CN-122023279-A - Online real-time detection method for casting blank section size

Abstract

The invention relates to an online real-time detection method for the cross section size of a casting blank, and belongs to the technical field of detection methods for the quality of casting blanks in the continuous casting production process. The method comprises the following steps of arranging cameras above and on two sides of a casting blank after flame cutting of a continuous casting machine, synchronously collecting multi-angle images of the section of the casting blank, calculating the real-time section size of the casting blank by means of image denoising, segmentation and edge extraction pretreatment and combining a pixel-actual size calibration relation and thermal expansion compensation, judging whether the size is qualified or not through a dynamic threshold value, and triggering early warning and adjustment response. The method has the beneficial effects that non-contact, full-section and high-precision detection of the section size of the casting blank is realized through multi-view industrial camera deployment, image preprocessing and feature extraction algorithm, a size calculation model and dynamic threshold judgment, and meanwhile, the method has the functions of abnormality early warning and data tracing and provides support for quality control and production parameter optimization of the casting blank.

Inventors

• HAN JIAN
• FAN JIA
• WU ZHIJIE
• HE FANG
• GUO FAN
• SHI LEI
• YANG GUIYU
• SU ZHENJUN

Assignees

• 邯郸钢铁集团有限责任公司
• 河钢股份有限公司邯郸分公司

Dates

Publication Date

20260512

Application Date

20260107

Claims (5)

