoiding the impact of online inspection on production efficiency. A second aluminum nitride spot image is acquired from the surface of the target steel and converted to a coordinate system to obtain an online spot map. This quickly generates an online spot map of the target steel in the coordinate system of a scanning electron microscope, achieving a unified coordinate system for the spot positions and ensuring accurate spot comparison. The reference spot map is used as prior knowledge for a pre-trained defect recognition model. The online spot map is input into the defect recognition model to obtain the defect classification results of the target steel, providing standard data for defect recognition and enhancing the model's ability to identify abnormal spot states. The embodiments of the present invention achieve accurate online identification of steel surface defects without damaging the performance of the steel, providing reliable technical support for real-time quality control of high-speed production lines for electrical steel. Attached Figure Description

[0038] Figure 1 This is a schematic diagram illustrating the implementation process of an online detection method for steel surface defects according to an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of the structure of an online steel surface defect detection device provided in an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of an online detection system for steel surface defects provided in an embodiment of the present invention. Detailed Implementation

[0041] The present application will be described more clearly below with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the function of the present application, but do not limit the present application in any way. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application. These all fall within the protection scope of the present application.

[0042] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0043] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0044] In the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0045] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0046] Furthermore, the term "multiple" mentioned in the embodiments of this application should be interpreted as two or more.

[0047] As a core soft magnetic material in equipment such as new energy vehicles and high-end transformers, the surface quality of electrical steel is crucial to the performance and reliability of end products. Minor defects such as surface scratches and indentations can disrupt magnetic domain continuity and affect equipment energy efficiency. The high-speed characteristics of modern electrical steel production lines place real-time and precise demands on defect detection.

[0048] Traditional machine vision inspection uses cameras to acquire images and combines them with algorithms to identify defects. While this enables non-contact inspection, it is susceptible to optical interference from substances such as residual oil films on the steel surface, making it difficult to accurately distinguish between minute defects and false features. Furthermore, the high-resolution equipment and the large number of samples required for model training contribute to high application costs. Eddy current testing and ultrasonic testing, among other non-destructive testing technologies, cannot match the speed of production lines, and their high equipment purchase and maintenance costs are prohibitive for small and medium-sized enterprises. The shortcomings of these existing technologies lie in their inability to balance inspection accuracy, speed, and cost, making comprehensive inspection on production lines difficult. There is an urgent need for an inspection solution that is adaptable to high-speed production lines, has strong anti-interference capabilities, and is cost-effective. Therefore, developing an online inspection method for steel surface defects that can overcome interference from steel surface conditions and image quality has significant practical implications and application value.

[0049] See Figure 1 This invention provides an online detection method for surface defects in steel, detailed below:

[0050] Step S101: Obtain a preset reference spot map; wherein, the reference spot map is obtained by establishing a coordinate system based on the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope.

[0051] In this embodiment of the invention, a reference spot map is obtained by establishing a coordinate system from the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope. This provides a standard reference for the spots of the defect-free steel and clarifies the reference position and distribution characteristics of the aluminum nitride spots.

[0052] Step S102: After the target steel leaves the annealing furnace, the residual heat of the target steel is used to generate raised aluminum nitride spots on the surface of the target steel through a pulse nitriding process.

[0053] In this embodiment of the invention, raised aluminum nitride spots are generated on the surface of the target steel through a pulse nitriding process. This forms raised and visually identifiable physical marks without damaging the steel's properties, thus avoiding the impact of online inspection on production efficiency.

[0054] Step S103: Acquire the second aluminum nitride spot image on the surface of the target steel, and transform the second aluminum nitride spot image into a coordinate system to obtain an online spot map.

[0055] In this embodiment of the invention, an online spot map of the target steel in the coordinate system of a scanning electron microscope is quickly generated, realizing the coordinate system of the spot positions and ensuring accurate comparison of the spots.

