Seamless steel tube closed-loop machining method based on online in-situ monitoring

Abstract

The invention provides a seamless steel tube closed-loop processing method based on online in-situ monitoring, which comprises the following steps: when a steel tube reaches a preset online in-situ monitoring station, imaging and identifying the surface of the steel tube through a high-resolution visual system, and planning a plurality of microscopic detection stations; carrying out multi-mode rapid scanning on the surface of the steel pipe by utilizing an atomic force microscope at each microscopic detection site, extracting microscopic characteristic parameters in real time, and forming a microscopic fingerprint data packet of the current steel pipe; inputting the microscopic fingerprint data packet into a pre-trained process inversion model to generate a process adjustment instruction; and the process adjustment instruction is issued to a corresponding execution mechanism in real time through a production line control system, the machining parameters of the current steel pipe or the subsequent steel pipe are adjusted, and microcosmic quality closed-loop control is formed. According to the technical scheme provided by the invention, the crossing from fixed process machining to self-adaptive microstructure regulation and control machining is realized, and the microstructure consistency and the machining precision of the seamless steel tube are improved.

Description

Technical Field

[0001] This application relates to the field of precision plastic processing and intelligent manufacturing technology of metallic materials, and in particular to a closed-loop processing method for seamless steel pipes based on online in-situ monitoring. Background Technology

[0002] Seamless steel pipes, as a fundamental economic material, are widely used in high-end equipment fields such as aviation, nuclear power, and medical devices. As the requirements for material performance in high-end equipment continue to increase, the processing quality of seamless steel pipes increasingly depends on the precise control of their microstructure, including characteristic parameters at the nanometer to micrometer scale, such as grain size, second phase distribution, and surface nano-roughness.

[0003] Currently, the industrial production of seamless steel pipes mainly adopts processing methods based on fixed process specifications, which have the following drawbacks: The process is rigid, using fixed processing parameters. When there are slight fluctuations in the composition of the incoming billet or inhomogeneity in the temperature field of the heating furnace, it cannot be adaptively adjusted, making it difficult to guarantee the consistency of microstructure between different batches and pipe sections; quality feedback is delayed. Microscopic inspection is conducted in the laboratory after processing. Once microstructure defects such as coarse grains or excessive harmful phases are found, the entire batch of products cannot be salvaged, and process adjustments are "post-hoc remediation," resulting in high costs; there is a lack of online process guidance at the nanoscale. For cutting-edge products requiring ultra-smooth inner walls and specific surface textures, their nanoscale roughness and surface nanomechanical properties are completely in a "black box" state during processing, making it impossible to fine-tune the process based on real-time monitoring data; the correlation between "process-microstructure-performance" relies on experience. Process engineers can only roughly correlate macroscopic parameters with final performance based on experience, and cannot establish a precise control model based on real-time microscopic data, severely restricting the improvement of product performance towards theoretical limits. Therefore, there is an urgent need to propose a processing method that can integrate nanoscale micro-characterization techniques into the continuous processing flow of seamless steel pipes and realize online in-situ monitoring and real-time feedback control. Summary of the Invention

[0004] This application provides a closed-loop processing method for seamless steel pipes based on online in-situ monitoring, which at least solves the technical problems of existing seamless steel pipe processing methods, such as lagging, passive, and crude micro-quality control, and the inability to conduct online in-situ monitoring and real-time feedback control of nanoscale micro-features during processing.

[0005] The first aspect of this application proposes a closed-loop processing method for seamless steel pipes based on online in-situ monitoring, the method comprising:

[0006] When the steel pipe reaches the preset online in-situ monitoring station, the surface of the steel pipe is imaged and identified by a high-resolution vision system, and multiple micro-inspection stations are planned. At each microscopic detection station, an atomic force microscope is used to perform multimodal rapid scanning on the surface of the steel pipe, extract microscopic feature parameters in real time, and form a microscopic fingerprint data package of the current steel pipe. The micro-fingerprint data packet is input into a pre-trained process inversion model to generate process adjustment instructions for subsequent processing steps; The process adjustment instructions are sent to the corresponding actuators in real time through the production line control system to adjust the processing parameters of the current steel pipe or subsequent steel pipes, thus forming a micro-quality closed-loop control.

[0007] Preferably, the preset online in-situ monitoring station is set after the hot rolling process and / or after the cold rolling process and / or before the heat treatment process in the seamless steel pipe continuous production line; The process involves imaging and identifying the surface of the steel pipe using a high-resolution vision system, and planning multiple microscopic inspection stations, including: A camera device is used to continuously image the surface of the steel pipe, generating a circumferential unfolded diagram. The circumferential unfolded image is subjected to grayscale conversion, enhancement and filtering to identify and mark macroscopic defect areas as avoidance areas; Based on the steel pipe ID, retrieve the prior micro-feature distribution information of steel pipes of the same specification or batch from the preset historical database to determine one or more prior regions of interest; An adaptive site planning algorithm is used to maximize the spatial dispersion of the planned micro-detection sites and their coverage of the prior interest region within a limited detection time. The adaptive site planning algorithm outputs the three-dimensional coordinates of N micro-detection sites, and combines the three-dimensional coordinates, the corresponding scanning mode, the scanning range, and the preset micro-feature extraction target to form a complete micro-detection site, where N is the total number of micro-detection sites.

