A strip steel production line and a strip steel micro-defect detection method

Abstract

The present application relates to the technical field of strip steel production detection, and particularly relates to a strip steel production line and a strip steel micro-defect detection method. The present application synchronously deploys detection units containing different modal sensors at a first node, a second node and a third node, and constructs an online detection network, thereby solving the technical problem of insufficient information dimension of traditional single visual detection and inability to comprehensively represent defect properties. A hyperspectral imager can analyze surface chemical composition (such as oxidation), a laser profilometer can obtain three-dimensional topography and zinc layer thickness, an infrared thermal imager can reflect the thermal field of the solidification process, and an eddy current instrument can detect subcutaneous defects. This multi-modal perception fusion technology first realizes the synchronous and comprehensive perception of the chemical composition, three-dimensional topography, thermodynamic state and internal quality of the surface defects of the strip steel, so that the accuracy of defect identification and classification realizes a qualitative leap.

Description

Technical Field

[0001] This invention relates to the field of strip steel production and inspection technology, and in particular to a strip steel production line and a method for detecting micro-defects in strip steel. Background Technology

[0002] Currently, galvanized steel strip is widely used in industries such as construction, home appliances, and automobiles. On a continuous hot-dip galvanizing production line, the steel strip undergoes a series of complex processes including uncoiling, straightening, welding, looping buffering, annealing, galvanizing, post-treatment, and coiling. During production, due to fluctuations in raw materials, drift in process parameters, or changes in equipment condition, the surface of the steel strip is highly susceptible to micron-level defects such as scratches, pits, oxide spots, zinc flow lines, and zinc slag particles. These micro-defects not only affect the product's appearance but also reduce its corrosion resistance and coating performance, becoming a key bottleneck restricting product quality improvement.

[0003] Currently, online inspection of strip steel surface quality in the industry mainly relies on a single vision inspection system installed at the end of the production line (such as before coiling). This system typically uses a high-resolution line scan camera to scan the strip steel surface and identifies defects through image processing algorithms. However, this traditional inspection method has significant technical limitations. Post-production inspection cannot intervene in real time: defects are only detected at the end of the production line, by which time the strip steel has already completed all galvanizing and post-processing. The inspection system can only record and sort, unable to intervene and control in real time during the generation or expansion of defects, resulting in high scrap rates and quality losses. The inspection information is limited, making it difficult to trace the root cause of defects: a single vision inspection can only obtain two-dimensional morphological information of defects, unable to determine the nature of the defects (such as whether they are oxides or subcutaneous defects), depth information, or whether the defect occurred during annealing, galvanizing, or cooling. The lack of cross-process defect evolution information makes it difficult for quality engineers to accurately trace the root cause of defects, resulting in low efficiency. Inspection accuracy is severely affected by the environment: the internal environment of the galvanizing line is harsh, with interference from high temperatures, zinc ash, oil fumes, etc. Traditional testing equipment suffers from insufficient protection, making lenses susceptible to contamination and resulting in high false alarm and false negative rates, as well as poor testing stability. The system also suffers from low intelligence and a lack of self-optimization capabilities: existing testing systems are relatively independent of the production line control system, forming information silos. Testing results cannot be directly and intelligently converted into process parameter adjustment instructions. Optimization of process parameters heavily relies on engineer experience, lacking data-driven self-learning and self-optimization mechanisms, making it difficult to adapt to changes in operating conditions such as steel grade or zinc ingot changes.

[0004] Therefore, there is an urgent need for an intelligent strip steel production line and testing method that can achieve online defect diagnosis, root cause tracing and real-time suppression, in order to break through the bottleneck of existing technologies. Summary of the Invention

[0005] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0006] A strip steel production line includes the following components arranged sequentially from left to right along the production line:

[0007] An uncoiling device is used to supply raw material strip steel;

[0008] A straightening device is used to level the raw material strip steel;

[0009] The shearing device is used to shear the head and tail of the leveled strip steel.

[0010] Welding equipment is used to connect the sheared strip to the tail of the previous coil of strip.

[0011] The inlet looper is used to store the welded strip steel and provide a buffer for subsequent processes;

[0012] A galvanizing apparatus for continuously hot-dip galvanizing the strip steel;

[0013] Finished product collection device, used to collect galvanized steel strips;

[0014] And a detection device, which is set on one side of the galvanizing treatment device, is used to perform online micro-defect detection on the surface of the strip steel at different treatment stages.

[0015] A method for detecting micro-defects in strip steel from a strip steel production line includes the following steps:

[0016] S1: Inside the galvanizing unit, detection units are deployed simultaneously at three process nodes to form an online detection network.

[0017] At the first node, namely in the transition channel between the outlet of the annealing furnace and the inlet of the galvanizing pot, a first detection unit is deployed, the first detection unit including at least a high-resolution linear CCD camera and a hyperspectral imager.

[0018] At the second node, that is, in the area after the air knife device at the galvanizing pot outlet and before the post-galvanizing cooling device, a second detection unit is deployed. The second detection unit includes at least a laser three-dimensional profilometer and an infrared thermal imager.

[0019] At the third node, namely the final quality inspection area before the export looper, a third inspection unit is deployed, which includes at least a surface gloss meter and an eddy current meter.

