Vacuum arc furnace molten pool state monitoring method based on image processing

Abstract

The present application relates to the technical field of image data processing, in particular to a vacuum consumable arc furnace molten pool state monitoring method based on image processing, comprising: acquiring infrared thermal images of the vacuum consumable arc furnace molten pool at current and historical multiple moments, and carrying out noise reduction processing on the infrared thermal images to obtain smooth images; extracting a stable contour based on the smooth images; calculating the multi-scale morphological entropy and the geometric fluctuation degree of the stable contour; respectively obtaining trend values of the multi-scale morphological entropy and the geometric fluctuation degree by using a trend analysis method; and issuing a warning signal when any trend value exceeds the corresponding preset upper limit. The present application solves the problems of unstable monitoring results and insufficient reliability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing. More particularly, the present application relates to a vacuum consumable arc furnace molten pool state monitoring method based on image processing. BACKGROUND

[0002] As a key metallurgical equipment commonly used for the preparation of high-purity metals and alloys, the vacuum consumable arc furnace is in a high-temperature, high-vacuum and strong electric arc radiation environment during the smelting process. The stability of the molten pool directly affects the composition uniformity, microstructure density of the final ingot, and the safety and energy efficiency level of the smelting process. In actual production, the molten pool of the vacuum consumable arc furnace is jointly affected by multiple factors such as electric arc discharge intensity, melting rate, furnace pressure and cooling conditions. The molten pool shape and temperature distribution will show significant dynamic fluctuations. If the molten pool fluctuates too much, it may cause uneven flow of liquid metal, blockage of inclusion floating, abnormal thickness of crystallization layer, and thus cause metallurgical defects such as inclusion segregation, shrinkage cavity, crack, and even cause smelting interruption or equipment damage. Therefore, real-time and accurate monitoring of the molten pool state and evaluation of its stability are important links to ensure the safe and efficient operation of the vacuum consumable arc furnace.

[0003] However, traditional molten pool state monitoring relies on indirect analysis methods of electric signals such as arc current and voltage waveform, or local temperature detection by thermocouples, infrared probes and other sensors. However, electric signal monitoring is greatly affected by arc fluctuations and electromagnetic interference, making it difficult to reflect the real geometric shape change of the molten pool; and the contact measurement method of temperature sensors is not suitable for high-vacuum high-temperature environment, which easily leads to equipment wear or measurement delay. In recent years, non-contact detection methods based on image processing have gradually attracted attention. In particular, infrared thermal imaging technology can continuously obtain the temperature field distribution information of the molten pool surface without affecting the furnace conditions, providing an important basis for shape feature extraction and dynamic state analysis. However, due to the light radiation interference caused by electric arc discharge, random noise in infrared images and thermal gradient blurring of the molten pool edge, traditional image processing algorithms based on single threshold segmentation or edge detection have problems such as misjudgment, fracture or contour drift when extracting the stable contour of the molten pool, resulting in unstable monitoring results and insufficient reliability. SUMMARY

[0004] To solve the problem of unstable monitoring results and insufficient reliability raised in the background art, the present application provides the following solutions.

[0005] In the scheme, the application provides a vacuum consumable arc furnace molten pool state monitoring method based on image processing, comprising: acquiring infrared thermal images of the vacuum consumable arc furnace molten pool at current and historical moments, and performing noise reduction processing on the infrared thermal images to obtain a smoothed image; extracting a stable contour based on the smoothed image, the extraction of the stable contour specifically comprising: calculating gradient amplitudes and local image entropies of each pixel point in the smoothed image, obtaining a corrected gray value of each pixel point after weighted fusion of the gradient amplitudes and the local image entropies, constructing a feature map based on the corrected gray value, performing adaptive threshold segmentation on the feature map to obtain a binary image, and taking a boundary of a target region in the binary image as the stable contour; calculating a multi-scale morphological entropy and a geometric fluctuation degree of the stable contour; the multi-scale morphological entropy is a ratio of a change amount of a stable contour area after morphological operation at each scale to a total change amount of the stable contour area after morphological operation at each scale; trend values of the multi-scale morphological entropy and the geometric fluctuation degree are respectively obtained by using a trend analysis method; and when any trend value exceeds a corresponding preset upper limit, an early warning signal is sent.