[0008] Furthermore, the optimization objective of the adaptive site planning algorithm is to maximize the spatial dispersion of the planned micro-detection sites and the coverage of the prior interest region. The spatial dispersion is achieved by maximizing the minimum Euclidean distance between any two detection stations, and the coverage is achieved by maximizing the sum of the prior interest region weights assigned to each station.

[0009] Preferably, the step of using an atomic force microscope to perform multimodal rapid scanning of the steel pipe surface and extracting microscopic feature parameters in real time includes: Acquire laser displacement signals and probe-sample interaction force signals generated by the atomic force microscope probe during scanning; The acquired raw signals are digitally low-pass filtered to suppress high-frequency mechanical vibration noise in the production line environment; Atomic force microscopes are periodically calibrated using standard grid samples to compensate for nonlinear and creep errors of the piezoelectric scanner and generate calibrated scan data. The calibrated scan data is routed to the morphology channel, nanomechanical channel, and electrical channel according to the data type. Spatially synchronize and pixel-level align the calibration data of each channel to form a fused multimodal dataset; Based on the fused multimodal dataset, the surface arithmetic mean height, ten-point height, surface skewness, and surface kurtosis of the scanning area within the current microscopic detection station are calculated. Based on the morphological and electrical data in the fused multimodal dataset, grain boundaries are automatically identified using an image segmentation algorithm, the distribution of equivalent grain diameters within the current scanning area is statistically analyzed, and the average grain size and grain size uniformity index within the current scanning area are calculated. Based on the force-distance curve data in the fused multimodal dataset, the contact portion of each curve is fitted, and the nanohardness and elastic modulus in the current scanning area are calculated pixel by pixel. The average value and coefficient of variation of the nanohardness and the average value and coefficient of variation of the elastic modulus in the area are statistically analyzed. Based on the fused multimodal dataset, pixel-level spatial alignment and correlation analysis are performed on the distribution maps of morphology data, electrical data, nanohardness, and elastic modulus. Regions in which at least two properties of morphology features, electrical signals, and nanomechanical properties at the same spatial location undergo co-mutation are identified. Regions in which co-mutation occurs are identified as second-phase particles. Then, all identified second-phase particles in the current scanning area are counted, and the area density and average spacing of second-phase particles in the current scanning area are calculated. The microscopic characteristic parameters include: surface arithmetic mean height, ten-point height, surface skewness and surface kurtosis, average grain size and grain size uniformity index, average value and coefficient of variation of nanohardness, average value and coefficient of variation of elastic modulus, and second phase particle area density and average spacing.

[0010] Furthermore, the periodic scanning calibration of the atomic force microscope using standard grid samples includes: By collecting scanning data from standard grid samples, a multivariable polynomial fitting function is established between the voltage applied to the piezoelectric scanner and the actual sample displacement. The fitting function uses ambient temperature and system running time as correction factors to compensate for the scanner's nonlinearity and creep errors.

[0011] Preferably, the process inversion model is a deep neural network or a convolutional neural network, the input of the model is a vector of the microscopic feature parameters of the current steel pipe, and the output is a vector of process adjustment amounts for subsequent processes. The adjustment vector includes one or more of the following: heat treatment temperature adjustment, holding time adjustment, cooling rate adjustment, straightening pressure adjustment, and grinding feed adjustment.

[0012] Furthermore, the output of the process inversion model also includes: confidence scores for each adjustment amount; When the confidence level of any adjustment is greater than the preset threshold, the process adjustment instruction is automatically sent to the actuator; Otherwise, the system retains the current process parameters and triggers a remote review process by human experts.

[0013] Preferably, the process adjustment command operates in at least one of the following ways: For personalized adjustments to subsequent processes of the current steel pipe, the instructions are bound to the steel pipe ID and executed as the steel pipe flows; Feedforward optimization for subsequent batches of steel pipes is used to adjust the initial process parameters of steel pipes of the same batch or similar specifications.

[0014] Preferably, the online in-situ monitoring station includes a magnetorheological damper, a vibration isolation foundation, and a closed chamber to form an active vibration isolation and constant temperature control device, enabling the atomic force microscope to work stably in the vibration and temperature rise environment of the production line.

[0015] Preferably, the method further includes: The final performance data of each steel pipe, the microscopic monitoring data during the process, and the corresponding process adjustment records are used as monitoring signals and fed back to the process inversion model training system. The model parameters are updated periodically or online to continuously optimize the inverse mapping between the microscopic state and the process parameters.

[0016] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects: This application proposes a closed-loop processing method for seamless steel pipes based on online in-situ monitoring. The method includes: when the steel pipe arrives at a preset online in-situ monitoring station, imaging and identifying the surface of the steel pipe using a high-resolution vision system to plan multiple microscopic inspection stations; at each microscopic inspection station, using an atomic force microscope to perform multimodal rapid scanning of the steel pipe surface, extracting microscopic feature parameters in real time, and forming a microscopic fingerprint data package of the current steel pipe; inputting the microscopic fingerprint data package into a pre-trained process inversion model to generate process adjustment instructions for subsequent processing steps; and issuing the process adjustment instructions to the corresponding actuators in real time through the production line control system to adjust the processing parameters of the current or subsequent steel pipes, forming a closed-loop control of microscopic quality. The technical solution proposed in this application realizes a leap from fixed process processing to adaptive microstructure control processing, improving the microstructure consistency and processing accuracy of seamless steel pipes.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of a closed-loop processing method for seamless steel pipes based on online in-situ monitoring, according to an embodiment of this application. Figure 2 This is a schematic diagram showing the installation positions of multiple atomic force microscopes (AFM) according to one embodiment of this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0020] This application proposes a closed-loop processing method for seamless steel pipes based on online in-situ monitoring. The method includes: when the steel pipe arrives at a preset online in-situ monitoring station, imaging and identifying the steel pipe surface using a high-resolution vision system to plan multiple microscopic inspection stations; at each microscopic inspection station, using an atomic force microscope to perform multimodal rapid scanning of the steel pipe surface, extracting microscopic feature parameters in real time to form a microscopic fingerprint data package for the current steel pipe; inputting the microscopic fingerprint data package into a pre-trained process inversion model to generate process adjustment instructions for subsequent processing steps; and issuing the process adjustment instructions in real time to the corresponding actuators through the production line control system to adjust the processing parameters of the current or subsequent steel pipes, forming a closed-loop control of microscopic quality. The technical solution proposed in this application achieves a leap from fixed-process processing to adaptive microstructure control processing, improving the microstructure consistency and processing accuracy of seamless steel pipes.