[0020] S2: Real-time synchronous acquisition of multimodal sensing data from three nodes, and assignment of a unique spatiotemporal code to the same strip steel passing through each node;

[0021] Defect features from different nodes are extracted and fused to construct a defect evolution map of the strip steel across process stages; wherein, the features include the oxide distribution features on the substrate surface from the first node, the three-dimensional thickness distribution and solidification thermal field features of the zinc layer from the second node, and the surface finish and subcutaneous defect features of the finished product from the third node.

[0022] S3: Associate the cross-process stage defect evolution map with the core process parameter set of the production line in the same time period. The process parameter set includes at least the temperature of each section of the annealing furnace, strip tension, zinc liquid temperature, air knife pressure and distance, and cooling rate.

[0023] By using an online incremental learning model, a dynamic mapping relationship is established between process parameter disturbances, multimodal feature anomalies, and final defect types. When a defect is detected, one or more of the most likely root cause process parameter deviations that led to the defect can be diagnosed in real time.

[0024] S4: Based on the root cause process parameter deviations diagnosed in step S3, generate two levels of control instructions:

[0025] Feedforward control: For diagnosed process deviations with fixed propagation delays, the setting parameters of the affected process section are adjusted in advance before the strip reaches the downstream affected process section.

[0026] Feedback control: For any real-time process deviations diagnosed, the process parameters at the current defect point are adjusted immediately;

[0027] S5: Record complete defect diagnosis records, control actions, and final quality ratings for each coil of finished strip steel; periodically perform cluster analysis on historical data to identify sensitive intervals of process parameters that cause the recurrence of similar defects, and automatically optimize the process parameter settings within these sensitive intervals to form a continuously evolving process knowledge base.

[0028] Furthermore, in step S1, the first node and the second node are both located between two adjacent processing layers inside the galvanizing device, and the detection unit is installed in a sealed protective chamber with a self-cleaning window, using helium positive pressure to prevent dust and zinc ash contamination.

[0029] Furthermore, the online incremental learning model described in step S3 adopts an ensemble learning algorithm based on concept drift detection. When the production line changes the steel grade or zinc ingot brand, the model can automatically identify the change in data distribution and start a new sub-model to learn without forgetting old knowledge.

[0030] Furthermore, in step S4, the triggering timing of the feedforward control is determined by calculating the precise transmission time of the strip from the root process deviation point to the downstream control point. This transmission time is dynamically calculated from the strip linear speed, looper inventory, and interlayer path length.

[0031] Furthermore, it also includes step S6: virtual tagging and hierarchical volume division:

[0032] For defective strips that cannot be completely eliminated through online control, the system performs virtual quality classification based on their defect maps and generates classification instructions.

[0033] During the export slitting process, the shearing device automatically cuts strips of different quality grades and guides them to different coilers for winding, thereby achieving quality homogenization within the same steel coil.

[0034] Furthermore, the construction of the defect evolution map across process stages described in step S2 is specifically achieved through the following steps:

[0035] S2.1: Using the aforementioned spatiotemporal coding, the multimodal sensing data from the three nodes are synchronized and aligned in the time and space dimensions to ensure that data points corresponding to the same physical location on the strip are associated.

[0036] S2.2: For each associated data point, extract the time-series change vector of its defect features along the process direction. The time-series change vector records the feature evolution of a specific location from the substrate state, through galvanizing and solidification, to becoming a finished product.

[0037] S2.3: Atlas generation and visualization: Arrange the time-series change vectors along the strip length direction to generate a two-dimensional defect evolution atlas with length position as the horizontal axis, process stage as the vertical axis, and feature vector values ​​as elements, and mark the initiation, propagation and morphological evolution trajectory of defects in the atlas.

[0038] Furthermore, the generation of two-level control instructions in step S4 further includes an arbitration and fusion step for conflicting control instructions:

[0039] S4.3: When multiple feedforward or feedback control commands are generated simultaneously for the same strip steel or the same process equipment, determine whether there is a conflict between these commands in the direction of control target or parameter adjustment.

[0040] S4.4: If a conflict exists, a decision is made based on a predefined arbitration rule base. The arbitration rule base sets the priority of different control objectives based on process principles and outputs a fused, conflict-free final control instruction sequence.

[0041] Furthermore, the automatic optimization of process parameter settings mentioned in step S5 is achieved through the following steps:

[0042] S5.1: Based on the cluster analysis results of historical data, identify the fluctuation range of process parameters that cause the same type of defect and define it as the sensitive interval;

[0043] S5.2: Within the sensitive range, a sequence optimization algorithm is used to systematically adjust the combination of process parameters in the form of simulation or small-batch trial production;

[0044] S5.3: Use the strip quality rating produced after each parameter adjustment as a feedback signal for the optimization algorithm;

[0045] S5.4: When a parameter combination that can reliably produce higher quality rated products is found, update the combination to the process knowledge base as the new standard setting value under the production conditions.

[0046] Furthermore, it also includes a detection network self-calibration step S0 performed prior to step S1:

[0047] S0.1: At each process node of the production line, a standard template with known standard micro-defect characteristics is introduced;

[0048] S0.2: During breaks in normal production line operation or planned shutdowns, drive the standard template through the field of view of each detection unit to collect its multimodal sensing data;

[0049] S0.3: Compare the collected data with the known features of the standard sample, calculate the measurement drift error of each sensing channel, and generate the corresponding compensation coefficients for real-time correction of subsequent production and testing data.

