Coiling furnace rotating hub strip steel insertion depth control system and method based on AI vision

Abstract

The invention relates to the technical field of steel production intelligent control, and particularly discloses a coiling furnace rotating hub strip steel insertion depth control system and method based on AI vision, an AI vision camera of an AI vision detection module captures the head and tail morphology of strip steel in real time, calculates, controls and records the strip steel insertion depth, circularly learns, optimizes and adjusts the insertion precision; the double-calibration-line speed measurement module is provided with two parallel calibration lines which are spaced by a certain distance in the vertical direction of the inlet channel of the coiling furnace, detects the time when the head of the strip steel passes through the two calibration lines, and calculates the passing speed of the strip steel; the multi-modal data fusion module receives a visual speed signal, a rolling mill encoder speed signal and a hot metal detector position signal, and outputs a fusion optimization speed through a Kalman filter; the insertion safety depth calculation module dynamically calculates the strip steel insertion depth according to the strip steel head shape; the control precision is improved, and the threading success rate is increased; the impact force of the strip steel head on the rotating hub is reduced; the service life of equipment is prolonged; scrap steel is reduced, and the cost is saved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of steel production, and particularly relates to a coiling furnace hub strip insertion depth control system and method based on AI vision. BACKGROUND

[0002] In the process of coiling furnace hub coiling strip, the strip insertion depth needs to be accurately controlled to avoid the strip head impacting the hub under high temperature environment, causing the internal rib plate of the hub to be severely deformed, damaging the hub and even cracking. Field statistical data shows that this problem shortens the hub life by 30%, but the traditional control method faces three technical bottlenecks:

[0003] Speed detection lag: The actual speed deviation from the mill line speed is often more than 0.2 m / s, and the actual strip insertion depth deviation is large.

[0004] Plate shape interference: The existing hot metal detector (HMD) cannot identify the appearance, and the position detection deviation is greater than 200 mm due to the head and tail distortion such as variable fish tail shape and sickle bend.

[0005] Depth control rigidity: The current strip insertion hub slot depth cannot be accurately quantified, and manual feedback adjustment can only be based on qualitative experience, and cannot dynamically adapt to speed and plate shape changes for accurate control. Therefore, a coiling furnace hub strip insertion depth control system and method based on AI vision are needed to solve the insertion depth control inaccuracy caused by speed detection distortion and head shape interference in the process of coiling furnace hub coiling strip head insertion. SUMMARY

[0006] In view of the problems in the prior art, the present application aims to provide a coiling furnace hub strip insertion depth control system and method based on AI vision, which is suitable for intelligent control of the strip coiling process and realizes high-precision strip insertion depth control of 1000±100 mm. The qualitative experience of strip insertion depth is converted into quantitative data measured based on AI image analysis, and intelligent feedback control is realized.

[0007] The technical scheme adopted by the present application to solve its technical problems is: a coiling furnace hub strip insertion depth control system based on AI vision, comprising an AI vision detection module, a double-calibration-line speed measurement module, a multi-modal data fusion module and an insertion safety depth calculation module,

[0008] The AI vision detection module includes an AI vision camera, which captures the head and tail shape of the strip in real time, calculates and controls the strip insertion depth, and optimizes the insertion accuracy through cyclic learning;

[0009] The dual calibration line speed measurement module sets two parallel calibration lines with a certain distance in the vertical direction of the coiling furnace entrance channel, detects the time when the strip head passes through the two calibration lines, and calculates the strip travel speed V v ;

[0010] The multimodal data fusion module receives the visual speed signal V v , rolling mill encoder speed signal V e And the hot metal detector position signal, through the Kalman filter output fusion optimization speed V f ;

[0011] The insertion safety depth calculation module: based on the formula strip insertion safety depth D safe =K×V f ×t resp +C, dynamically calculate the strip insertion depth according to the strip head shape; when the strip insertion depth D≥D safe Start the hub.