[0021] The following description, with reference to the accompanying drawings, illustrates a closed-loop processing method for seamless steel pipes based on online in-situ monitoring, according to an embodiment of this application.

[0022] Example 1 Figure 1 This is a flowchart illustrating a closed-loop processing method for seamless steel pipes based on online in-situ monitoring, according to an embodiment of this application. Figure 1 As shown, the method includes: Step 1: When the steel pipe arrives at the preset online in-situ monitoring station, the surface of the steel pipe is imaged and identified by a high-resolution vision system to plan multiple micro-inspection stations; In this embodiment of the disclosure, the preset online in-situ monitoring station is set after the hot rolling process, and / or after the cold rolling process, and / or before the heat treatment process in the seamless steel pipe continuous production line.

[0023] It should be noted that the online in-situ monitoring station includes a magnetorheological damper, a vibration isolation foundation, and a closed chamber, which together form an active vibration isolation and constant temperature control device, enabling the atomic force microscope to work stably under the vibration and temperature rise environment of the production line.

[0024] In this embodiment of the disclosure, the imaging and identification of the steel pipe surface using a high-resolution vision system, and the planning of multiple microscopic detection stations, include: A camera device is used to continuously image the surface of the steel pipe, generating a circumferential unfolded diagram. The imaging device can be a linear CCD camera, an area-array CMOS camera with a rotating mechanism, or multiple fixed cameras stitched together for imaging.

[0025] The circumferential unfolded image is subjected to grayscale conversion, enhancement and filtering to identify and mark macroscopic defect areas as avoidance areas; Based on the steel pipe ID, retrieve the prior micro-feature distribution information of steel pipes of the same specification or batch from the preset historical database to determine one or more prior regions of interest; An adaptive site planning algorithm is used to maximize the spatial dispersion of the planned micro-detection sites and their coverage of the prior interest region within a limited detection time. The adaptive site planning algorithm outputs the three-dimensional coordinates of N micro-detection sites, and combines the three-dimensional coordinates, the corresponding scanning mode, the scanning range, and the preset micro-feature extraction target to form a complete micro-detection site, where N is the total number of micro-detection sites.

[0026] It should be noted that the optimization objective of the adaptive site planning algorithm is to maximize the spatial dispersion of the planned micro-detection sites and the coverage of the prior interest region. The spatial dispersion is achieved by maximizing the minimum Euclidean distance between any two detection stations, and the coverage is achieved by maximizing the sum of the prior interest region weights assigned to each station.

[0027] It should be noted that when a steel pipe enters the monitoring station, a high-resolution linear CCD camera (with a resolution of no less than 2048 pixels / line, and a line frequency synchronized with the steel pipe feed speed) fixed at the station entrance first performs continuous imaging of the steel pipe surface, generating a seamless circumferential unfolded image. The recognition and planning process of the vision system is as follows: Image preprocessing and coarse defect localization: The unfolded image is converted to grayscale, contrast enhanced, and median filtered to eliminate interference from water stains, oxide scale reflections, etc. Subsequently, edge detection and morphological operations based on the Canny operator are used to identify and mark the boundaries of macroscopic defects (such as visible scratches, dents, and roller marks). These areas are recorded as "avoidance zones," which will be actively avoided by subsequent AFM scanning stations.

[0028] Microscopic Feature Region Prediction: Based on the unique identifier (ID) of the steel pipe, the system retrieves prior knowledge of the typical microscopic feature distribution of steel pipes of the same specification, batch, or furnace number from the historical database. For example, historical data may indicate that after rolling, the outer surface circumferential positions corresponding to 1 / 4 and 3 / 4 wall thicknesses of this type of steel pipe are prone to deformation texture. The system will prioritize these positions as candidate detection areas.

[0029] Adaptive Site Planning Algorithm: Combining the "avoidance zone" and the "prior interest zone," the system runs an adaptive site planning algorithm. The goal of this algorithm is to maximize the representation of the micro-state variability of the entire steel pipe within a finite detection time window. The core of the algorithm is as follows: Input: Length L of the steel pipe, list of areas to be avoided, and weighted graph of prior regions of interest.

[0030] Constraints: Total number of detection sites N (e.g., 6), single-site scanning time .

[0031] Optimization objective: To maximize the spatial dispersion of planned sites (e.g., maximizing the minimum Euclidean distance between sites) and the coverage of prior interest regions.