[0050] The advantages of this invention are:

[0051] 1. This invention solves the technical problem of insufficient information dimensions and inability to comprehensively characterize defect properties in traditional single-vision inspection by simultaneously deploying detection units containing different modal sensors at the first node (after annealing), the second node (after galvanizing), and the third node (final quality inspection), and constructing an online detection network. A hyperspectral imager can analyze surface chemical composition (such as oxidation), a laser profilometer can acquire three-dimensional morphology and zinc layer thickness, an infrared thermal imager can reflect the thermal field during solidification, and an eddy current analyzer can detect subcutaneous defects. This multimodal perception fusion technology achieves, for the first time, simultaneous and comprehensive perception of the chemical composition, three-dimensional morphology, thermodynamic state, and internal quality of surface defects in strip steel, resulting in a qualitative leap in the accuracy of defect identification and classification.

[0052] 2. This invention solves the technical challenge of the invisible generation and development process of defects and the difficulty in tracing their root causes by assigning spatiotemporal codes to the same strip of steel and constructing a defect evolution map across process stages. This method links the characteristic changes of the same defect in the annealing, galvanizing, and finished product stages, forming a clear defect evolution trajectory. This upgrades quality analysis from static point detection to dynamic line tracing, enabling a clear determination of whether the defect is due to oxidation during annealing, zinc dross formation during galvanizing, or stress cracking during cooling, providing unprecedentedly clear targets for precise process intervention.

[0053] 3. This invention addresses the technical problems of traditional quality analysis, such as reliance on human experience, slow response, and inability to adapt to changes in operating conditions, by associating defect evolution maps with real-time process parameters and establishing a dynamic mapping relationship using an online incremental learning model. This model can automatically learn the nonlinear relationship between defects and complex process parameters. When a defect is detected, it can diagnose the most likely root cause of the process parameter deviation (such as "the temperature in the Xth stage of the annealing furnace is 2°C lower" or "air knife pressure fluctuation") in real time, achieving intelligent and real-time diagnosis of quality problems.

[0054] 4. This invention solves the technical problem of traditional control methods being passive, lagging, and unable to prevent defect propagation by generating two-level collaborative control commands (feedforward and feedback) based on diagnostic results. Feedback control can quickly correct immediate deviations; feedforward control can adjust the parameters of the target process section in advance, based on the precise time delay of strip transport, before defects propagate downstream, thereby proactively cutting off the defect propagation chain and suppressing defects at the nascent or escalating stage. This predictive intervention mode reduces the generation of defective products at the source.

[0055] 5. This invention solves the technical problem of production process knowledge relying on personal experience, being difficult to accumulate, and being continuously optimized by recording historical data, performing cluster analysis, and automatically optimizing the sensitive range of process parameters. The system can automatically identify the sensitive range of process parameters that lead to defect recurrence and use optimization algorithms to find a better process window. This enables the process knowledge base of the entire production system to continuously and autonomously evolve, resulting in increasingly stable product quality and achieving a leap from automation to intelligence.

[0056] 6. This invention solves the technical problem that traditional inspection equipment can only perform end-of-line inspections and cannot achieve process sensing by integrating a detection device on one side of the galvanizing unit. This layout allows the detection system to penetrate deep into the core process areas of the production line, providing a physical basis for online, real-time acquisition of surface condition information of strip steel during key processing stages (such as after annealing and after galvanizing), and creating the prerequisites for achieving process quality control.

[0057] In summary, this invention deeply integrates sensing technology, data fusion, machine learning algorithms, and production line control logic to construct a closed-loop quality control method with comprehensive sensing, intelligent diagnosis, forward-looking control, and continuous evolution capabilities. This fundamentally changes the traditional strip steel quality inspection and control model, significantly improving product quality, production efficiency, and intelligence level. Attached Figure Description

[0058] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0059] Figure 1 This is a schematic diagram of the layout structure of the strip steel production line in this invention.

[0060] In the diagram: 1. Uncoiling device; 2. Straightening device; 3. Shearing device; 4. Welding device; 5. Inlet piston device; 6. Galvanizing treatment device; 7. Finished product collection device; 8. Detection device. Detailed Implementation

[0061] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1:

[0063] Figure 1 This is a schematic diagram of the layout structure of the strip steel production line in this invention. This embodiment provides a solution as shown in the attached diagram. Figure 1 The image shows a continuous hot-dip galvanizing production line for strip steel. This production line is arranged according to the strip steel's running direction (e.g., ...). Figure 1 As indicated by the middle arrow, from left to right, the following are mechanically connected and work in concert: uncoiling device 1, straightening device 2, shearing device 3, welding device 4, inlet looper device 5, galvanizing treatment device 6, finished product collection device 7, and a detection device 8 deeply integrated into the production line. All devices are uniformly scheduled and controlled through a central control system (not shown in the figure, such as an industrial PC or PLC system).

[0064] This embodiment employs a dual uncoiler configuration, including a first uncoiler and a second uncoiler, arranged side-by-side. This device is used to carry and unwind raw cold-rolled steel coils. When the first uncoiler is about to run out of coils, the second uncoiler prepares the next coil in advance, achieving continuous and uninterrupted production through subsequent welding. A tension control system is installed during the uncoiling process to ensure that the strip enters the production line with stable tension.

[0065] The straightening device 2 is located after the uncoiling device 1 and is a multi-roll straightener. Its function is to eliminate wavy, warped, and other strip shape defects generated during the rolling or coiling process of the raw strip, so that the strip remains flat before entering the continuous processing section. The reduction amount of the straightening rolls can be set automatically or manually through the control system according to the thickness and material of the strip.

