Steel plate defect detection method and system, electronic equipment and storage medium

Abstract

The invention discloses a steel plate defect detection method and system, electronic equipment and a storage medium, and is characterized in that the method comprises the following steps: synchronously collecting a surface optical image, ultrasonic detection data and infrared thermal imaging data of a steel plate, and carrying out space-time registration and noise suppression processing on the multi-modal data; high-precision multi-defect detection: through multi-modal data fusion and a deep learning model, the single-frame processing time consumption is less than or equal to 15ms, the omission ratio is less than or equal to 0.3%, and the industrial high-speed continuous operation requirement is met; the interference such as oil contamination and light reflection is effectively inhibited through multi-modal data cross validation, the false alarm rate is reduced, the environmental adaptability reaches-20 DEG C to 60 DEG C / 95% RH, and the stability of complex working conditions is ensured; manual work is replaced by full-automatic detection, so that the cost is saved, a defect data driving process is optimized, the yield is improved by 3%-5%, and the operation and maintenance efficiency is improved by 70%; the modular design supports detection of multiple metal plates, the model supports online learning and remote upgrading, seamless connection with an industrial Internet of Things platform is achieved, and the technology expansibility is high.

Description

Technical Field

[0001] The present invention relates to the technical field of steel plates, and in particular to a steel plate defect detection method and system, electronic equipment, and a storage medium. Background Art

[0002] Steel plate, a key product of the steel industry, is widely used in industries such as chemicals, machinery manufacturing, aerospace, automotive, and home appliances. Modern steel plate production and manufacturing places increasingly stringent quality requirements on steel plate. Due to factors such as equipment, production processes, and the environment, various surface defects can appear on the steel plate. Surface defects not only affect the product's appearance but also limit its application, hinder further productivity improvements, and directly reduce the steel plate's quality. With downstream industries continuously pursuing high-quality raw materials, such as the automotive, home appliance, and power companies' demand for zero-defect strip surface quality, steel companies must provide even higher-quality products. Consequently, the issue of steel plate surface quality testing has garnered widespread attention.

[0003] Current steel plate defect detection mainly relies on manual visual inspection, single sensors (such as X-rays or ultrasound), or traditional image processing technology, which has the following drawbacks: 1. Manual inspection: low efficiency, high cost, and greatly affected by operator experience; 2. Single sensor technology: can only detect specific types of defects (e.g., ultrasonic detection of internal cracks but not surface rust); 3. Traditional image processing: relies on manual feature extraction, is sensitive to complex background noise, and has poor generalization ability. Summary of the Invention

[0004] In order to overcome the deficiencies of the prior art, one of the objectives of the present invention is to provide a steel plate defect detection method and system, electronic equipment, and storage medium.

[0005] One of the objectives of the present invention is achieved by the following technical solution: A method for detecting defects in a steel plate, characterized in that it comprises the following steps: S1: Synchronously collects surface optical images, ultrasonic detection data, and infrared thermal imaging data of steel plates; S2: performing spatiotemporal registration and noise suppression processing on the multimodal data; S3: Input the preprocessed data into a multi-branch deep learning model, wherein the model includes: Image branch: uses improved ResNet-50 to extract surface defect features; Acoustic branch: Processing ultrasonic spectrum features based on 1D-CNN; Thermal imaging branch: Analyze temperature field distribution through Transformer architecture; S4: The cross-modal attention fusion module integrates the features of each branch and outputs the defect type, location and size information; S5: Generate a test report based on the test results and feed it back to the production line control system.

[0006] Steel plate defect detection system, including: Multimodal acquisition module, including a high-resolution linear array camera, an ultrasonic probe array, and an infrared thermal imager; a signal processing module configured to perform preprocessing operations of S2; Edge computing module, deploying multi-branch deep learning models; Control the output module, link with the production line PLC and trigger the sorting mechanism.

[0007] An electronic device comprises a memory, a processor and a computer program stored in the memory, wherein the processor implements the steps of the method as described in S1-S5 when executing the program.

[0008] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in S1-S5.

[0009] The spatiotemporal registration in step S2 specifically includes: aligning the timestamps of the image and the ultrasonic data based on the encoder signal; and achieving the unification of multi-sensor spatial coordinates through feature point matching.

[0010] Furthermore, the calculation formula of the cross-modal attention fusion module is:

[0011] Among them, Q, K, and V come from the eigenvectors of different modal branches respectively.

[0012] Furthermore, the ultrasonic probe array is arranged in a cross-staggered manner, and the distance between adjacent probes is less than 1 / 2 of the ultrasonic wavelength.

[0013] Furthermore, it also includes the use of knowledge distillation technology in the model inference stage to compress the teacher model (parameter number ≥ 100 million) into the student model (parameter number ≤ 20 million).

[0014] Furthermore, the processor integrates an FPGA acceleration unit for real-time calculation of the fast Fourier transform of the ultrasonic signal.