[0056] Step S104: Use the benchmark spot map as prior knowledge for the pre-trained defect recognition model, input the online spot map into the defect recognition model, and obtain the defect classification result of the target steel.

[0057] In this embodiment of the invention, the baseline spot map provides standard data for defect identification, enhancing the model's ability to identify abnormal spot conditions.

[0058] The defect identification model is a machine learning model specifically designed for online inspection of high-speed steel production lines. It is capable of real-time identification of minute, real defects on the surface of electrical steel. In steel inspection scenarios, non-real defects such as rolling marks, oil films, uneven lighting, camera shake, or dust obstruction may occur. These surface appearance differences do not affect product function or service life, but they can easily interfere with traditional inspection judgments.

[0059] The defect identification model is based on the following criteria: Real defects cause changes in the material properties of the steel surface (such as the appearance of crystal defects on the steel surface), which in turn causes anomalies in the distribution area, location, and size of the aluminum nitride spots generated by the pulse nitriding process; non-real defects, on the other hand, do not have a substantial impact on the spot characteristics. This invention focuses on changes in spot characteristics based on this difference, achieving accurate identification of real defects and eliminating interference from non-real defects.

[0060] For example, the defect recognition model of this embodiment of the invention employs a convolutional neural network to deeply fuse prior knowledge of a baseline speckle map. The defect recognition model of this embodiment of the invention is described in detail below:

[0061] I. Constructing a Defect Detection Model

[0062] The model employs a dual-branch parallel input structure, with both inputs fed simultaneously into the convolutional neural network for processing, specifically including:

[0063] The online spot map branch (3-channel input) takes three types of images related to the actual spots of the target steel as input, and directly stitches them into a 3-channel format input model to provide the real spot features of the steel to be detected.

[0064] Channel 1: Original online spot map (original image of the spots on the steel to be detected);

[0065] Channel 2: Gray-scale normalized online speckle map (the original online speckle map is uniformly calibrated in terms of gray-scale values ​​to eliminate interference from differences in lighting and equipment acquisition);

[0066] Channel 3: Edge Enhancement Online Blob Map (This enhances the edges of the original online blob map, making the outlines of blobs and potential defects clearer).

[0067] The baseline spot map branch (2-channel input) takes two types of reference images related to spots on defect-free steel samples as input, which are directly stitched together into a 2-channel format and fed into the model to provide prior criteria for spots on defect-free steel samples, as detailed below:

[0068] Channel 1: Spot Location Distribution Map (A spatial distribution reference map of spots in defect-free steel, clearly identifying the location of normal spots);

[0069] Channel 2: Normalized distribution map of spot size (standard reference map of spot size for defect-free steel, clarifying the size range of normal spots).

[0070] After the model acquires the input data, it first extracts features using three types of conventional convolutional kernels. The features are extracted in parallel by each type of kernel before being aggregated, as detailed below:

[0071] The two-branch inputs are processed by the following three types of convolutional kernels, extracting corresponding features simultaneously without interference:

[0072] Size feature convolution kernel: filters out noise and impurities that do not conform to the spot size, and only retains the features of spots with normal size;

[0073] Location feature convolution kernel: captures the arrangement pattern of spots and clarifies the position and spacing between adjacent spots;

[0074] Topological feature convolution kernel: obtains the topological relationships of spots.

[0075] After extraction by the convolutional kernel, the two branches of input data each obtain corresponding spot feature information (the online branch is the spot feature of the steel to be detected, and the baseline branch is the spot feature of the defect-free steel sample). Based on the positional correspondence of these two sets of features, the difference features between the two sets of features are obtained and input into the activation layer, pooling layer and fully connected layer of the convolutional neural network.

[0076] II. Training the Defect Detection Model

[0077] Using a baseline blob map as prior knowledge, combined with defect labels corresponding to the online blob map, the model's dependence on defect samples is reduced through label and multi-task training. The specific process is as follows:

[0078] First, the defect labels are standardized: multi-classification one-hot encoding is used for defect types, and defect location and defect size are normalized.