[0006] The above technical scheme can not only quantitatively depict the stability and fluctuation intensity of the molten pool shape, but also realize early warning before the molten pool state abnormally fluctuates, thereby effectively improving the safety, stability and intelligent control level of the smelting process, and providing reliable data support and technical support for operation optimization and fault prevention of the vacuum consumable arc furnace.

[0007] Further, the calculation of the local image entropy comprises: taking any pixel point in the smoothed image as a center, selecting a local neighborhood window; counting gray values of all pixel points in the local neighborhood window, and calculating probabilities of each gray value appearing in the local neighborhood window; the local image entropy of the pixel point is: , is a probability of the gray value appearing in the local neighborhood window, is a total number of different gray values.

[0008] The above technical scheme effectively quantifies the gray complexity and information content of the local region by constructing a local neighborhood in the smoothed image and calculating the entropy value of the pixel gray distribution, can highlight the regions with rich texture or significant edge changes in the molten pool image, and suppress the interference of flat or noise regions, thereby enhancing the discriminability of the image features.

[0009] Further, the weight value of the weighted fusion ranges from 0 to 1.

[0010] ​Further, the binary image is subjected to a closing operation.

[0011] The technical solution above fills the possible holes in the target region and smoothes the boundary by performing a closing operation on the binary image, i.e., a morphological operation of expansion followed by corrosion, thereby enhancing the connectivity and integrity of the contour.

[0012] Further, the current time The geometric fluctuation degree of the stable contour is: , and are the center point and the area of the stable contour at the current time ; and are the average center point and the average area of the stable contour at the historical multiple times before the current time ; is a preset hyperparameter; is the diagonal line length of the smoothed image.

[0013] The technical solution above quantifies the geometric fluctuation degree as an index that can reflect the spatial displacement and morphological change of the molten pool contour by comprehensively considering the center position offset and area change of the stable contour, which can sensitively capture the dynamic change characteristics of the molten pool at different time points, and the index will significantly increase when the molten pool is affected by the fluctuation of electric arc energy or abnormal melting state, thereby realizing timely identification of the unstable state of the molten pool geometry.

[0014] Further, the adaptive threshold segmentation adopts a local Otsu algorithm.

[0015] Further, the trend analysis method is a double exponential weighted moving average algorithm.

[0016] Further, the target region is a region corresponding to a pixel point whose feature value after correction is greater than a set threshold.

[0017] Further, the infrared thermal image is subjected to noise reduction processing by using a median filter algorithm.

[0018] The technical solution above effectively suppresses isolated noise points in the image caused by thermal radiation fluctuation or environmental interference by using a median filter algorithm to reduce the noise of the infrared thermal image, while maintaining the integrity of the molten pool edge and detail structure.

[0019] Further, the sending of the early warning signal further includes: when the trend values of the multi-scale morphological entropy and the geometric fluctuation degree both exceed the corresponding preset upper limits, a secondary early warning signal is sent.

[0020] The present application has the advantages that:

[0021] The present application realizes high-precision and real-time monitoring of the state of the molten pool by systematic image processing and dynamic feature analysis of infrared thermal images of the vacuum consumable arc furnace molten pool; image quality is ensured by noise reduction processing, stable contours are extracted by gradient and local entropy fusion, the connectivity and accuracy of the contours are effectively enhanced by adaptive threshold segmentation and closing operation processing, and the complexity and dynamic fluctuations of the molten pool shape are quantified by calculation of multi-scale morphological entropy and geometric fluctuation degree, the change trend of the state of the molten pool is captured by trend analysis, and sensitive early warning of abnormal states is realized; the present application can accurately reflect the geometric and thermodynamic characteristics of the molten pool in a complex thermal radiation environment, not only improves the monitoring precision and stability, but also can distinguish between short-term fluctuations and sustained abnormalities, realizes graded early warning, significantly enhances the safety, intelligent level and process control ability of the smelting process, and provides reliable data support and technical support for molten pool state management and fault prevention. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 is a flow chart of an image processing-based vacuum consumable arc furnace molten pool state monitoring method according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0024] An image processing-based vacuum consumable arc furnace molten pool state monitoring method embodiment.