[0012] Specifically, the dual-calibration line speed measurement module sets up two calibration lines with a spacing of L on the winding channel. When the strip head touches the first calibration line, the timing starts T1, and when it touches the second calibration line, the timing stops T2;

[0013] Calculation of basic strip insertion speed

[0014] Thermal expansion compensation model L real =L×[1+α(T-25)], where α=1.2×10 -5 / ℃ is the expansion coefficient of the bracket between the two calibration lines;

[0015] Speed ​​calculation after temperature compensation

[0016] Specifically, the AI ​​visual detection module divides the head area and extracts the fishtail deformation index and sickle curvature R c ;

[0017] Calculation logic of flatness compensation coefficient K:

[0018] When W f >0.15: K=1.2+0.2(W f -0.15),

[0019] When R c >1.5° / m: K=1.3+0.3(R c -1.5),

[0020] Normal plate shape K = 1.0;

[0021] An AI vision analysis model is used to segment the tail region and extract the actual size of the strip insertion depth, which is used to feedback and adjust the strip insertion depth control in the next round.

[0022] Specifically, in the strip insertion safety depth formula of the insertion safety depth calculation module, K is a shape compensation coefficient, t resp ≤50ms is the system response time, and C is the basic insertion depth of the strip head.

[0023] Specifically, the strip insertion safety depth D safe When the value is in the range of 950mm to 1300mm, the hub is started, and when the value exceeds the limit, a three-level protection mechanism is activated.

[0024] Software truncation: when the set safety depth D safe >1300mm, the value of D safe is forcibly set to 1250mm.

[0025] Limit depth: when the strip insertion depth reaches the limit maximum depth D MAX =1350mm;

[0026] Emergency switching: when the vision fails, switch to the mill roll encoder detected mill line speed mode V e .

[0027] A working method of a coiling furnace hub strip insertion depth control system based on AI vision, comprising the following steps:

[0028] S1, when the strip head passes through the first calibration line, trigger the AI vision camera to collect the head image and start timing;

[0029] S2, stop timing when the head touches the second calibration line, and trigger the AI vision camera to analyze and calculate the strip head passing speed V v ;

[0030] S3, multi-modal data fusion strip head passing speed V v , mill line speed V e and HMD position signal, output the optimal speed V f ;

[0031] S4, calculate the shape compensation coefficient K based on the strip head topography characteristics;

[0032] S5, calculate the safety depth according to the formula D safe =K×V f ×0.05+C;

[0033] S6, start the hub when the strip insertion depth D safe ≥D

[0034] S7, when the strip steel is turned out for re-rolling, the AI vision camera captures the feature image formed after the strip steel is inserted into the hub clamping slot and turned out, analyzes, calculates and records the actual insertion depth, and feeds back, calculates and adjusts the reasonable insertion depth of the next round of strip steel.

[0035] Specifically, in the multi-modal data fusion in step S3, a federated learning framework is constructed to locally estimate the speed value V local as the basis, and the optimized global fusion speed V f is the core variable of hierarchical cooperation, aggregates global knowledge, forms a closed loop of "edge perception-cloud decision-edge execution", and provides a reliable speed reference for the strip steel insertion depth control of the coiler hub.

[0036] Specifically, in the step S6, when the strip steel insertion depth D is greater than or equal to D safe , the initial rotating speed is started according to the calculation n1=60*V f / (2*π*R), and three layers of protection mechanisms are included.

[0037] The present application has the following beneficial effects:

[0038] The AI vision-based coiler hub strip steel insertion depth control system and method designed by the present application improve the control precision, and the strip steel insertion depth error can be controlled within ±100mm; improve the threading success rate, reduce the steel stacking accident, and reduce the accident; reduce the impact force of the strip steel head on the hub, prolong the service life of the equipment; reduce scrap steel, save cost, and generate benefits. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 is the overall architecture diagram of the AI vision-based coiler hub strip steel insertion depth control system.

[0040] Figure 2 is the flowchart of the double-calibration-line speed measurement principle.

[0041] Figure 3 is the flowchart of the AI vision module.

[0042] Figure 4 is the flowchart of multi-modal data fusion.