[0079] Using online blob maps, baseline blob maps, and defect labels as complete samples, a multi-task parallel output header is designed after the fully connected layer, corresponding to defect type, location, size, and confidence level, respectively.

[0080] Establish a joint loss function for the model, including classification loss, location loss, size regression loss, and topology consistency loss. Use the Adam optimizer and validate the output results every 100 rounds. Stop training when the classification accuracy, location bias, and size error all meet the preset model accuracy threshold, and save the model weights.

[0081] III. Identifying defects in the steel under inspection using the trained defect detection model.

[0082] The online spot map of the steel to be inspected and the corresponding baseline spot map are obtained and input into the defect detection model. The model generates prediction results from the multi-task output head, restores the output results to the actual defect type, defect location, and defect size, and outputs the prediction confidence score to indicate the reliability of the defect identification results.

[0083] The embodiments of the present invention achieve accurate online identification of steel surface defects without damaging the performance of the steel, providing reliable technical support for real-time quality control of high-speed production lines for electrical steel.

[0084] In one possible implementation, obtaining a preset baseline spot map includes:

[0085] A coordinate system is established with the top left corner of the first aluminum nitride spot image as the origin and pixels as the unit length;

[0086] Based on a preset grayscale threshold, the position coordinates and size parameters of each aluminum nitride spot in the first aluminum nitride spot image are extracted to obtain a preset baseline spot map.

[0087] This invention provides a reliable standard for spot image comparison and defect identification by establishing a standardized coordinate system and accurately constructing a preset benchmark spot map according to the extracted spot position coordinates and size parameters.

[0088] In one possible implementation, raised aluminum nitride spots are generated on the surface of the target steel using a pulsed nitriding process, including:

[0089] High-purity ammonia gas is pulsed and injected onto the target steel using a solenoid valve to generate active nitrogen atoms.

[0090] By combining active nitrogen atoms with aluminum in the target steel, aluminum nitride spots are generated in situ on the surface of the target steel, forming raised spots that are firmly attached to the target steel matrix.

[0091] This invention provides a stable and identifiable physical marker for identifying defects on steel surfaces by generating raised aluminum nitride spots in situ on the steel surface.

[0092] In some embodiments, when acquiring aluminum nitride spot images of the target steel surface online, impurities such as residual oil films on the steel surface can easily cause optical interference to the imaging process. Conventional lighting equipment and imaging methods make it difficult to separate the interference signal from the spot signal, resulting in low spot image recognition. Therefore, acquiring a second aluminum nitride spot image of the target steel surface may include:

[0093] The surface of the target steel is illuminated by a coaxial white light source, and the surface of the target steel is scanned by a linear array camera system equipped with a polarizer.

[0094] Whenever the target steel moves a certain distance, the control line array camera system simultaneously acquires images of the target steel surface using blue light and near-infrared light. The blue light image and the near-infrared light image are then differentially calculated, and the aluminum nitride spot signal is retained to obtain the second aluminum nitride spot image.

[0095] In this embodiment of the invention, by illuminating with a coaxial white light source and scanning with a linear array camera system equipped with a polarizer, blue light and near-infrared light images of the steel surface are simultaneously acquired and differential calculations are performed. This effectively separates the optical interference signals of impurities such as oil film on the steel surface, accurately preserves the aluminum nitride spot signals, significantly improves the recognition of the second aluminum nitride spot image, and provides high-quality data support for coordinate transformation and defect identification.

[0096] In some embodiments, during the transformation of the second aluminum nitride spot image to a coordinate system, direct matching of the contours of the online detection image and the scanning electron microscope image may produce deviations due to the different scales, affecting the accuracy of the online spot map. Therefore, transforming the second aluminum nitride spot image to a coordinate system and generating an online spot map may include:

[0097] Obtain the third aluminum nitride spot image of the first batch of offline steel surfaces on the same day in the coordinate system using a scanning electron microscope;

[0098] Based on the third aluminum nitride spot image, transform the second aluminum nitride spot image to the coordinate system;

[0099] The position coordinates and size parameters of aluminum nitride spots in the second aluminum nitride spot image under the detection coordinate system are used to obtain an online spot map.