1. 1. The online real-time detection method for the cross section size of the casting blank is characterized by comprising the following steps of: (1) The method comprises the steps of disposing and calibrating an industrial camera system, arranging a main camera right above a casting blank, and arranging side cameras on two sides of the casting blank in the width direction respectively to form a triangular view angle layout so as to ensure no vision blind area; (2) An image data processing center is built, an industrial control computer is adopted in the field data processing center, and image processing software developed based on Python+OpenCV is carried, so that 'image acquisition-preprocessing-feature extraction' automation is realized; (3) Calculating and judging the cross section size of the casting blank, constructing a full-section size calculation-dynamic threshold judgment secondary model based on the preprocessed image and the calibration relation, and realizing accurate calculation of the size and abnormal early warning; (4) Outputting and executing response of the detection result, and outputting the result and triggering process adjustment through multiple channels after size calculation and abnormality judgment are completed.
2. 2. The method for online real-time detection of the cross-sectional dimension of a casting blank according to claim 1, wherein the step (1) comprises the following steps: (11) Camera mounting and layout: 1) Setting up fixed brackets on two sides and above the casting blank outlet roller way for installing cameras to form a triangular view angle layout: 2) Two lateral cameras are respectively arranged on two sides of the casting blank in the width direction, and are 1.5-2.0 m away from the surface of the casting blank, and the axis of each lens is perpendicular to the side face of the casting blank and is used for capturing the width and the thickness of a section; 3) The main camera is arranged right above the casting blank and is 2.0-2.5 m away from the surface of the casting blank, and the axis of the lens is perpendicular to the section of the casting blank and is used for capturing the diagonal and edge deformation of the section; 4) The high-temperature-resistant protective cover and the compressed air purging device are additionally arranged outside the camera, and the bracket adopts an anti-vibration design; (12) The camera selection parameters comprise selecting an industrial area array camera, wherein the resolution is more than or equal to 2048 multiplied by 1536 pixels, the frame rate is more than or equal to 30fps, selecting an industrial fixed focus lens with a polarizing filter, and environmental suitability, namely, the working temperature is-20-60 ℃, the protection level is IP67, and electromagnetic interference is prevented; (13) Performing system calibration, and establishing a mapping relation of pixel to actual size: 1) During cold state calibration, a standard size calibration plate is placed at the outlet position of a casting blank, three calibration plate images of cameras are synchronously collected, four vertexes of the calibration plate are identified through a corner detection algorithm, pixel coordinates of each vertex in the images are calculated, and a mapping formula of pixel distances and actual sizes is established: actual size=pixel distance×k K is a calibration coefficient, and is calculated by the actual size of the calibration plate and the pixel distance in the image; 2) Thermal expansion compensation, namely collecting the surface temperature of a casting blank in real time, and calculating thermal state size correction according to the material of the casting blank: Corrected size=image calculation size× [ 1+α× (T-T 0 ) ] Wherein T 0 is the cold temperature, and 25 ℃.
3. 3. The method for online real-time detection of the cross-sectional dimension of a casting blank according to claim 1, wherein the step (2) comprises the following steps: (1) And (3) synchronously collecting images: 1) The method comprises the steps of acquiring image data of three cameras at the frequency of 30fps, synchronously acquiring the image data of the three cameras, respectively marked as I_lateral 1 (t), I_lateral 2 (t) and I_upper (t), wherein t is time, and simultaneously reading casting blank temperature, drawing speed and steel grade information from a continuous casting machine PLC through an OPC protocol; 2) The data frame format is set as a time stamp, a camera number, image data, casting blank temperature, pulling speed, a steel grade code and a check bit, so that data traceability is ensured; 3) Storing strategy, namely caching the real-time image in a memory, compressing and storing the processed size data and the key image to an industrial database every 1 minute, wherein the retention time is more than or equal to 3 months; (22) Image preprocessing: 1) Denoising, namely removing image noise caused by dust by adopting a Gaussian filter algorithm, and removing isolated noise points generated by molten steel splashing by adopting a median filter algorithm; 2) The reflection inhibition is carried out, namely a reflection area is segmented and shielded according to gray level distribution on the surface of a casting blank through a self-adaptive threshold segmentation algorithm, and gray level average values of adjacent areas are used for filling; 3) The section segmentation, namely extracting the edge of the section of the casting blank by adopting a Canny edge detection algorithm, filling the edge gap by combining with morphological closing operation, and finally segmenting a complete section area of the casting blank; (23) Abnormal image processing: 1) Calculating a definition evaluation index of each frame of image, judging that the lens is polluted if the continuous 5 frames of images are blurred, triggering a compressed air purging device to purge strongly, and replacing the compressed air purging device with a size average value of the effective images of the previous 3 frames; 2) And (3) diagnosing camera faults, namely triggering fault alarm if a certain camera does not output images or the transmission error rate of image data is more than 5% after 10 seconds, marking the camera data invalid, and starting the cross calculation of the other two cameras.
4. 4. The method for online real-time detection of the cross-sectional dimension of a casting blank according to claim 1, wherein the step (3) comprises the following steps: (31) Full section size calculation: 1) Core size calculation: The width calculation, namely identifying the left edge and the right edge of the casting blank section along the horizontal direction through a side camera image, calculating the pixel distance between the edges, and combining the calibration coefficient and thermal expansion compensation to obtain the real-time width W (t): w (T) = [ pixel distance_left and right (T) ×k ] × [ 1+α× (T) -T 0 ) ] 2) The thickness calculation comprises the steps of identifying the upper edge and the lower edge of a casting blank section along the vertical direction through an upper camera image, calculating the pixel distance between the edges, and obtaining the real-time thickness H (t) by combining a calibration coefficient and thermal expansion compensation; 3) The diagonal deviation calculation, namely identifying four vertexes of a casting blank section through an upper camera image, namely A, B, C and D respectively, and calculating the actual lengths L1=AC (t) and L2=BD (t) of the two diagonals, wherein the diagonal deviation delta L (t) = |L1-L2|; 4) The size verification is that the average value of the calculation results of the three cameras is taken as the final size, and the single-view angle error is reduced; (32) Dynamic threshold determination: establishing a dynamic size threshold library based on the casting blank steel grade and the drawing speed: 1) Setting a threshold value, namely establishing a relation between the threshold value and the pull speed through linear regression according to historical production data: W _ upper limit (v) =a x v + b, lower limit of W (v) () =c×v+d Wherein a, b, c and d are fitting coefficients, and are determined by a process test; 2) Abnormality determination: ① The first-stage early warning is triggered when W (t) exceeds [ W_lower limit (v), W_upper limit (v) ] and H (t) exceeds the thickness qualification range or delta L (t) >0.05mm, and the first-stage early warning is prompted to be abnormal in size and needs to be concerned; ② The second-level early warning is triggered when W (t) exceeds [ W_lower limit (v) -0.03, W_upper limit (v) +0.03] and H (t) exceeds the thickness qualification range of +/-0.03 mm, or DeltaL (t) >0.08mm and the duration is more than or equal to 3 seconds, and the serious abnormality is judged to be needed to be immediately regulated; 3) And predicting the dimension W_pred (t+2s) of 2 seconds in the future according to the dimension change trend of the first 5 seconds, and triggering early warning 2 seconds in advance if the predicted value exceeds a secondary early warning threshold.
5. 5. The method for online real-time detection of the cross-sectional dimension of a casting blank according to claim 1, wherein the step (4) comprises the following steps: (41) Displaying in real time, and dynamically displaying a three-dimensional visual image of a casting blank section, a real-time size curve, a dynamic threshold range, a camera state and key parameters on a man-machine interaction interface of a central control room; (42) Alarm and adjustment response: 1) Primary early warning, namely flashing a yellow indicator lamp of a central control room and buzzing at low frequency, wherein an early warning window is popped up by an HMI to prompt an operator to check the pressure of a roller way of the continuous casting machine; 2) And (43) data tracing, namely automatically generating a daily/weekly size detection report which comprises an average size value, abnormal times, early warning processing records and key abnormal images, and supporting to derive an Excel/PDF format for quality tracing and process parameter optimization.