[0032] Output: 3D coordinates (axial positions) of N detection stations Circumferential angle Specifically, the algorithm can be modeled as a constrained optimization problem and solved using simulated annealing: ,in, Where is the radius of the steel pipe. It is the smallest included angle in the circumference. For the prior weight function, , This is the balance coefficient.

[0033] The objective function consists of two core terms, employing a combination of maximizing the minimum distance and maximizing the feature weights: 1. First item: Spatial distribution uniformity control

[0034] Physical meaning: This is The Euclidean distance between the two closest points among the sampling points (the geometric distance when unfolded on the surface of the steel pipe).

[0035] By using max(min(...)), the algorithm forces the most crowded point pairs as far apart as possible. This ensures that AFM monitoring is not confined to a small area, thus avoiding incorrect adjustment of processing parameters due to local sample bias.

[0036] 2. Second item: Feature-driven weight capture

[0037] Physical meaning: It is based on the "salience weight" identified by the vision system (e.g., abnormal reflectivity, complex texture, or belonging to a high-stress risk area in the current scanned area).

[0038] By maximizing the weight sum, the algorithm guides the probe to preferentially land in areas with "story," ensuring that the AFM captures the microscopic features most valuable for subsequent process adjustments.

[0039] The planned station coordinates are converted into the motion path of a six-axis robotic arm equipped with an AFM probe (with a repeatability of ±2μm). The path planning needs to consider the smoothness of the robotic arm joint movements, avoid collisions with steel pipes and other components in the workstation, and optimize the movement time.

[0040] It should be noted that the motion mechanism equipped with the AFM probe is not limited to a six-axis robotic arm, but can also be a three-axis Cartesian coordinate robot, a precision slide with two-dimensional rotation function, or other multi-degree-of-freedom mechanism that can achieve precise positioning of the AFM probe.

[0041] Step 2: At each microscopic detection station, use an atomic force microscope to perform multimodal rapid scanning on the surface of the steel pipe, extract microscopic feature parameters in real time, and form a microscopic fingerprint data package of the current steel pipe; In this embodiment of the disclosure, the step of using an atomic force microscope to perform multimodal rapid scanning of the steel pipe surface and extract microscopic feature parameters in real time includes: 1. Acquire laser displacement signals and probe-sample interaction force signals generated by the atomic force microscope probe during scanning; 2. Perform digital low-pass filtering on the acquired raw signals to suppress high-frequency mechanical vibration noise in the production line environment; It should be noted that the online in-situ monitoring station, through a magnetorheological damper and vibration isolation foundation, forms an active vibration isolation system, attenuating vibrations transmitted from the production line to the AFM probe by at least 40dB in the frequency range above 1Hz. Under this premise, the acquired raw signal undergoes further digital low-pass filtering to suppress residual high-frequency mechanical vibration noise. Simultaneously, the AFM scanning controller employs an adaptive feedforward compensation algorithm based on vibration sensor feedback to correct the piezoelectric scanner's drive signal in real time, thus offsetting the impact of periodic vibrations on the scanning trajectory.

[0042] 3. Periodic scanning calibration of the atomic force microscope is performed using standard grid samples to compensate for the nonlinear and creep errors of the piezoelectric scanner and generate calibrated scanning data; It should be noted that, as Figure 2 As shown, multiple AFM probes can be arranged, with each AFM mounted on an XYZ axis motion table (or robotic arm), which can characterize a specified position based on vision.

[0043] The raw laser displacement signal and probe-sample interaction force signal acquired by the AFM probe are first processed by digital filtering (such as using a Butterworth low-pass filter with a cutoff frequency 10 times the scanning frequency, or a Chebyshev filter, or an elliptic filter, or a Kalman filter) to suppress noise introduced by high-frequency mechanical vibrations in the production line. Simultaneously, periodic scanning calibration is performed using a built-in standard grid sample to compensate for the nonlinearity and creep errors of the piezoelectric scanner. The calibration model can be expressed as: ,in, The voltage applied to the scanner, For the actual sample location, For ambient temperature, For system uptime, This is a multivariate polynomial fitting function established through calibration experiments.

[0044] 4. The calibrated scan data is routed to the morphology channel, nanomechanical channel, and electrical channel according to the data type; 5. Spatially synchronize and pixel-level align the calibration data of each channel to form a fused multimodal dataset; 6. Based on the fused multimodal dataset, calculate the surface arithmetic mean height, ten-point height, surface skewness, and surface kurtosis of the scanning area within the current microscopic detection station; 7. Based on the morphological and electrical data in the fused multimodal dataset, grain boundaries are automatically identified using an image segmentation algorithm, the distribution of equivalent grain diameters within the current scanning area is statistically analyzed, and the average grain size and grain size uniformity index within the current scanning area are calculated. It should be noted that the watershed algorithm is used for segmentation, automatically identifying grain boundaries and calculating the equivalent diameter of the grains. Distribution, calculate average grain size and grain size uniformity index ,in The standard deviation is denoted as .

[0045] 8. Based on the force-distance curve data in the fused multimodal dataset, fit the contact portion of each curve, calculate the nanohardness and elastic modulus of the current scanning area pixel by pixel, and statistically analyze the average value and coefficient of variation of the nanohardness and the average value and coefficient of variation of the elastic modulus in the area. It should be noted that by fitting the contact portion of each force-distance curve, the surface nanohardness is calculated pixel-by-pixel using the Hertzian contact model or the Oliver-Pharr method. and elastic modulus The distribution of the data was plotted, and the average value and coefficient of variation were calculated.