[0100] In this embodiment of the invention, by taking advantage of the consistency in materials and processes between the first batch of offline steel materials and the target steel materials, and using their scanning electron microscope images as coordinate templates, the coordinates of the spots on the surface of the target steel materials can be obtained conveniently and quickly in the coordinate system without the need to observe the target steel materials with a scanning electron microscope.

[0101] In some embodiments, when transforming the coordinates of the second aluminum nitride spot image based on the third aluminum nitride spot image, if only spot features are considered for coordinate matching, the matching accuracy may be insufficient due to uneven spot distribution or local image defects, making it difficult to achieve accurate coordinate alignment. Therefore, transforming the second aluminum nitride spot image into a coordinate system based on the third aluminum nitride spot image may include:

[0102] Based on the grain boundary network of the steel, the grain boundary contrast feature points are enhanced and extracted in the third and second aluminum nitride spot images, respectively.

[0103] Based on the matching relationship between the grain boundary contrast feature points of the third and second aluminum nitride spot images, the second aluminum nitride spot image is transformed into a coordinate system.

[0104] In this embodiment of the invention, by utilizing the stability of the steel grain boundary network, the grain boundary contrast feature points of the third aluminum nitride spot image and the second aluminum nitride spot image are matched to achieve accurate alignment of the second aluminum nitride spot image with the coordinate system, thereby improving the coordinate transformation accuracy.

[0105] In some embodiments, if the same penalty weight is applied to regions with spots and regions without spots in the baseline blob map during defect recognition model training, the model may not pay enough attention to defect recognition in key areas with spots, potentially leading to low accuracy in key area recognition. Therefore, during the training of the defect recognition model, the penalty weight for incorrect defect recognition in the baseline blob map's marked areas can be set to greater than 1 in the model's loss function.

[0106] In this embodiment of the invention, setting the penalty weight for defect identification errors in key areas of the baseline blob map in the model loss function to be greater than 1 can enhance the model's learning focus on the key area and significantly improve the defect identification accuracy of the blob area.

[0107] In the input data of the defect recognition model, the baseline spot map and the online spot map are aligned by coordinates to achieve pixel-level correspondence. The differences between the baseline spot map and the online spot map can reflect the surface defects of the steel, strengthening the correlation judgment of spot distribution. To meet adaptation requirements, the loss function of the defect recognition model applies higher penalty weights to the identification errors in key areas, ensuring that the model's learning direction meets the detection requirements.

[0108] In this embodiment of the invention, stable and distinguishable aluminum nitride spots are generated in situ on the steel surface through a pulsed nitriding process. Optical interference signals are separated by an image acquisition scheme using multiple light sources and polarizers. Using the scanning electron microscope (SEM) images of the first batch of offline steel samples taken that day as coordinate templates, a unified coordinate system can be quickly achieved without requiring individual SEM inspection of each target steel sample. The accuracy of coordinate transformation is improved by matching grain boundary feature points in the steel, and the identification of defects in key areas is enhanced by adjusting the weights of the model loss function. This achieves accurate online identification of steel surface defects without compromising steel performance. This embodiment of the invention is adapted to the real-time detection requirements of high-speed production lines for electrical steel, providing efficient and reliable technical support for quality control during the production process.

[0109] This invention provides an online detection method for surface defects in steel, detailed below:

[0110] (1) Obtain the baseline spot map.

[0111] Defect-free steel samples were selected and subjected to pulse nitriding marking followed by polishing. A light etching process was then performed using 4% nitric acid alcohol. Utilizing the difference in corrosion resistance between AlN and the steel matrix, the AlN spots appeared raised under a scanning electron microscope (SEM). The spot images of the defect-free steel samples were observed using SEM at 5000x magnification. Spots were extracted based on a preset grayscale threshold, and their location coordinates and diameters were determined to obtain a baseline spot map.