Description

Online real-time detection method for casting blank section size Technical Field The invention relates to an online real-time detection method for the cross section size of a casting blank, and belongs to the technical field of casting blank quality detection devices and methods in continuous casting production processes. Background The cross section size of casting blank (such as the width and thickness of slab, the side length and diagonal line of square billet) is one of the core quality indexes of continuous casting production, and the qualified cross section size can ensure the material utilization rate (reduce the trimming amount) of the subsequent rolling process, and avoid the problems of steel clamping during rolling, uneven product thickness and the like caused by size deviation. Currently, the detection methods of the casting blank section size are mainly divided into two types of off-line sampling detection and on-line detection, and have obvious limitations: 1. The off-line spot check method is that samples are manually selected after casting blanks are cooled, and the critical dimension of the section is measured by using a caliper and a laser calliper. The method has three main defects: (1) The hysteresis is serious, the detection result only reflects the casting blank state at the sampling detection moment, and the dynamic change of the section size (for example, the size shrinkage rate of the hot casting blank after cooling is 0.3% -0.5% and the off-line value cannot represent the actual hot size) caused by the abnormal taper of the crystallizer, the fluctuation of the pull speed and the uneven roller pressure in the production process cannot be captured in real time; (2) The accuracy and the representativeness are insufficient, wherein the manual measurement error is generally more than 0.15mm, and a single casting blank is only subjected to sampling inspection for 1-2 sections, so that the full-length size fluctuation cannot be covered, and batch 'hidden unqualified products' are easily caused to flow into a downstream process; (3) The efficiency is low, the single detection needs to take 5-10 minutes, the casting blank transferring flow is required to be interrupted, and the operation rate of the continuous casting production line is affected. 2. Some enterprises try to adopt a contact thickness gauge (such as a roller sensor) or a single laser line scanning technology, but an inherent short plate exists: (1) The contact thickness gauge needs to be in direct contact with a high-temperature casting blank (the thermal state temperature reaches 800-1000 ℃), the sensor is fast in abrasion (the service life is less than 2 weeks), the surface of the casting blank is easy to scratch to form cracks, and the section width and diagonal line cannot be measured; (2) The single laser line scanning is that only line data of a certain section of the section can be obtained, the full section (such as 'bulging' deformation of the edge of a plate blank is easy to miss detection) cannot be covered, and the data integrity is insufficient (the effective data rate is less than 85%) due to molten steel splashing and on-site dust shielding; (3) The adaptability is poor, the existing detection method is mostly aimed at fixing the steel grade and the section (such as adapting only 1800mm wide slab), when the casting blank specification is switched, the equipment needs to be debugged again (consuming more than 2 hours), and the flexible production requirement can not be met. With the improvement of the requirements of the steel industry on the production with high precision, flexibility and intelligence (such as the requirement of +/-0.08 mm of the dimensional tolerance of the section of a casting blank of a high-end automobile plate), the traditional detection method cannot be adapted. Industry needs a non-contact full-section coverage, high-precision and strong-environmental-adaptability casting blank section size online detection technology, and the specific requirements are as follows: (1) Real-time requirements are that abnormal section sizes of casting blanks need to be identified within 3 seconds, so that batch scrapping in the subsequent rolling process is avoided; (2) The full section coverage requirement is that the deformation of the section (comprising edges and diagonal lines) of the casting blank is required to be completely captured, so that the partial defect missing judgment is avoided; (3) The environment adaptability requirement is that the continuous casting site high temperature (the environment temperature is less than or equal to 60 ℃), dust (the concentration is less than or equal to 10mg/m < 3 >) and molten steel reflection interference can be tolerated; (4) The flexible adaptation requirement is that when the casting blank specification (the width is 800-220mm and the thickness is 150-300 mm) is switched, shutdown and debugging are not needed. In addition, although the laser radar-based c