[0046] Furthermore, the Johnson-Kendall-Roberts (JKR) model or the Derjaguin-Muller-Toporov (DMT) model can be used to calculate the surface nanohardness pixel by pixel. and elastic modulus The distribution of the data was plotted, and the average value and coefficient of variation were calculated.

[0047] The above calculation method is selected based on the probe tip shape and the sample adhesion force.

[0048] 9. Based on the fused multimodal dataset, pixel-level spatial alignment and correlation analysis is performed on the morphology data, electrical data, distribution map of nanohardness, and distribution map of elastic modulus. Regions in which at least two properties of morphology features (protrusions, depressions, or compositional contrast), electrical signals (such as conductivity), and nanomechanical properties (hardness or modulus) at the same spatial location undergo co-mutation are identified. Regions in which co-mutation occurs are determined as second-phase particles. Then, all identified second-phase particles in the current scanning area are counted, and the area density and average spacing of second-phase particles in the current scanning area are calculated. It should be noted that by combining the morphological protrusions with abrupt changes in electrical / mechanical properties, the second-phase particles are identified and their area density is calculated. (Number of particles per unit area) and average spacing .

[0049] The microscopic characteristic parameters include: surface arithmetic mean height, ten-point height, surface skewness and surface kurtosis, average grain size and grain size uniformity index, average value and coefficient of variation of nanohardness, average value and coefficient of variation of elastic modulus, and second phase particle area density and average spacing.

[0050] Furthermore, the periodic scanning calibration of the atomic force microscope using standard grid samples includes: By collecting scanning data from standard grid samples, a multivariable polynomial fitting function is established between the voltage applied to the piezoelectric scanner and the actual sample displacement. The fitting function uses ambient temperature and system running time as correction factors to compensate for the scanner's nonlinearity and creep errors.

[0051] It should be noted that all extracted feature parameters, representative images of key areas (such as the area containing the largest grain and the deepest scratch), and data quality indicators (signal-to-noise ratio, calibration status) are packaged to generate a structured "microscopic fingerprint data package". This data package is compressed using a data compression algorithm (such as Zstandard) to reduce transmission latency and is uploaded in real time to the process inversion decision server via industrial Ethernet.

[0052] It should be noted that at each microscopic inspection station, the AFM probe completes multi-mode scanning (including morphology, nanomechanics, and electrical modes) within seconds. The real-time processing unit quantitatively extracts key microscopic feature parameters from the scan data, including but not limited to: nanoscale surface roughness, micro-area grain size and morphology statistics, number density and distribution of second-phase particles, surface nanohardness and elastic modulus distribution, and residual stress gradient information. These parameters constitute the "microscopic fingerprint" of the steel pipe at the current processing stage.

[0053] At each designated microscopic inspection station (MIS), the AFM probe performs rapid, multi-mode scanning. The selection criteria for each MIS include: 1) avoiding macroscopic defects; 2) surface curvature within the AFM probe's scanning range (typically a curvature radius > 5 mm); and 3) reflecting the representative microstructure of the currently scanned area. A typical MIS distribution strategy is as follows: along the steel pipe axis, at least one MIS is planned at the head, middle, and tail; at each axial position, at least two MIS are planned circumferentially (e.g., at 0° and 180°) to assess circumferential uniformity.

[0054] Step 3: Input the micro-fingerprint data package into the pre-trained process inversion model to generate process adjustment instructions for subsequent processing steps; In this embodiment of the disclosure, the process inversion model is a deep neural network or a convolutional neural network. The input of the model is a vector of the microscopic feature parameters of the current steel pipe, and the output is a vector of process adjustment amounts for subsequent processes. It should be noted that the process inversion model can also be a machine learning model such as Random Forest, Gradient Boosting Decision Tree (GBDT), or an "expert system" based on physical rules.

[0055] The adjustment vector includes one or more of the following: heat treatment temperature adjustment, holding time adjustment, cooling rate adjustment, straightening pressure adjustment, and grinding feed adjustment.

[0056] It should be noted that the output of the process inversion model also includes: confidence scores for each adjustment quantity; When the confidence level of any adjustment is greater than the preset threshold, the process adjustment instruction is automatically sent to the actuator; Otherwise, the system retains the current process parameters and triggers a remote review process by human experts.

[0057] Specifically, the extracted "microscopic fingerprint" parameter set is standardized according to feature categories: for each feature parameter, the mean of that parameter in the historical training dataset is subtracted, and then divided by its standard deviation, mapping the data of each dimension to a similar numerical range. The standardized feature parameter vector is then input into a pre-trained process inversion intelligent model (such as a deep neural network). This model learns and establishes an inverse mapping relationship between "processing parameters" and "generated microstructures." Based on real-time microscopic features, the model inversely infers the adjustment amount of parameters required for subsequent processing steps to achieve the target microstructure (preset according to product grade and application). For example, based on the currently detected large grain size and excessively deep surface hardening layer, the model determines that "the cooling rate in the previous stage is insufficient" and decides to "increase the cooling water flow rate after final rolling by 8% for the next steel pipe" or "adjust the tempering temperature curve for this steel pipe in subsequent heat treatment processes." The decision instructions are sent in real time to the corresponding actuators (such as valves, heaters, and mill servo systems) through the production line control system.