[0112] The process of extracting the spot coordinates is as follows: with the top left corner of the SEM image as the origin (0,0) and pixels as the unit length, set the SEM image coordinate system and record the coordinates of each AlN spot in the SEM image coordinate system.

[0113] (2) Perform pulse nitriding marking on the annealed steel.

[0114] Approximately 5 meters behind the annealing furnace outlet, utilizing the residual temperature of the target steel (820±10℃) after annealing, ammonia gas with a purity greater than 99.99% is controlled by a solenoid valve to... A pulse jet is applied to the target steel at a flow rate of 2-3 seconds. By controlling the temperature, time, and flow rate parameters, a diameter of [missing information] is generated on the surface of the target steel. The AlN spots have a distribution density of [missing information]. about.

[0115] The spot growth process is as follows: The high-temperature steel surface acts as a catalyst, promoting the decomposition of ammonia gas and producing highly active nitrogen atoms [N]. These active nitrogen atoms [N] are adsorbed onto the steel surface and rapidly combine with Al elements in the steel through short-range surface diffusion, forming AlN clusters in situ on the steel surface. Because the ammonia gas injection lasts only 2-3 seconds, the active nitrogen atoms [N] have not yet diffused deep into the steel; therefore, the formed AlN clusters exist only on the steel surface, with a diameter of approximately [missing information]. It appears as raised spots.

[0116] After pulse nitriding marking, the treated target steel continues to cool using its own residual heat and air cooling, allowing the AlN spots on the steel surface to adhere firmly to the substrate. Ammonia pulse injection is performed in a sealed chamber filled with inert gas to prevent explosions and ensure process stability.

[0117] (3) Obtain an online spot map of the target steel.

[0118] The first batch of offline steel samples were pulsed nitriding and then polished. They were then lightly etched with 4% nitric acid alcohol. The spot images of the first batch of offline steel samples were observed using SEM at 5000x magnification, and the position coordinates and diameter of the spots were collected according to the SEM image coordinate system.

[0119] The target steel surface is illuminated by a coaxial white light source with a hue angle of 20-40°, and the surface is scanned using a linear array camera system equipped with polarizers. Coaxial light can highlight the edges and contours of the raised AlN spots on the steel surface, avoid shadow interference, and enhance the contrast between light and dark at the edges of the AlN spots.

[0120] Whenever the target steel moves a certain distance, the control line array camera system simultaneously acquires images of the target steel surface using blue light and near-infrared light. The acquired blue light image and near-infrared light image are differentially calculated to remove oil film interference fringes and retain aluminum nitride spot signals, thus obtaining an aluminum nitride spot image of the target steel.

[0121] For the same type of steel produced under the same process conditions, the grain boundary network of the steel grains remains stable. Therefore, by using image processing algorithms, the grain boundary contrast feature points in the aluminum nitride spot image of the target steel and the SEM image of the first batch of offline steel on the same day are enhanced and extracted respectively. Based on the matching relationship of the grain boundary contrast feature points in these two images, the coordinates of the aluminum nitride spot image of the target steel are transformed into the SEM image coordinate system to determine the position coordinates and diameter of the aluminum nitride spots of the target steel, thus obtaining the online spot map of the target steel.

[0122] Instead of directly matching easily changing AlN spots, the image processing algorithm matches stable and unchanging grain boundary contrast feature points. By aligning the two images, the position of the AlN spots attached to or on the grain boundary within the SEM image coordinate system is determined, thus achieving accurate coordinate transformation. Edge enhancement processing is applied to both the SEM and optical images to highlight shared line features, suppress modal differences, and make the grain boundary contours clearer and more prominent.

[0123] (4) Surface defect identification based on spot map.