[0015] Furthermore, the lighting system of the line array camera adopts a double-sided LED light source, and the brightness can be automatically adjusted according to the reflectivity of the steel plate surface.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. High-precision multi-defect detection: Through multimodal data fusion and deep learning models, joint detection of surface and internal defects is achieved, with improved average accuracy, supporting the identification of tiny defects as small as 0.1mm, and covering more than 90% of defect types in ISO standards.

[0017] 2. High-speed real-time response: Embedded hardware acceleration design supports real-time detection at a production line speed of 5m / s, with single-frame processing time ≤15ms and missed detection rate ≤0.3%, meeting the needs of industrial high-speed continuous operation.

[0018] 3. Strong anti-interference ability: Multimodal data cross-validation effectively suppresses interference such as oil stains and reflections, reduces false alarm rate, and has an environmental adaptability of -20℃~60℃ / 95%RH, ensuring stability in complex working conditions.

[0019] 4. Efficient cost reduction and efficiency improvement: Fully automated testing replaces manual labor to save costs, defect data drives process optimization, improves yield rate by 3%-5%, and improves operation and maintenance efficiency by 70%.

[0020] 5. Flexible and scalable architecture: The modular design supports multi-metal plate detection, the model supports online learning and remote upgrades, seamlessly connects to the industrial Internet of Things platform, and has strong technical scalability. The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following is a detailed description of the preferred embodiments with the help of the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of this embodiment. DETAILED DESCRIPTION

[0022] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0023] It should be noted that when a component is referred to as being "fixed to" another component, it may be directly on the other component or there may also be a central component. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may also be a central component. When a component is considered to be "set on" another component, it may be directly set on the other component or there may also be a central component. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used in this specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0025] See also Figure 1 , a steel plate defect detection method, characterized in that it comprises the following steps: S1: Synchronously collects surface optical images, ultrasonic detection data, and infrared thermal imaging data of steel plates; S2: Perform spatiotemporal registration and noise suppression on multimodal data; S3: Input the preprocessed data into a multi-branch deep learning model, which includes: Image branch: uses improved ResNet-50 to extract surface defect features; Acoustic branch: Processing ultrasonic spectrum features based on 1D-CNN; Thermal imaging branch: Analyze temperature field distribution through Transformer architecture; S4: The cross-modal attention fusion module integrates the features of each branch and outputs the defect type, location and size information; S5: Generate a test report based on the test results and feed it back to the production line control system.

[0026] Steel plate defect detection system, including: Multimodal acquisition module, including a high-resolution linear array camera, an ultrasonic probe array, and an infrared thermal imager; a signal processing module configured to perform preprocessing operations of S2; Edge computing module, deploying multi-branch deep learning models; The control output module is linked with the production line PLC and triggers the sorting mechanism. When the processor executes the program, the steps of the S1-S5 method are implemented. When the program is executed by the processor, the steps of the S1-S5 method are implemented. The spatiotemporal registration in step S2 specifically includes: aligning the timestamps of the image and ultrasonic data based on the encoder signal; achieving the unification of multi-sensor spatial coordinates through feature point matching. The calculation formula of the cross-modal attention fusion module is:

[0027] Among them, Q, K, and V come from the eigenvectors of different modal branches respectively. The ultrasonic probe array is arranged in a cross-staggered manner, and the distance between adjacent probes is less than 1 / 2 of the ultrasonic wavelength. It also includes the use of knowledge distillation technology in the model inference stage to compress the teacher model (parameter number ≥ 100 million) into the student model (parameter number ≤ 20 million). The processor integrates an FPGA acceleration unit for real-time calculation of the fast Fourier transform of the ultrasonic signal. The lighting system of the linear array camera uses a double-sided LED light source, and the brightness can be automatically adjusted according to the reflectivity of the steel plate surface.

[0028] Experimental data and effect verification To verify the effectiveness of the present invention, multiple comparative experiments were conducted on core indicators such as detection accuracy, real-time performance, and small defect recognition capability. The specific data are as follows: 1. Detection accuracy comparison experiment Experimental setup: Dataset: 1,200 steel plate samples (1m × 2m in size) collected from industrial sites, containing 200 images of each of six types of defects (scratches, pores, inclusions, cracks, rust, and indentations), and 200 images of normal samples.

[0029] Comparison method: Traditional method 1: based on SVM+handcrafted features (HOG+LBP); Traditional method 2: Single-modal CNN (image branch only); Existing patented solution: single sensor + ResNet-34; The present invention: multimodal fusion MBDN model; Evaluation indicators: mAP (mean average precision), F1-Score.

[0030] Experimental results:

[0031] Conclusion: Multimodal data fusion improves mAP by 9.1%, and the inference speed meets real-time requirements (≤20ms).