[0043] Figure 5 is the multi-modal data fusion federated learning framework diagram.

[0044] Figure 6 is the flowchart of the hub start control. DETAILED DESCRIPTION

[0045] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely further describe the technical solutions in the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0046] like Figures 1-6 As shown, an AI vision-based coiling furnace hub strip insertion depth control system includes an AI vision detection module, a dual-calibration line speed measurement module, a multimodal data fusion module, and an insertion safety depth calculation module. The AI ​​vision detection module includes an AI vision camera, which captures the strip head and tail morphology in real time, calculates and controls the strip insertion depth, performs cyclic learning, and optimizes and adjusts the insertion accuracy. The dual-calibration line speed measurement module sets two parallel calibration lines at a certain distance in the vertical direction of the coiling furnace entrance channel, detects the time it takes for the strip head to pass through the two calibration lines, and calculates the strip travel speed V v ; The multimodal data fusion module receives the visual speed signal V v , rolling mill encoder speed signal V e And the hot metal detector position signal, through the Kalman filter output fusion optimization speed V f ; Insertion safety depth calculation module: Based on the formula strip insertion safety depth D safe =K×V f ×t resp +C, dynamically calculate the strip insertion depth according to the strip head shape; K in the formula is the plate shape compensation coefficient, t resp ≤50ms is the system response time, C is the basic insertion depth of the strip steel head.

[0047] When the strip insertion depth D≥D safe Start the hub. Strip steel insertion safety depth D safe ∈[950,1300]mm, the hub is started, and the three-level protection mechanism is activated when the limit is exceeded: software cutoff: when the safety depth D is set safe >1300mm, set D compulsorily safe =1250mm; Limit depth: detect when the strip insertion depth reaches the maximum limit depth D MAX =1350mm; Emergency switch: switch to the mill linear speed mode V detected by the mill roll encoder when visual failure occurs e .

[0048] The technical solution adopted by the present invention includes dual-calibration line spatiotemporal detection + head shape AI compensation + multimodal data fusion control + tail actual depth detection feedback, specifically including:

[0049] 1.Dual calibration line speed measurement module.

[0050] Two calibration lines are set at the inlet channel of the coiling furnace, with a distance L of mm (accuracy ±0.5 mm). When the head of the strip touches the calibration line:

[0051] Trigger high-precision timer (error ≤0.1 ms);

[0052] Calculate the real-time threading speed

[0053] Compensate model L through thermal expansion real = L × [1 + α (T-25)], eliminate temperature influence, ensure speed detection accuracy ±0.03 m / s.

[0054] 2. AI visual head shape recognition.

[0055] Capture head image with 200 million pixel infrared array camera (2048 × 1 resolution, 200 fps), realize through improved YOLOv11 model:

[0056] Fish tail deformation detection: calculate the maximum width ratio W f =(W max / W0)-1;

[0057] Falciform bend recognition: extract centerline curvature radius R c ;

[0058] Output plate shape compensation coefficient K:

[0059]

[0060] 3. Dynamic calculation of insertion safety depth and actual depth detection feedback.

[0061] Build depth control model: D safe = K·V f ·t real +C;

[0062] Where:

[0063] V f : Kalman filter fusion speed (fusion of visual speed V v and encoder speed V e );

[0064] t resl : system response time ≤50 ms;

[0065] C: basic insertion amount 300 mm;

[0066] When D safe ∈[950,1300] mm, start the hub, and activate the three-level protection mechanism when it exceeds the limit.

[0067] 4. Multi-modal data fusion federated learning.

[0068] A multi-modal data fusion federated learning framework is constructed to estimate the edge local speed value V local Based on this, the edge visual node and the encoder node are realized in real time to ensure the timeliness of control and the optimized global fusion speed V f The core variable of hierarchical collaboration is constituted to aggregate global knowledge, improve accuracy and robustness, dynamically calibrate the local model, realize continuous evolution, and form a closed loop of "edge perception-cloud decision-edge execution", avoiding visual interference from steam and other related factors, and providing a reliable speed reference for the insertion control of the coiling furnace rotating hub.