[0066] The shearing device 3 is located between the straightening device 2 and the welding device 4, and employs a hydraulic or mechanical flying shear. Its main functions are twofold: first, to cut off the tail end of the previous coil of strip (which may contain uneven thickness or poor surface quality) and the head end of the next coil (which may have uncoiling damage); second, to prepare a flat, vertical strip end face for welding. The cut surface of the sheared strip should be neat to facilitate high-quality subsequent welding.

[0067] In this embodiment, welding device 4 preferably uses a narrow lap resistance welding machine, but a laser welding machine can also be used. Its core function is to firmly connect the tail end of the previous coil of strip after shearing to the head end of the next coil of strip. During the welding process, it is necessary to ensure that the weld strength is not lower than that of the base material, the weld thickness is uniform, and that it will not break in subsequent processes (especially when passing through tension rollers and loopers). After welding, a scraper or grinding equipment is usually used for preliminary treatment of weld protrusions.

[0068] Inlet looper device 5: This embodiment uses a vertical (tower-type) looper, which consists of multiple steering rollers and a movable looper carriage, capable of storing hundreds of meters of strip steel. Its core function is "buffering." When the inlet section (uncoiling, straightening, shearing, welding) needs to be paused or slowed down due to coil changing, welding, or other operations, the inlet looper device 5 can continuously release the stored strip steel to supply the subsequent high-speed galvanizing treatment device 6, thereby ensuring the continuous and stable operation of the core process section (annealing, galvanizing), unaffected by the discrete operations of the inlet section. The looper's filling degree and release speed are monitored in real time by sensors and precisely adjusted by the control system.

[0069] Galvanizing Unit 6: This galvanizing unit 6 integrates the following main sub-units according to the process sequence: Cleaning Section: The strip steel first enters this section and undergoes multiple processes such as alkaline spraying, mechanical brush washing, electrolytic cleaning, and hot water rinsing to thoroughly remove contaminants such as rolling oil, grease, and iron powder from its surface, providing an absolutely clean metal surface for annealing and galvanizing. Annealing Furnace Section: The cleaned strip steel enters a full-radiation tube annealing furnace. This furnace is typically divided into a preheating section, a heating section, a soaking section, and a cooling section. In an environment filled with a high-purity nitrogen-hydrogen protective atmosphere, the strip steel is heated to a recrystallization temperature above the Ac1 point (e.g., approximately 700-850°C for low-carbon steel) and held at this temperature for a certain period to eliminate work hardening after cold rolling, restore its plasticity and toughness, and obtain a microstructure conducive to galvanizing (such as the homogenization of ferrite grains). Subsequently, the strip steel is precisely cooled in a controlled cooling section to an entry temperature slightly higher than the zinc bath temperature (approximately 460-480°C). Galvanizing Section: The cooled strip steel is vertically immersed in a zinc pot containing molten zinc (the zinc temperature is generally maintained at approximately 460°C). The strip steel is turned by submerged rollers in the zinc liquid, completing the zinc layer adhesion. Subsequently, the strip steel is drawn out of the zinc liquid and vertically passed through an air knife device. The air knife sprays a high-speed, uniform stream of nitrogen (or air) onto both sides of the strip steel surface, precisely scraping away excess zinc liquid, thereby controlling the zinc layer thickness (e.g., 60-275 g / m³ per side) and its uniformity. This is crucial for obtaining good coating surface quality. Post-Galvanizing Cooling and Post-Treatment Section: The zinc layer on the strip steel surface after passing through the air knife is not yet fully solidified. It first enters the air cooling section, where controlled air cooling causes preliminary solidification. It then enters a water quenching tank for rapid cooling to optimize the mechanical properties and adhesion of the coating. The cooled strip steel enters the chemical treatment section, where chromate passivation or environmentally friendly chromium-free passivation treatment forms an extremely thin conversion film on the zinc layer surface, significantly improving its resistance to white rust corrosion and coating adhesion. Finally, the strip steel passes through a drying device to remove surface moisture.

[0070] The finished product collection device 7 is located at the very end of the production line and can be one or more coilers. Its function is to rewind the galvanized steel strip that has undergone all processing into finished steel coils that are tight inside, even outside, and have neat edges. The coiling tension needs to be precisely controlled to prevent damage to the strip surface or interlayer abrasion. In some configurations, a cross-cutting unit may also be included to cut the strip into single steel sheets of fixed length.