[0124] The baseline speckle map is used as prior knowledge input into the defect recognition model to be trained. Based on the defective steel speckle map with defect labels, the defect recognition model is trained to obtain a pre-trained defect recognition model. In the loss function of the model training, a defect recognition error penalty weight of 1.3 is applied to the regions in the baseline speckle map that are marked with AlN speckles, so that the model can accurately identify phenomena such as speckle breakage and missing spots caused by real defects such as scratches and indentations.

[0125] The online spot map of the target steel is input into the trained defect recognition model to obtain the defect classification result of the target steel.

[0126] Traditional defect detection models need to learn the features of defects from massive amounts of data. However, in industrial scenarios, steel defect samples are relatively scarce. Moreover, surface texture changes caused by rolling marks, oil film, uneven lighting, camera shake, dust obstruction, etc., are harmless surface appearance differences that do not affect product function and service life. Traditional defect detection models are trained only on steel surface images, which can easily lead to model misjudgment.

[0127] The baseline spot map identifies the location and size of AlN spots in the absence of defects. Therefore, the disappearance or deformation of these AlN spots may be caused by real defects in the target steel that affect performance. During training, the model will be guided by the improved loss function to focus on the changes in these key areas.

[0128] In this embodiment of the invention, pulsed nitriding is performed using the residual heat of steel annealing to generate firmly raised AlN spots in situ on the steel surface. This requires no additional processing equipment and does not damage the steel's properties, providing stable and identifiable physical markers for defect detection. An imaging scheme combining coaxial white light and a polarizer is used to simultaneously acquire images and perform differential calculations to remove interference. By matching spot coordinates with grain boundary contrast feature points, accurate coordinate system transformation is achieved. This allows for the acquisition of an online spot map in the SEM coordinate system without requiring SEM inspection of the target steel. Using the baseline spot map as prior knowledge, the ability to identify key areas is enhanced by adjusting the penalty weights of the model's loss function, reducing misjudgments caused by harmless appearance differences. This enables accurate online identification of steel surface defects, meeting the real-time quality control requirements of high-speed steel production lines.

[0129] See Figure 2 This invention provides an online steel surface defect detection device 2, comprising:

[0130] The reference map acquisition module 21 is used to acquire a preset reference spot map; wherein, the reference spot map is obtained by establishing a coordinate system based on the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope;

[0131] The surface spot marking module 22 is used to generate raised aluminum nitride spots on the surface of the target steel by using the residual heat of the target steel after it leaves the annealing furnace through a pulse nitriding process.

[0132] The online map generation module 23 is used to acquire images of the second aluminum nitride spots on the surface of the target steel, convert the images of the second aluminum nitride spots to a coordinate system, and generate an online spot map.

[0133] The defect detection module 24 is used to use the benchmark spot map as prior knowledge for the pre-trained defect recognition model, input the online spot map into the defect recognition model, and identify defects in the target steel.

[0134] In one possible implementation, the baseline map acquisition module 21 is used to establish a coordinate system with the upper left corner of the first aluminum nitride spot image as the origin and pixels as the unit length.

[0135] Based on a preset grayscale threshold, the position coordinates and size parameters of each aluminum nitride spot in the first aluminum nitride spot image are extracted to obtain a preset baseline spot map.

[0136] In one possible implementation, the surface spot marking module 22 is used to control the pulse injection of high-purity ammonia gas onto the target steel via a solenoid valve to generate active nitrogen atoms;

[0137] By combining active nitrogen atoms with aluminum in the target steel, aluminum nitride spots are generated in situ on the surface of the target steel, forming raised spots that are firmly attached to the target steel matrix.

[0138] In one possible implementation, the online map generation module 23 is used to illuminate the surface of the target steel material with a coaxial white light source and to scan the surface of the target steel material with a line scan camera system.