[0032] 2. Real-time test (production line speed compatibility) Experimental conditions: Steel plate moving speed: 1m / s to 10m / s Detection system hardware: NVIDIA Jetson AGX Xavier + PCIe-6321 acquisition card Data throughput: Image (8192×512@200fps) + Ultrasonic (100MHz sampling) result:

[0033] Conclusion: At a speed of ≤5m / s, the missed detection rate is ≤0.5%, meeting the needs of industrial high-speed detection.

[0034] 3. Verification of small defect detection capability Test sample: Artificially prepared micro defects (size 0.05mm×0.05mm to 0.3mm×0.3mm); Defect types: surface scratches, internal micropores; Detection threshold analysis:

[0035] Conclusion: The detection rate of the present invention for 0.1mm×0.1mm defects is ≥98%, and its sensitivity is significantly better than that of the traditional method.

[0036] 4. False alarm rate test (anti-interference ability) Interference scenario: Surface oil, water stains, and reflections; Ambient temperature fluctuation (±15°C); False positive statistics:

[0037] Conclusion: The complementarity of multimodal data effectively suppresses the influence of a single interference source, with a false alarm rate of ≤1.5%.

[0038] 5. Verification of the effectiveness of multimodal data (ablation experiment) Experimental design: Option A: Image data only Option B: Image + Ultrasound Solution C: Image + Infrared Solution D: Full modality (image + ultrasound + infrared) Results comparison:

[0039] in conclusion: Ultrasonic waves significantly contribute to internal defect (pore) detection (AP increased by 36.4%); Infrared data improves crack detection (+21.6%); Full modal fusion achieves optimal overall performance.

[0040] 6. System robustness test Test items: Long-term operation stability: continuous operation for 24 hours, sampling and testing every hour; Extreme environment: temperature (-10℃ to 50℃), humidity (30%-95%RH).

[0041] result:

[0042] Conclusion: The embedded hardware platform remains stable in harsh environments and meets industrial-grade reliability requirements.

[0043] 7. Industrial field test (case study at a steel plant) Testing period: November 2024-April 2025 Steel plate type: cold rolled steel plate (thickness 2-10mm) Production line speed: 4m / s Detection targets: surface defects (scratches, indentations), internal defects (inclusions, pores) Statistical results:

[0044] Conclusion: The present invention increases the defect detection rate by 42.7%, achieves zero false detection shutdown, and significantly reduces production costs.

[0045] The above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by technicians in this field on the basis of the present invention fall within the scope of protection required by the present invention.

Claims

1. A method for detecting defects in steel plates, characterized in that: The following steps are involved: S1: Synchronously collects surface optical images, ultrasonic detection data, and infrared thermal imaging data of steel plates; S2: performing spatiotemporal registration and noise suppression processing on the multimodal data; S3: Input the preprocessed data into a multi-branch deep learning model, wherein the model includes: Image branch: uses improved ResNet-50 to extract surface defect features; Acoustic branch: Processing ultrasonic spectrum features based on 1D-CNN; Thermal imaging branch: Analyze temperature field distribution through Transformer architecture; S4: The cross-modal attention fusion module integrates the features of each branch and outputs the defect type, location and size information; S5: Generate a test report based on the test results and feed it back to the production line control system.

2. Steel plate defect detection system, characterized in that, include: Multimodal acquisition module, including a high-resolution linear array camera, an ultrasonic probe array, and an infrared thermal imager; a signal processing module configured to perform the preprocessing operation of S2 in claim 1; An edge computing module deploying the multi-branch deep learning model according to claim 1; Control the output module, link with the production line PLC and trigger the sorting mechanism.

3. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the processor executes the program, the steps of the method according to claim 1 are implemented.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to claim 1 are implemented.

5. The method according to claim 1, wherein The spatiotemporal registration in step S2 specifically includes: aligning the timestamps of the image and the ultrasonic data based on the encoder signal; and achieving the unification of multi-sensor spatial coordinates through feature point matching.

6. The method according to claim 1, characterized in that The calculation formula of the cross-modal attention fusion module is: ; Among them, Q, K, and V come from the eigenvectors of different modal branches respectively.

7. The system according to claim 2, wherein: The ultrasonic probe array is arranged in a cross-staggered manner, and the distance between adjacent probes is less than 1 / 2 of the ultrasonic wavelength.

8. The method according to claim 1, characterized in that It also includes the use of knowledge distillation technology in the model inference stage to compress the teacher model (parameter number ≥ 100 million) to the student model (parameter number ≤ 20 million).

9. The electronic device according to claim 3, wherein: The processor integrates an FPGA acceleration unit for real-time calculation of the fast Fourier transform of ultrasonic signals.

10. The system according to claim 2, wherein: The lighting system of the line array camera adopts double-sided LED light sources, and the brightness can be automatically adjusted according to the reflectivity of the steel plate surface.

Images