[0058] The model infers adjustment suggestions based on the real-time acquired microscopic feature parameter vectors. The real-time microscopic features specifically refer to the quantified set of microscopic physical parameters extracted in real-time from the currently inspected steel pipe at the planning site (MIS) through AFM scanning and processing in step C. This set includes data in at least the following dimensions: Topographic dimensions: surface arithmetic mean roughness Sa (unit: nm), maximum height Sz (unit: nm).

[0059] Structural dimension: average grain size (Unit: μm), Grain size uniformity index (Dimensionless).

[0060] Mechanical dimension: Average nanometer hardness (Unit: GPa), coefficient of variation of hardness distribution (%). (Among them, the coefficient of variation of nanohardness) , The standard deviation of the hardness at the sampling points. (This is the average value.) Phase composition dimension: Second phase particle area density (Unit: particles / mm²), average interparticle spacing (Unit: μm). These parameters together constitute the high-dimensional, digital "microstate" input for the model to perform reverse reasoning.

[0061] The process inversion model generates specific, differentiated process adjustment instructions based on the deviation between real-time micro-features and target values. Below are examples of several typical deviation scenarios and corresponding adjustment strategies: Scenario 1: The grain size is too large and the uniformity is poor. If the average grain size is detected... Greater than the target value If the content is above 10% and the uniformity index U < 0.85, the model judges it as insufficient recrystallization or excessively rapid growth.

[0062] Adjustment strategy: For subsequent heat treatment processes, decide to increase the solution temperature (e.g., increase by 15-30℃) and / or extend the holding time (e.g., extend by 15-25%) to provide greater phase transformation driving force, promote more complete recrystallization and grain refinement.

[0063] Execution method: The adjusted temperature-time curve is bound to the ID of the steel pipe and sent to the heat treatment furnace control system.

[0064] Scenario 2: Surface nano-roughness Sa exceeds the standard. If the detected Sa value is higher than the target upper limit (e.g., Sa>20 nm), the model judges it to be caused by severe surface friction or adhesion during the previous rolling or straightening process, or uneven cooling.

[0065] Adjustment strategy: In subsequent finishing processes, adjust the centerless grinding or polishing process parameters. For example, increase the grinding wheel feed rate by 5-10%, reduce the workpiece speed by 8-12%, or replace with a finer-grit grinding wheel / polishing media.

[0066] Execution method: The instruction is sent to the CNC system of the finishing machine tool.

[0067] Scenario 3: The density of second-phase particles is abnormally high. If detected... The levels were significantly higher than the historical baseline, and the model indicated that improper precipitation control during heating or cooling may have led to the formation of harmful brittle phases.

[0068] Adjustment strategy: Decide to adjust subsequent cooling processes. For example, for supersaturated precipitation, accelerate the cooling rate (e.g., change air cooling to mist cooling or water cooling); for carbides that require spheroidization treatment, decide to adopt a specific spheroidization annealing temperature profile (e.g., hold at a temperature in the two-phase region for a long time).

[0069] Execution method: The command is sent to the control unit of the cooling section or annealing furnace.

[0070] Scenario 4: The distribution of nano-hardness is extremely uneven ( >20% indicates localized deformation or uneven phase transformation of the material, potentially leading to residual stress concentration.

[0071] Adjustment strategies: The decision is to add a stress-relief annealing process in subsequent steps, or to adjust the straightening force distribution mode of the straightening machine (such as using multi-point variable pressure straightening).

[0072] Execution method: The instruction is sent to the controller of the newly added annealing station or intelligent straightening machine.

[0073] Adjustment amount of model output Typically, it's a vector with a sign (+ / -) and a confidence level, such as: adjustment vector. It can be represented as: in, For temperature adjustment amount, This is the adjustment amount for the straightening pressure. The amount in parentheses represents the feed rate adjustment. The confidence score is given for the corresponding item.

[0074] Only when the overall confidence level exceeds the preset threshold (like The adjustment instruction will only be executed automatically when the value is 0.8; otherwise, a manual confirmation process will be triggered. The decision will be made if and only if the judgment criterion is met. When the condition is met, the instruction is automatically sent to the actuator; otherwise, the system will maintain the current process parameters and simultaneously trigger a remote review process by human experts to ensure the safety and robustness of the closed-loop control.

[0075] Step 4: The process adjustment command is sent to the corresponding actuator in real time through the production line control system to adjust the processing parameters of the current steel pipe or subsequent steel pipes, forming a micro-quality closed-loop control.

[0076] In this embodiment of the disclosure, the process adjustment command operates in at least one of the following ways: For personalized adjustments to subsequent processes of the current steel pipe, the instructions are bound to the steel pipe ID and executed as the steel pipe flows; Feedforward optimization for subsequent batches of steel pipes is used to adjust the initial process parameters of steel pipes of the same batch or similar specifications.

[0077] In this embodiment of the disclosure, the method further includes: The final performance data of each steel pipe, the microscopic monitoring data during the process, and the corresponding process adjustment records are used as monitoring signals and fed back to the process inversion model training system. The model parameters are updated periodically or online to continuously optimize the inverse mapping between the microscopic state and the process parameters.

[0078] It should be noted that the adjustment command immediately applies to the subsequent processes of the steel pipe itself (such as adjusting the temperature curve of the heat treatment furnace it is about to enter), or to the steel pipes that are processed later (such as adjusting the rolling process), to achieve "single-piece self-adaptation" or "batch feedforward optimization".

[0079] After processing, the final performance data of the steel pipe is associated with and stored along with all microscopic monitoring data and process adjustment records during the process, which is used to continuously optimize the process inversion model.