[0139] Whenever the target steel moves a certain distance, the control line array camera system simultaneously acquires images of the target steel surface using blue light and near-infrared light. The blue light image and the near-infrared light image are then differentially calculated, and the aluminum nitride spot signal is retained to obtain the second aluminum nitride spot image.

[0140] In one possible implementation, the online map generation module 23 is also used to acquire the third aluminum nitride spot image of the first batch of offline steel surfaces on the same day in the coordinate system under a scanning electron microscope;

[0141] Based on the third aluminum nitride spot image, transform the second aluminum nitride spot image to the coordinate system;

[0142] The position coordinates and size parameters of aluminum nitride spots in the second aluminum nitride spot image under the detection coordinate system are used to obtain an online spot map.

[0143] In one possible implementation, the online map generation module 23 is further configured to enhance and extract grain boundary contrast feature points in the third aluminum nitride spot image and the second aluminum nitride spot image, respectively, based on the grain boundary network of the steel.

[0144] Based on the matching relationship between the grain boundary contrast feature points of the third and second aluminum nitride spot images, the second aluminum nitride spot image is transformed into a coordinate system.

[0145] In this embodiment of the invention, a standardized reference is established through a baseline map acquisition module, and a surface spot marking module generates firmly raised aluminum nitride spots without damaging the steel substrate, significantly reducing inspection costs. The online map generation module employs a coordinate transformation method that combines multi-source acquisition with grain boundary feature point matching, effectively suppressing optical interference and ensuring precise alignment between the online spot image and the baseline, providing reliable data support for defect identification. The device achieves real-time online detection of steel surface defects through multi-module collaboration, enhancing the targeting of defect identification based on prior knowledge of the baseline spots, and enabling online inspection on high-speed production lines.

[0146] See Figure 3 The diagram shows a schematic of the online steel surface defect detection system 3 provided in an embodiment of the present invention, which is described in detail below:

[0147] like Figure 3As shown, the online steel surface defect detection system 3 of this embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32. When the processor 30 executes the computer program 32, it implements the steps in the above-described method embodiments. Alternatively, when the processor 30 executes the computer program 32, it implements the functions of each module in the above-described device embodiments.

[0148] For example, the computer program 32 can be divided into one or more modules / units, which are stored in the memory 31 and executed by the processor 30 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 32 in the online steel surface defect detection system 3.

[0149] The online steel surface defect detection system 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that... Figure 3 This is merely an example of the online steel surface defect detection system 3 and does not constitute a limitation on the online steel surface defect detection system 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, the online steel surface defect detection system 3 may also include input / output devices, network access devices, buses, etc.

[0150] The processor 30 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0151] The memory 31 can be an internal storage unit of the online steel surface defect detection system 3, such as a hard disk or RAM. The memory 31 can also be an external storage device of the online steel surface defect detection system 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the online steel surface defect detection system 3. Furthermore, the memory 31 can include both internal and external storage units of the online steel surface defect detection system 3. The memory 31 is used to store the computer program 32 and other programs and data required by the online steel surface defect detection system 3. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0152] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.

[0153] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0154] This invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the methods described in the above-described method embodiments.

[0155] Computer programs include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0156] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0157] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. An online detection method for surface defects in steel, characterized in that, include: A preset reference spot map is obtained; wherein, the reference spot map is obtained by establishing a coordinate system based on the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope; After the target steel leaves the annealing furnace, the residual heat of the target steel is used to generate raised aluminum nitride spots on the surface of the target steel through a pulse nitriding process. Acquire images of second aluminum nitride spots on the surface of the target steel, transform the second aluminum nitride spot images into the coordinate system, and generate an online spot map; The baseline spot map is used as prior knowledge for the pre-trained defect recognition model. The online spot map is then input into the defect recognition model to identify defects in the target steel. The acquisition of the second aluminum nitride spot image on the surface of the target steel includes: The surface of the target steel is illuminated by a coaxial white light source, and the surface of the target steel is scanned by a line scan camera system; Whenever the target steel moves a certain distance, the linear array camera system is controlled to simultaneously acquire images of the target steel surface using blue light and near-infrared light. The blue light image and the near-infrared light image are differentially calculated, and the aluminum nitride spot signal is retained to obtain the second aluminum nitride spot image. The step of converting the second aluminum nitride spot image to the coordinate system and generating an online spot map includes: Obtain an image of the third aluminum nitride spot on the surface of the first batch of offline steel materials on the same day in the coordinate system using a scanning electron microscope; Based on the third aluminum nitride spot image, the second aluminum nitride spot image is transformed into the coordinate system; The position coordinates and size parameters of aluminum nitride spots in the second aluminum nitride spot image under the coordinate system are detected to obtain the online spot map.