[0069] Embodiment one: 3500 furnace coiling line coiling furnace strip insertion depth control.

[0070] Hardware configuration example.

[0071] Calibration line: a feature line on the coiling furnace bottom guide device, embedded with zirconia ceramic rods (temperature resistance 1500℃), interval L, unit mm.

[0072] Camera: Basler raL2048-48gm line array camera, 2048 pixels, 200fps.

[0073] Lens: Schneider Kreuznach Xenoplan 35mm fixed focus lens.

[0074] Light source: Laser Components 808nm pulse laser, peak power 500W.

[0075] Processor: NVIDIA Jetson AGX Orin, computing power 200TOPS.

[0076] Control flow details.

[0077] 1. Speed detection stage.

[0078] When the strip head touches the first calibration line:

[0079] Trigger the camera to take continuous pictures (exposure time 80μs);

[0080] Start PTP synchronization timer T1;

[0081] When the head leaves the second calibration line;

[0082] Stop timer T2;

[0083] Basic speed calculation V=L / (T2-T1);

[0084] Temperature compensation: Lreal = L x [1 + 1.2e -5 (T-25)];

[0085] Temperature compensation speed calculation V v = L x [1 + 1.2e -5 (T-25)] / (T2-T1).

[0086] 2. Data fusion stage.

[0087] Where the Kalman gain K is updated every 50 ms.

[0088] 3. Depth calculation stage.

[0089] Fish tail deformation (W f = 0.18), then K = 1.2 + 0.2 x (0.18-0.15) = 1.26;

[0090] Take V f = 1.5 m / s, t real = 0.05 s;

[0091] D safe = 1.26 x 1.5 x 0.05 + 0.3 = 1.0245 m.

[0092] 4. Hub control stage.

[0093] Based on the above image analysis, when the strip steel cuts into the hub slot and the strip steel insertion depth D1 is greater than or equal to 1024.5 mm, the hub is started, and the building hub control stage is entered.

[0094] The present application solves the core problem of the insertion depth control of the coiling furnace hub through the triple technical innovation of "double calibration line space-time detection + AI visual topography compensation + federal learning data fusion".

[0095] Breakthrough in detection accuracy: visual speed measurement error ≤0.03 m / s, 5 times higher than traditional encoder measurement accuracy.

[0096] Adaptive capability: the shape compensation coefficient K dynamically responds to head variations, covering more than 90% of distortion conditions.

[0097] Real-time control: from visual recognition to hub action response t real ≤50 ms, meeting the 2 m / s high-speed strip passing requirement.

[0098] System robustness: three-level protection mechanism ensures that the maximum strip steel insertion depth D MAX ≤1300 mm, eliminating mechanical damage caused by the impact of the strip head on the hub.

[0099] The present application is not limited to the above-mentioned embodiments, and any person should know that the structural changes made under the inspiration of the present application, any technical solutions with the same or similar to the present application, fall within the scope of protection of the present application.

[0100] The technology, shape, and configuration parts not described in detail in the present application are well-known technology.

Claims

1. A coiling furnace hub strip steel insertion depth control system based on AI vision, characterized in that: Including AI visual detection module, dual-calibration line speed measurement module, multimodal data fusion module and insertion safety depth calculation module, The AI ​​visual inspection module includes an AI visual camera, which captures the shape of the strip head and tail in real time, calculates and controls the recording of the strip insertion depth, performs cyclic learning, and optimizes and adjusts the insertion accuracy; The dual calibration line speed measurement module sets two parallel calibration lines with a certain distance in the vertical direction of the coiling furnace entrance channel, detects the time when the strip head passes through the two calibration lines, and calculates the strip travel speed V v ; The multimodal data fusion module receives the visual speed signal V v , rolling mill encoder speed signal V e And the hot metal detector position signal, through the Kalman filter output fusion optimization speed V f ; The insertion safety depth calculation module: based on the formula strip insertion safety depth D safe =K×V f ×t resp +C, dynamically calculate the strip insertion depth according to the strip head shape; when the strip insertion depth D≥D safe Start the hub.