[0071] The detection device 8 is arranged on one side of the galvanizing treatment device 6, and its functional logic runs through the entire quality closed loop. The core of this embodiment lies in the method executed by the aforementioned detection device 8. The detection device 8 is a system composed of a hardware sensor network, a data acquisition and communication system, an edge computing unit, a central analysis server, and a control interface. This invention sets up three of the most representative process nodes inside the galvanizing treatment device 6, and simultaneously deploys functionally complementary detection units to form a three-dimensional online detection network. The first node and the first detection unit are located in the transition channel between the outlet of the annealing furnace and the inlet of the zinc pot. This position is a window for observing the pretreatment effect and the state of the substrate before it enters the pot. The strip steel has been annealed, the surface is clean and in a high-temperature activated state, but it has not yet come into contact with the zinc liquid. The first detection unit is deployed at this node. The core of this unit includes a high-resolution linear CCD camera, which uses a linear scanning camera with a resolution of not less than 4K, equipped with a high-brightness LED linear light source, and acquires bright field and dark field images of the strip steel surface at an extremely high scanning frequency (matching the highest linear speed of the strip steel). It is mainly used to detect macroscopic and microscopic morphological defects such as scratches, pits, and roll marks. The hyperspectral imager covers the visible to near-infrared band (e.g., 400-1000nm). Instead of acquiring ordinary color images, this device acquires the continuous spectral curve of each pixel. Since different substances (such as pure iron, various iron oxides FeO, Fe2O3, Fe3O4) have characteristic absorption or reflection peaks in specific bands, analyzing the spectral curves allows for non-contact, quantitative analysis of the chemical composition of the steel strip surface. This is particularly useful for detecting subtle oxidation phenomena (oxidation color, oxidation spots) that are difficult to detect with the naked eye, as well as the distribution of residual contaminants. In this invention, all optical components of the first detection unit are integrated and installed within a sealed protective chamber. The observation window of the protective chamber facing the steel strip is made of high-temperature resistant quartz glass and equipped with a compressed air self-cleaning system. This system continuously sprays clean gas onto the window surface to form a positive pressure air curtain, effectively isolating zinc ash, oil vapor, and dust, ensuring the window remains clean and image acquisition is stable. The second node and the second detection unit are located in the area after the air knife device at the zinc pot outlet and before the post-galvanizing cooling device (water quenching). At this location, the quality of the galvanizing process is observed. After the strip steel has completed galvanizing and air knife scraping, the zinc layer is in the initial stage of liquid-to-solid transition, providing the richest information on its surface morphology, thickness distribution, and solidification thermal behavior. The second detection unit is deployed at this node. This unit includes a laser 3D profilometer, employing a scanner based on the laser triangulation principle or confocal principle. It projects a line laser or array laser onto the strip steel surface and, by receiving the reflected light, can reconstruct the 3D morphology of the strip steel surface in real time with high precision, achieving measurement accuracy down to the micrometer level. It is mainly used to detect 3D characteristic defects such as zinc flow lines, zinc undulations, and zinc slag particles, and can non-contactly measure the local thickness distribution of the zinc layer to assess the uniformity of the air knife operation. A high-speed infrared thermal imager with a spectral response range in the mid-wave infrared (e.g., 3-5 μm) is also included.This equipment is used to capture the temperature field distribution of the zinc layer during the solidification process after the strip leaves the air knife. Uniform cooling creates a uniform temperature field, while uneven airflow, uneven substrate temperature, or zinc liquid flow problems can lead to hot spots or cold streaks in the temperature field. The third node and the third detection unit of this invention are located after the exit of the galvanizing treatment device 6, but before the final quality inspection area of ​​the exit looper (if any) or the finished product collection device 7 (coiler). The third detection unit is deployed at this node. This unit includes: a surface gloss / roughness meter: using the principle of laser scattering or confocal white light interference, to measure the gloss (GI value) and micro-roughness (Ra value) of the strip surface. The gloss of the finished galvanized sheet is an important commodity attribute and must meet customer requirements. This instrument can detect gloss differences caused by uneven passivation, minor scratches, etc. Eddy current detector: using a multi-frequency eddy current probe. Eddy currents can penetrate the surface zinc layer and sense changes in the electromagnetic properties of the substrate metal beneath it. Subsurface defects beneath the zinc layer, such as non-metallic inclusions, micro-delamination, or micro-peeling caused by poor coating adhesion, cannot be detected by traditional vision systems. However, these defects can cause changes in the phase and amplitude characteristics of eddy current signals, making them effectively detectable. The detection units at these three nodes are connected to a central analysis server and an edge computing gateway via high-speed industrial Ethernet (such as EtherCAT or Profinet) to achieve real-time synchronous data acquisition and uploading.