[0080] Taking the manufacture of cold-rolled thin-walled seamless steel pipes made of high-temperature alloys (such as Inconel 718) for aero-engines as an example, the specific implementation of this method is illustrated.

[0081] Processing line configuration: The online in-situ AFM monitoring and feedback control station of the present invention is set between the cold rolling mill and the subsequent vacuum solution heat treatment furnace.

[0082] Example of key parameters: Target microstructure: uniform grain size of ASTM grade 10 or finer, surface nano-roughness Sa < 20 nm, and no microcracks or anomalous precipitates.

[0083] AFM station scanning speed: using peak force tapping mode, single point (20μm×20μm) scanning time ≤ 8 seconds.

[0084] Process inversion model: A convolutional neural network (CNN) trained based on historical production data and physical simulation data. The input is a multidimensional microscopic feature vector, and the output is a suggestion for adjusting the solution temperature and time.

[0085] Processing procedure: Initial processing: The steel pipe is cold rolled according to the standard process.

[0086] Initial monitoring and feedback: The rolled steel pipe enters the AFM monitoring station. The vision system positions the pipe, and the AFM performs a rapid scan at two circumferential stations at the tail end. The edge processor extracts the average grain size (currently estimated at 12 μm) and detects trace amounts of elongated grains and dislocation entanglements on the surface.

[0087] Real-time decision-making: The process inversion model analyzes the data and determines that "the current microstructure exhibits rolling deformation texture and high residual strain energy. If processed according to the original solution treatment process (standard temperature 980℃), it may lead to uneven grains after recrystallization." The model decides: "For this branch of steel pipe, increase the solution treatment temperature to 995℃ and extend the holding time by 10%" to promote more complete recrystallization and grain homogenization.

[0088] Instruction Execution: This decision instruction is sent from the production line control system to the host computer of the vacuum heat treatment furnace, indexed by the steel pipe ID. When the steel pipe enters the furnace, the furnace temperature control system automatically calls the customized heating program (995℃) for it.

[0089] Data closed loop and learning: The metallographic inspection results of the steel pipe after final heat treatment (such as the actual grain size reaching ASTM 10.5 grade) are used as a monitoring signal and fed back to the system to optimize the parameters of the process inversion model, making its next prediction and decision more accurate.

[0090] This method seamlessly integrates the microscopic insight of atomic force microscopy into the macroscopic processing flow, forming a new generation of intelligent processing method for seamless steel pipes with real-time microscopic quality control as its core.

[0091] The closed-loop processing method for seamless steel pipes based on online in-situ monitoring proposed in this embodiment has the following advantages: Achieving microstructure-oriented adaptive processing: The processing objective is deepened from meeting macroscopic size standards to meeting microstructure standards, and the process can be automatically adjusted based on real-time monitoring results, so that the microstructure of each steel pipe can converge to the ideal target, significantly improving batch consistency.

[0092] Transforming offline sampling inspection into online full inspection eliminates batch accidents of microscopic quality: By conducting in-situ "microscopic live inspection" of key process points of each steel pipe, near-full inspection of microscopic quality monitoring is achieved. Microscopic abnormalities can be detected and corrected in a timely manner during processing, avoiding the scrapping of the entire batch.

[0093] Overcoming the challenges of nanoscale process monitoring: Real-time monitoring and feedback methods have been provided for the nanoscale surface quality processing of ultra-precision steel tubes (such as grating ruler tubes and hydraulic servo cylinders), enabling related processes to move from "experience-based trial and error" to a "data-driven" precise control stage.

[0094] Build an evolvable process knowledge base: The continuously accumulated full-link data of "process parameters - real-time microstructure - final performance" enables the process inversion model to continuously evolve, and eventually realize the "reverse manufacturing" mode that automatically generates and executes the optimal processing parameters by specifying the target performance or microstructure.

[0095] In summary, the closed-loop processing method for seamless steel pipes based on online in-situ monitoring proposed in this embodiment achieves a leap from fixed process processing to adaptive microstructure control processing, thereby improving the microstructure consistency and processing accuracy of seamless steel pipes.

[0096] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0097] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0098] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A closed-loop processing method for seamless steel pipes based on online in-situ monitoring, characterized in that, The method includes: When the steel pipe reaches the preset online in-situ monitoring station, the surface of the steel pipe is imaged and identified by a high-resolution vision system, and multiple micro-inspection stations are planned. At each microscopic detection station, an atomic force microscope is used to perform multimodal rapid scanning on the surface of the steel pipe, extract microscopic feature parameters in real time, and form a microscopic fingerprint data package of the current steel pipe. The micro-fingerprint data packet is input into a pre-trained process inversion model to generate process adjustment instructions for subsequent processing steps; The process adjustment instructions are sent to the corresponding actuators in real time through the production line control system to adjust the processing parameters of the current steel pipe or subsequent steel pipes, thus forming a micro-quality closed-loop control.