[0025] As Figure 1 shown, the flow chart of the image processing-based vacuum consumable arc furnace molten pool state monitoring method of an embodiment of the present application includes the following steps:

[0026] S1: Obtain infrared thermal images of the vacuum consumable arc furnace molten pool at the current and historical multiple time points, and perform noise reduction processing on the infrared thermal images to obtain smoothed images.

[0027] In a preferred embodiment, the noise reduction processing is performed using a median filter algorithm. Specifically, for the high-frequency noise and random thermal radiation interference present in the obtained infrared thermal images, a suitable window size is first selected, for example or The window is slid in the image pixel by pixel, the gray values in the neighborhood of each pixel point are sorted, and the original gray value of the pixel point is replaced by the median, thereby effectively suppressing isolated noise points and maintaining the structural features of the molten pool edge region.

[0028] For example, when processing the infrared thermal image collected at a certain time, a window size of 3x3 is selected The median filter window is scanned and calculated for the whole image, and the scattered noise in the filtered image due to thermal radiation fluctuation is significantly reduced, and the transition of the molten pool profile edge is smoother, providing high-quality input for subsequent gradient calculation and stable profile extraction.

[0029] S2: Calculate the gradient amplitude and local image entropy of each pixel point in the smoothed image, fuse the two by weighting to correct the gray value, construct a feature map and obtain a binary image by adaptive threshold segmentation, and extract the boundary of the target region as a stable profile.

[0030] In a preferred embodiment, calculating the local image entropy comprises: selecting a local neighborhood window centered on any pixel point in the smoothed image; counting the gray values of all pixel points in the local neighborhood window and calculating the probability of each gray value appearing in the local neighborhood window;

[0031] Pixel point The local image entropy of is: , The gray value The probability of appearing in the local neighborhood window, The total number of different gray values.

[0032] By constructing a local neighborhood window centered on each pixel point in the smoothed image, counting the gray scale distribution in the neighborhood and calculating the local image entropy accordingly, the complexity of the gray scale change around the pixel and the amount of information are reflected. The regions with rich texture and significant edge change in the image have higher local entropy values, while the flat and stable regions have lower entropy values. In this way, the local thermal distribution inhomogeneity and dynamic change characteristics of the vacuum consumable arc furnace molten pool infrared image can be accurately described without relying on global features, providing more discriminative structural information for subsequent gradient fusion and stable profile extraction, thereby improving the sensitivity and stability of the molten pool state monitoring.

[0033] The gradient amplitude and local image entropy are fused by weighting to obtain the corrected gray value of each pixel point, wherein the weight of the weighted fusion is in the range [0, 1], a feature map is constructed based on the corrected gray value, the feature map is segmented by adaptive threshold to obtain a binary image, and the boundary of the target region in the binary image is taken as a stable profile. The target region is the region corresponding to the pixel points whose corrected feature values are greater than a set threshold, and the adaptive threshold segmentation uses a local Otsu algorithm.

[0034] The gradient amplitude and the local image entropy are fused by weighting, so that the pixel gray value reflects the edge strength and the local information complexity, thereby giving consideration to the structural change and the texture feature of the image in the feature construction. The area with obvious heat distribution change in the molten pool image can be effectively enhanced, and the noise or the interference of the smooth area can be suppressed. Then, the feature map is constructed based on the modified gray value, and the adaptive threshold segmentation is adopted, so that the adaptive region extraction under different image brightness and contrast conditions can be realized, and the stability and accuracy of the target region boundary extraction can be ensured.

[0035] Further, the binary image is subjected to a closing operation, and the binary image is further subjected to a closing operation after being obtained. Through the morphological operation of first expansion and then corrosion, the possible holes in the target region can be effectively filled, and the boundary can be smoothed. Under the premise of keeping the overall contour shape unchanged, the boundary breakage and the small gap caused by the infrared image noise or the local gray value fluctuation can be eliminated, so that the connectivity and the integrity of the target region are enhanced. Through the closing operation, the extracted stable contour is more continuous, smoother and has higher structural consistency, so that more accurate data basis can be provided for the subsequent calculation of the molten pool geometric feature and the state trend analysis.