[0072] To achieve end-to-end tracking of the same strip steel from substrate to finished product, this invention introduces a precise spatiotemporal coding mechanism. When the strip steel enters the first detection unit (i.e., the annealing furnace exit), the system generates a unique ID (e.g., "Coil_20240617_001") for the starting point of the strip steel coil. Simultaneously, a high-precision encoder (typically installed on the drive roller of the production line) begins continuously measuring the running length of the strip steel with millimeter-level accuracy. The central system timestamps and assigns a corresponding precise length position label (Length_Pos) to each frame of data (image, spectrum, contour, temperature, etc.) acquired from the first detection unit. The timestamp and length position constitute the spatiotemporal coordinates of that data point. Since the strip steel runs at a basically constant speed, and the transmission path and delay from the first node to the second and third nodes (considering changes in looper inventory) can be accurately calculated using a physical model (the formula involves linear velocity V, path length L, looper height H, etc.), the central system can predict and automatically align and correlate the strip steel data detected at the second and third nodes, which belong to the same physical length position, with the data from the first node. For example, the strip steel passing through the first node at time T1 and position L1 will arrive at the second node at the predicted time T2 (T1+ΔT), and its detection data should be correlated. The central analysis server runs advanced data analysis software and performs the following steps: First, the CCD image is filtered, enhanced, and segmented to extract the geometric features (area, perimeter, aspect ratio), grayscale features, and texture features of defects. Then, for the hyperspectral data, principal component analysis (PCA) or spectral angle mapping (SAM) is used to extract the feature vectors that differ most from the standard "clean iron-based" spectrum, quantitatively characterizing the degree of oxidation or contaminant index. For 3D contour data, statistical features of surface height distribution (such as root mean square deviation Sq, skewness Ssk) and the depth and volume features of defects are extracted. For infrared thermographic data, statistical features of the temperature field (average temperature, temperature gradient, area of ​​high-temperature regions) are extracted. For eddy current data, trajectory features of the impedance plane (such as amplitude and phase angle) are extracted. For gloss data, the distribution features of GI values ​​are extracted. The system arranges all features from three nodes corresponding to the same strip (e.g., a 10-meter strip from length L_start to L_end) into a multidimensional feature matrix according to their spatiotemporal coordinates. The horizontal axis of the matrix is ​​the length position of the strip (L_start, L_start+ΔL,…,L_end). The vertical axis is the process stage (first stage: substrate after annealing; second stage: initial solidification after galvanizing; third stage: final product). The matrix elements are the multidimensional feature vectors extracted at that location and stage.Data visualization technology can project a multi-dimensional matrix into an intuitive and understandable defect evolution map across process stages. For example, the software interface can simultaneously display: a) an oxidation index thermal map at substrate location Lx; b) a three-dimensional height map and temperature distribution map of the same location Lx after galvanizing; and c) a gloss distribution and eddy current signal map of the same location Lx after finishing. With these three maps aligned vertically, abnormal defects such as high oxidation index, corresponding plating protrusions and high temperatures, and low gloss of the finished product can be tracked. The central system uses industrial communication protocols such as OPC UA to obtain a set of core process parameters in real time from the production line's distributed control system (DCS) for the same time period. These parameters include, but are not limited to: the set temperature and actual temperature curves of the annealing furnace preheating section, heating section, soaking section, and rapid cooling section; dew point and oxygen content in each section of the furnace; production line master speed and tension in each section; zinc liquid temperature and aluminum content; air knife pressure, blade lip distance, and angle; cooling fan speed and water quenching tank temperature, etc. These parameters are synchronized with the detection data stream in time-series format. Model building and training include: Initially, the system runs in supervised learning mode. Engineers manually label defect types (e.g., oxide spots, zinc slag particles, air knife streaks) when analyzing defect evolution maps, and judge their most likely root causes based on experience (e.g., low temperature in the soaking zone of the annealing furnace, blockage on one side of the air knife, abnormal zinc-aluminum content). Defect-process parameter deviation pairing data is used as the initial training set. The system employs an online ensemble learning model based on concept drift detection, for example, using a Hoeffding Adaptive Tree as the base learner, combined with the ADWIN (Adaptive Window) algorithm to detect changes in data distribution. This model can learn online; as production continues, new data (feature maps + process parameters + manually / automatically labeled root causes) is continuously input, and the model updates its internal parameters in real time, making the diagnosis increasingly accurate. When the online detection network detects significant defect features (e.g., excessive oxidation index, three-dimensional protrusions), the system immediately triggers the diagnostic process, and the model analyzes the dynamic relationship between process parameters and multimodal features within a time window before and after the defect occurs. The system ultimately outputs a diagnostic report, for example, detecting an oxide spot defect at length Lx with a confidence level of 92%. The diagnostic results are immediately translated into control actions, forming a quality closed loop. This is for immediate and quickly correctable process deviations. For example, the system diagnoses that fluctuations in the current air knife pressure are causing uneven zinc layer thickness. The control module immediately generates instructions to fine-tune the air knife pressure regulating valve, stabilizing the pressure back to the set value within milliseconds to seconds. This invention can achieve feedforward control for diagnosed process deviations with a fixed physical propagation delay. For example, the system diagnoses at the first node that the strip steel (assuming