2. The method as described in claim 1, characterized in that, The preset online in-situ monitoring station is set after the hot rolling process and / or after the cold rolling process and / or before the heat treatment process in the seamless steel pipe continuous production line. The process involves imaging and identifying the surface of the steel pipe using a high-resolution vision system, and planning multiple microscopic inspection stations, including: A camera device is used to continuously image the surface of the steel pipe, generating a circumferential unfolded diagram. The circumferential unfolded image is subjected to grayscale conversion, enhancement and filtering to identify and mark macroscopic defect areas as avoidance areas; Based on the steel pipe ID, retrieve the prior micro-feature distribution information of steel pipes of the same specification or batch from the preset historical database to determine one or more prior regions of interest; An adaptive site planning algorithm is used to maximize the spatial dispersion of the planned micro-detection sites and their coverage of the prior interest region within a limited detection time. The adaptive site planning algorithm outputs the three-dimensional coordinates of N micro-detection sites, and combines the three-dimensional coordinates, the corresponding scanning mode, the scanning range, and the preset micro-feature extraction target to form a complete micro-detection site, where N is the total number of micro-detection sites.

3. The method as described in claim 2, characterized in that, The optimization objective of the adaptive site planning algorithm is to maximize the spatial dispersion of the planned micro-detection sites and the coverage of the prior interest region. The spatial dispersion is achieved by maximizing the minimum Euclidean distance between any two detection stations, and the coverage is achieved by maximizing the sum of the prior interest region weights assigned to each station.

4. The method as described in claim 1, characterized in that, The method of using atomic force microscopy to perform multimodal rapid scanning of the steel pipe surface and extract microscopic feature parameters in real time includes: Acquire laser displacement signals and probe-sample interaction force signals generated by the atomic force microscope probe during scanning; The acquired raw signals are digitally low-pass filtered to suppress high-frequency mechanical vibration noise in the production line environment; Atomic force microscopes are periodically calibrated using standard grid samples to compensate for nonlinear and creep errors of the piezoelectric scanner and generate calibrated scan data. The calibrated scan data is routed to the morphology channel, nanomechanical channel, and electrical channel according to the data type. Spatially synchronize and pixel-level align the calibration data of each channel to form a fused multimodal dataset; Based on the fused multimodal dataset, the surface arithmetic mean height, ten-point height, surface skewness, and surface kurtosis of the scanning area within the current microscopic detection station are calculated. Based on the morphological and electrical data in the fused multimodal dataset, grain boundaries are automatically identified through a preset algorithm, the distribution of equivalent grain diameters in the current scanning area is statistically analyzed, and the average grain size and grain size uniformity index in the current scanning area are calculated. The preset algorithm includes: image segmentation algorithm, watershed algorithm, semantic segmentation model based on deep learning, and edge detection algorithm based on gray-level gradient. Based on the force-distance curve data in the fused multimodal dataset, the contact portion of each curve is fitted, and the nanohardness and elastic modulus in the current scanning area are calculated pixel by pixel. The average value and coefficient of variation of the nanohardness and the average value and coefficient of variation of the elastic modulus in the area are statistically analyzed. Based on the fused multimodal dataset, pixel-level spatial alignment and correlation analysis are performed on the distribution maps of morphology data, electrical data, nanohardness, and elastic modulus. Regions in which at least two properties of morphology features, electrical signals, and nanomechanical properties at the same spatial location undergo co-mutation are identified. Regions in which co-mutation occurs are identified as second-phase particles. Then, all identified second-phase particles in the current scanning area are counted, and the area density and average spacing of second-phase particles in the current scanning area are calculated. The microscopic characteristic parameters include: surface arithmetic mean height, ten-point height, surface skewness and surface kurtosis, average grain size and grain size uniformity index, average value and coefficient of variation of nanohardness, average value and coefficient of variation of elastic modulus, and second phase particle area density and average spacing.

5. The method as described in claim 4, characterized in that, The periodic scanning calibration of the atomic force microscope using standard grid samples includes: By collecting scanning data from standard grid samples, a multivariable polynomial fitting function is established between the voltage applied to the piezoelectric scanner and the actual sample displacement. The fitting function uses ambient temperature and system running time as correction factors to compensate for the scanner's nonlinearity and creep errors.

6. The method as described in claim 1, characterized in that, The process inversion model is a deep neural network or a convolutional neural network. The input of the model is a vector of the microscopic feature parameters of the current steel pipe, and the output is a vector of process adjustment amounts for subsequent processes. The adjustment vector includes one or more of the following: heat treatment temperature adjustment, holding time adjustment, cooling rate adjustment, straightening pressure adjustment, and grinding feed adjustment.

7. The method as described in claim 6, characterized in that, The output of the process inversion model also includes: confidence scores for each adjustment quantity; When the confidence level of any adjustment is greater than the preset threshold, the process adjustment instruction is automatically sent to the actuator; Otherwise, the system retains the current process parameters and triggers a remote review process by human experts.

8. The method as described in claim 1, characterized in that, The process adjustment command operates in at least one of the following ways: For personalized adjustments to subsequent processes of the current steel pipe, the instructions are bound to the steel pipe ID and executed as the steel pipe flows; Feedforward optimization for subsequent batches of steel pipes is used to adjust the initial process parameters of steel pipes of the same batch or similar specifications.

9. The method as described in claim 1, characterized in that, The online in-situ monitoring station includes a magnetorheological damper, a vibration isolation foundation, and a closed chamber, forming an active vibration isolation and constant temperature control device to enable the atomic force microscope to work stably in the vibration and temperature rise environment of the production line.

10. The method as described in claim 1, characterized in that, The method further includes: The final performance data of each steel pipe, the microscopic monitoring data during the process, and the corresponding process adjustment records are used as monitoring signals and fed back to the process inversion model training system. The model parameters are updated periodically or online to continuously optimize the inverse mapping between the microscopic state and the process parameters.

Images