2. The online detection method for steel surface defects according to claim 1, characterized in that, The step of transforming the second aluminum nitride spot image to the coordinate system based on the third aluminum nitride spot image includes: Based on the grain boundary network of the steel, the grain boundary contrast feature points are enhanced and extracted in the third aluminum nitride spot image and the second aluminum nitride spot image, respectively. Based on the grain boundary contrast feature point matching relationship between the third aluminum nitride spot image and the second aluminum nitride spot image, the second aluminum nitride spot image is transformed into the coordinate system.

3. The online detection method for steel surface defects according to claim 1 or 2, characterized in that, The process of obtaining a preset baseline spot map includes: The coordinate system is established with the top left corner of the first aluminum nitride spot image as the origin and pixels as the unit length. Based on a preset grayscale threshold, the position coordinates and size parameters of each aluminum nitride spot in the first aluminum nitride spot image are extracted to obtain the preset reference spot map.

4. The online detection method for steel surface defects according to claim 1, characterized in that, During the pre-training process of the defect recognition model, the penalty weight for defect recognition errors in the region of the baseline blob map that identifies AlN blobs is set to be greater than 1 in the model loss function.

5. The online detection method for steel surface defects according to claim 1, characterized in that, The process of generating raised aluminum nitride spots on the surface of the target steel using pulsed nitriding includes: High-purity ammonia gas is pulsed and injected onto the target steel using a solenoid valve to generate active nitrogen atoms. The active nitrogen atoms combine with the aluminum element in the target steel to generate raised aluminum nitride spots on the surface of the target steel that are firmly attached to the substrate of the target steel.

6. An online detection device for steel surface defects, characterized in that, include: The benchmark map acquisition module is used to acquire a preset benchmark spot map; wherein, the benchmark spot map is obtained by establishing a coordinate system from the first aluminum nitride spot image on the surface of a defect-free steel sample under a scanning electron microscope; The surface spot marking module is used to generate raised aluminum nitride spots on the surface of the target steel by means of the residual heat of the target steel after it leaves the annealing furnace through a pulse nitriding process. An online map generation module is used to acquire a second aluminum nitride spot image of the target steel surface, convert the second aluminum nitride spot image to the coordinate system, and generate an online spot map; the target steel surface is illuminated by a coaxial white light source, and the surface of the target steel is scanned by a line scan camera system; whenever the target steel moves a certain distance, the line scan camera system is controlled to simultaneously acquire images of the target steel surface using blue light and near-infrared light, the blue light image and near-infrared light image are differentially calculated, the aluminum nitride spot signal is retained, and the second aluminum nitride spot image is obtained; a third aluminum nitride spot image of the first batch of offline steel surfaces of the day is acquired in the coordinate system under a scanning electron microscope; based on the third aluminum nitride spot image, the second aluminum nitride spot image is converted to the coordinate system; the position coordinates and size parameters of the aluminum nitride spots in the second aluminum nitride spot image in the coordinate system are detected to obtain the online spot map; The defect detection module is used to use the benchmark spot map as prior knowledge for the pre-trained defect recognition model, input the online spot map into the defect recognition model, and identify the defects of the target steel.

7. An online detection system for steel surface defects, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images