length Ly) about to enter the zinc pot has insufficient activity due to a low temperature in a certain zone of the annealing furnace, predicting that it will exhibit uneven zinc bloom or poor adhesion defects at the second node (after galvanizing). However, this strip steel Ly requires time Δt (propagation delay) to travel from the first node to the zinc pot.Traditional methods can only address defects after they appear at the second node (which is too late). However, the feedforward control module of this invention accurately calculates the transmission delay Δt = (path length + loop equivalent length) / current strip speed. At Δt before the strip Ly segment reaches the zinc pot, an instruction is sent to the zinc pot process control system in advance, for example, at [T+Δt], slightly increasing the zinc bath temperature by 2°C or adjusting the air knife distance corresponding to that strip segment. When the problematic substrate Ly arrives at the zinc pot, it faces pre-adjusted galvanizing process conditions, thus compensating for or mitigating the adverse effects of upstream processes. For example, for the same strip segment, it may be necessary to adjust both the annealing furnace temperature and the production line tension simultaneously. An arbitration rule base can be pre-set in the central controller, with rules based on process principles (e.g., temperature control has higher priority than speed fine-tuning, and ensuring tension stability takes precedence over local temperature compensation). The controller performs conflict detection and resolution on concurrent instructions, generating a safe, orderly, and executable final control instruction sequence, which is then sent to the production line DCS for execution. For defective strip sections that have occurred and cannot be completely eliminated through online control, the system does not simply reject the entire coil. Instead, based on the severity, type, and distribution density of its defect profile, and according to internal company or customer standards, it performs virtual quality grading (e.g., Grade A: no defects; Grade B: minor flaws allowed; Grade C: requires downgrading). The system marks the strip section (defined by length range) with a virtual quality grade label in the database. When the strip reaches the exit shearing device 3, the control system reads the virtual grading instruction. At the location where coiling is required (e.g., switching from Grade B to Grade C), the flying shear automatically cuts the strip. Multiple coilers (part of the finished product collection device 7) receive the instruction, with Grade A strip coiled to coiler number 1 and Grade B strip coiled to coiler number 2, thereby achieving quality homogenization within the same physical coil, maximizing product value, and reducing mixed-grade losses. The system continuously archives all production data for each coil (original process parameters, inspection characteristics, diagnostic records, control actions, final quality rating, customer feedback) to form a digital twin archive. The background data analysis module is activated periodically (e.g., weekly) to perform cluster analysis on historical data. It can automatically detect, for example, that when the temperature in zone 3 of the annealing furnace is within the range [T1, T2] and the dew point inside the furnace is higher than P, the probability of oxide spot defects increases significantly. This range of [T1, T2] and P is defined as the process parameter sensitive range. The system then calls a Bayesian optimization algorithm to explore a virtual process window within this sensitive range. It simulates the impact of different parameter combinations (temperature, dew point, and speed) on the final quality rating to find a robust process window that maximizes the expected quality rating. Once a validated better parameter combination is found, the system prompts engineers for review and can choose to update it as the new standard process formula under these production conditions, storing it in the process knowledge base. This cycle ensures that the entire production system's process settings have the ability to continuously and autonomously optimize and adapt, no longer relying entirely on the experience of experienced technicians.Taking the production of a galvanized steel coil as an example, the workflow of this invention is as follows: Production Start-up: The raw material coil is loaded, and the production line starts according to the standard process schedule. Each unit of the detection network completes self-calibration (through the standard template) and enters standby mode. Continuous Production and Synchronous Sensing: The strip steel passes through each process sequentially. At the first node, the system monitors the surface cleanliness and oxidation state of the substrate in real time; at the second node, it monitors the uniformity of the zinc layer and the solidification process in real time; at the third node, it conducts a comprehensive inspection of the final product. All data is synchronized, encoded, and uploaded in real time. Defect Occurrence and Intelligent Diagnosis: Assuming that at time T, the first node detects an abnormally high oxidation index in a section of strip steel (position L100-L110). The system immediately activates the diagnostic model. The model analyzes the associated process parameters and finds that there were intermittent fluctuations in the thermocouple of a certain zone of the annealing furnace a few minutes earlier. The diagnostic report outputs that the substrate at position L100-L110 is slightly oxidized, and the root cause may be the instantaneous temperature fluctuation in zone 5 of the annealing furnace. The system calculates that the time delay Δt for this section of strip steel to travel from the first node to the zinc pot is approximately 120 seconds. At an appropriate time before T+120 seconds, the feedforward control module sends a preparatory command to the zinc pot control system: the strip arriving at [T+Δt] is expected to have slightly lower activity, and it is recommended to slightly raise the zinc bath temperature by 1.5°C for 20 seconds. When this strip passes the second node, the data from the 3D profilometer and thermal imager are analyzed in detail. The results show that the coating quality in this area is not significantly different from the surrounding area, and the feedforward control has suppressed potential defects. The system records this successful intervention case. Grading and Archiving: After the entire roll is produced, the system performs virtual grading based on the full roll inspection map, with 99% classified as Grade A and 1% as Grade B (due to other minor reasons). During winding, Grade B segments are automatically cut to another winding machine. All production data, diagnostic logs, and control records are archived for subsequent knowledge base optimization.