2. The coiling furnace hub strip steel insertion depth control system based on AI vision according to claim 1 is characterized in that: The dual-calibration line speed measurement module sets up two calibration lines with a spacing of L on the winding channel. When the strip head touches the first calibration line, the timing starts T1, and when it touches the second calibration line, the timing stops T2; Calculation of basic strip insertion speed Thermal expansion compensation model L real =L×[1+α(T-25)], where α=1.2×10 -5 / ℃ is the expansion coefficient of the bracket between the two calibration lines; Speed ​​calculation after temperature compensation 3. The coiling furnace hub strip steel insertion depth control system based on AI vision according to claim 1 is characterized in that: The AI ​​visual detection module divides the head area and extracts the fishtail deformation index and sickle curvature R c ; Calculation logic of flatness compensation coefficient K: When W f >0.15: K=1.2+0.2(W f -0.15), When R c >1.5° / m: K=1.3+0.3(R c -1.5), Normal plate shape K = 1.0; An AI visual analysis model is used to segment the tail area and extract the actual size of the strip insertion depth, which is used as feedback to adjust the next round of strip insertion depth control.

4. The coiling furnace hub strip steel insertion depth control system based on AI vision according to claim 1 is characterized in that: In the strip steel insertion safety depth formula of the insertion safety depth calculation module, K is the plate shape compensation coefficient, t resp ≤50ms is the system response time, C is the basic insertion depth of the strip steel head.

5. The coiling furnace hub strip steel insertion depth control system based on AI vision according to claim 1 is characterized in that: The strip is inserted into the safety depth D safe The hub is started when the speed is ∈[950,1300]mm. When the speed exceeds the limit, the three-level protection mechanism is activated: Software cutoff: When setting safety depth D safe >1300mm, set D compulsorily safe =1250mm; Limit depth: detect when the strip insertion depth reaches the maximum limit depth D MAX =1350mm; Emergency switch: switch to the mill linear speed mode V detected by the mill roll encoder when visual failure occurs e .

6. The working method of the coiling furnace hub strip steel insertion depth control system based on AI vision according to any one of claims 1 to 5, characterized in that: The following steps are involved: S1: When the strip head passes the first calibration line, the AI ​​visual camera is triggered to capture the head image and start timing; S2. When the head touches the second calibration line, the timing stops and the AI ​​visual camera is triggered to perform visual analysis and calculate the speed V of the strip head. v ; S3, Multimodal data fusion steel head travel speed V v , rolling mill linear speed V e and HMD position signal, output optimal speed V f ; S4. Calculate the flatness compensation coefficient K based on the strip head shape characteristics; S5. According to formula D safe =K×V f ×0.05+C to calculate the safety depth; S6, the strip is punched into the hub slot to a depth of D ≥ D safe Start the hub when S7. When the strip is rotated out and rolled again, after the head becomes the tail, the AI ​​visual camera grabs the strip and inserts it into the hub slot and rotates it out to form a characteristic image, analyzes, calculates and records the actual insertion depth, and feeds back, calculates and adjusts the reasonable insertion depth of the next round of strip.

7. The working method of the coiling furnace hub strip steel insertion depth control system based on AI vision according to claim 6 is characterized in that: In the multimodal data fusion in step S3, a federated learning framework is constructed to estimate the local speed V local Based on the optimized global fusion speed V f The core variables that constitute layered collaboration aggregate global knowledge to form a closed loop of "edge perception-cloud decision-making-edge execution", providing a reliable speed benchmark for the coiler hub strip insertion depth control.

8. The working method of the coiling furnace hub strip steel insertion depth control system based on AI vision according to claim 6 is characterized in that: The strip insertion depth D≥D in step S6 safe According to the calculation, n1=60×V f / (2×π×R) initial speed to start the hub, and also includes three layers of protection mechanisms.

Images