[0073] This embodiment details a next-generation strip steel production line and micro-defect detection method that deeply integrates multimodal perception, big data analysis, machine learning, and intelligent control. Through strategic deployment, construction of a full defect lifecycle map, root cause diagnosis and proactive control, and feedback of quality decisions to production execution and process optimization, it forms a complete, self-learning intelligent quality closed-loop ecosystem.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A strip steel production line, characterized in that, Including those arranged sequentially from left to right along the production line: Uncoiling device (1), used to provide raw material strip steel; Straightening device (2) is used to level the raw material strip steel; Shearing device (3) is used to shear the head and tail of the leveled strip steel; Welding device (4) is used to connect the sheared strip to the tail of the previous coil of strip. The inlet looper (5) is used to store the welded strip and provide a buffer for subsequent processes; A galvanizing treatment device (6) is used to continuously hot-dip galvanize the strip steel; Finished product collection device (7) is used to collect galvanized steel strips; And a detection device (8), which is set on one side of the galvanizing treatment device (6), for online micro-defect detection of the strip surface at different treatment stages.

2. A method for detecting micro-defects in strip steel applied to a strip steel production line as described in claim 1, characterized in that, Includes the following steps: S1: Inside the galvanizing unit, detection units are deployed simultaneously at three process nodes to form an online detection network. At the first node, namely in the transition channel between the outlet of the annealing furnace and the inlet of the galvanizing pot, a first detection unit is deployed, the first detection unit including at least a high-resolution linear CCD camera and a hyperspectral imager. At the second node, that is, in the area after the air knife device at the galvanizing pot outlet and before the post-galvanizing cooling device, a second detection unit is deployed. The second detection unit includes at least a laser three-dimensional profilometer and an infrared thermal imager. At the third node, namely the final quality inspection area before the export looper, a third inspection unit is deployed, which includes at least a surface gloss meter and an eddy current meter. S2: Real-time synchronous acquisition of multimodal sensing data from three nodes, and assignment of a unique spatiotemporal code to the same strip steel passing through each node; Defect features from different nodes are extracted and fused to construct a defect evolution map of the strip steel across process stages; wherein, the features include the oxide distribution features on the substrate surface from the first node, the three-dimensional thickness distribution and solidification thermal field features of the zinc layer from the second node, and the surface finish and subcutaneous defect features of the finished product from the third node. S3: Associate the cross-process stage defect evolution map with the core process parameter set of the production line in the same time period. The process parameter set includes at least the temperature of each section of the annealing furnace, strip tension, zinc liquid temperature, air knife pressure and distance, and cooling rate. By using an online incremental learning model, a dynamic mapping relationship is established between process parameter disturbances, multimodal feature anomalies, and final defect types. When a defect is detected, one or more of the most likely root cause process parameter deviations that led to the defect can be diagnosed in real time. S4: Based on the root cause process parameter deviations diagnosed in step S3, generate two levels of control instructions: Feedforward control: For diagnosed process deviations with fixed propagation delays, the setting parameters of the affected process section are adjusted in advance before the strip reaches the downstream affected process section. Feedback control: For any real-time process deviations diagnosed, the process parameters at the current defect point are adjusted immediately; S5: Record complete defect diagnosis records, control actions, and final quality ratings for each coil of finished strip steel; periodically perform cluster analysis on historical data to identify sensitive intervals of process parameters that cause the recurrence of similar defects, and automatically optimize the process parameter settings within these sensitive intervals to form a continuously evolving process knowledge base.

3. The method for detecting micro-defects in strip steel production lines according to claim 2, characterized in that, In step S1, the first node and the second node are both located between two adjacent processing layers inside the galvanizing device. The detection unit is installed in a sealed protective chamber with a self-cleaning window, and dust and zinc ash contamination is prevented by positive helium pressure.

4. The method for detecting micro-defects in strip steel production lines according to claim 1, characterized in that, The online incremental learning model described in step S3 adopts an ensemble learning algorithm based on concept drift detection. When the production line changes the steel grade or zinc ingot brand, the model can automatically identify the change in data distribution and start a new sub-model to learn without forgetting old knowledge.

5. The method for detecting micro-defects in strip steel production lines according to claim 1, characterized in that, In step S4, the triggering timing of the feedforward control is determined by calculating the precise transmission time of the strip from the root process deviation point to the downstream control point. This transmission time is dynamically calculated from the strip linear speed, looper inventory, and interlayer path length.

6. The method for detecting micro-defects in strip steel production lines according to claim 1, characterized in that, It also includes step S6: Virtual tagging and hierarchical volume division: For defective strips that cannot be completely eliminated through online control, the system performs virtual quality classification based on their defect maps and generates classification instructions. During the export slitting process, the shearing device automatically cuts strips of different quality grades and guides them to different coilers for winding, thereby achieving quality homogenization within the same steel coil.

7. The method for detecting micro-defects in strip steel production lines according to claim 2, characterized in that, The construction of the defect evolution map across process stages described in step S2 is specifically achieved through the following steps: S2.1: Using the aforementioned spatiotemporal coding, the multimodal sensing data from the three nodes are synchronized and aligned in the time and space dimensions to ensure that data points corresponding to the same physical location on the strip are associated. S2.2: For each associated data point, extract the time-series change vector of its defect features along the process direction. The time-series change vector records the feature evolution of a specific location from the substrate state, through galvanizing and solidification, to becoming a finished product. S2.3: Atlas generation and visualization: Arrange the time-series change vectors along the strip length direction to generate a two-dimensional defect evolution atlas with length position as the horizontal axis, process stage as the vertical axis, and feature vector values ​​as elements, and mark the initiation, propagation and morphological evolution trajectory of defects in the atlas.

8. The method for detecting micro-defects in strip steel production lines according to claim 2, characterized in that, Step S4, which generates two-level control instructions, further includes an arbitration and fusion step for conflicting control instructions: S4.3: When multiple feedforward or feedback control commands are generated simultaneously for the same strip steel or the same process equipment, determine whether there is a conflict between these commands in the direction of control target or parameter adjustment. S4.4: If a conflict exists, a decision is made based on a predefined arbitration rule base. The arbitration rule base sets the priority of different control objectives based on process principles and outputs a fused, conflict-free final control instruction sequence.

9. The method for detecting micro-defects in strip steel production lines according to claim 2, characterized in that, The automatic optimization of process parameter settings mentioned in step S5 is achieved through the following steps: S5.1: Based on the cluster analysis results of historical data, identify the fluctuation range of process parameters that cause the same type of defect and define it as the sensitive interval; S5.2: Within the sensitive range, a sequence optimization algorithm is used to systematically adjust the combination of process parameters in the form of simulation or small-batch trial production; S5.3: Use the strip quality rating produced after each parameter adjustment as a feedback signal for the optimization algorithm; S5.4: When a parameter combination that can reliably produce higher quality rated products is found, update the combination to the process knowledge base as the new standard setting value under the production conditions.

10. The method for detecting micro-defects in strip steel during a strip steel production line according to claim 2, characterized in that, It also includes the detection network self-calibration step S0 performed before step S1: S0.1: At each process node of the production line, a standard template with known standard micro-defect characteristics is introduced; S0.2: During breaks in normal production line operation or planned shutdowns, drive the standard template through the field of view of each detection unit to collect its multimodal sensing data; S0.3: Compare the collected data with the known features of the standard sample, calculate the measurement drift error of each sensing channel, and generate the corresponding compensation coefficients for real-time correction of subsequent production and testing data.

Images