[0036] S3: calculating a multi-scale morphological entropy and a geometric fluctuation degree of the stable contour.

[0037] In a preferred embodiment, the multi-scale morphological entropy is a ratio of a change amount of the stable contour area after the morphological operation at each scale to a total change amount of the stable contour area after the morphological operation at each scale. By performing the morphological operation on the stable contour at different scales, the ratio of the contour area change amount at each scale to the total change amount is calculated to represent the structural complexity and the morphological stability of the stable contour in the multi-scale space. The morphological evolution characteristics of the molten pool contour at different scales can be comprehensively reflected. When the molten pool state is stable, the area change distribution at each scale is balanced, and the ratio change is gentle. When the energy or heat flow distribution in the molten pool appears abnormal fluctuation, the area change at the local scale is significant, so that the ratio distribution is abnormal.

[0038] the current time the geometric fluctuation degree of the stable contour is: , and are the center point and the area of the stable contour at the current time ; and are the average center point and the average area of the stable contour at the historical multiple times before the current time ; is a preset hyperparameter; is a diagonal line length of the smoothed image.

[0039] By comprehensively considering the center position offset and area change of the stable contour at the current moment, a geometric fluctuation degree index is constructed to quantify the dynamic fluctuation of the molten pool contour in the spatial position and the shape scale. The relative deviation of the center point and the area reflects the movement of the geometric center of gravity of the molten pool and the expansion or contraction degree of the shape, respectively, and the balance of the influences of the two is realized through a weighting coefficient, so that the index can capture both the local geometric deformation and the overall thermal dynamic change. When the molten pool is affected by the fluctuation of the electric arc energy or the unstable melting state, the index will significantly increase, so as to be used for timely identifying the unstable state of the molten pool, and improving the real-time monitoring accuracy and the early warning sensitivity of the smelting process.

[0040] S3: The trend values of the multi-scale morphological entropy and the geometric fluctuation degree are obtained by using a trend analysis method, and when any trend value exceeds the corresponding preset upper limit, a first-level early warning signal is sent.

[0041] In a preferred embodiment, the trend analysis method is a double exponential weighted moving average algorithm. The sending of the early warning signal further includes: when the trend values of the multi-scale morphological entropy and the geometric fluctuation degree exceed the corresponding preset upper limits at the same time, a second-level early warning signal is sent.

[0042] The double exponential weighted moving average algorithm is used to analyze the trends of the multi-scale morphological entropy and the geometric fluctuation degree, and through the weighted smoothing processing of the current and historical characteristic values, the dynamic tracking and noise suppression of the change trend of the molten pool state are realized. The stability recognition ability for long-term trends can be enhanced while keeping sensitive to short-term fluctuations, so as to more accurately reflect the continuity and abnormality of the molten pool shape change. When the trend values of the two characteristics exceed the upper threshold at the same time, it is determined that the molten pool is in an obviously abnormal state and a second-level early warning signal is sent, so as to distinguish transient disturbance from persistent instability, and realize higher level safety protection and process control.

[0043] The scheme of the present application realizes high-precision dynamic monitoring of the molten pool state of the vacuum consumable arc furnace by organically combining multiple image processing technologies such as infrared thermal imaging, image noise reduction, median filtering, gradient amplitude and local image entropy fusion, local adaptive threshold segmentation and closing operation processing. The real molten pool contour can be stably extracted under complex thermal radiation interference and noise environment, and the spatial changes of the shape and temperature distribution of the molten pool can be accurately reflected. At the same time, through the calculation and trend analysis of the multi-scale morphological entropy and the geometric fluctuation degree, the stability and fluctuation characteristics of the molten pool at different time scales are quantified, so as to realize early identification and hierarchical early warning of the abnormal state of the molten pool. Overall, the scheme of the present application not only improves the accuracy and robustness of the molten pool contour extraction, but also enhances the sensitivity and early warning ability of the dynamic change of the molten pool, providing reliable technical support for the safety control, production optimization and intelligent management of the smelting process.

[0044] In the description of the present specification, the meaning of "a plurality of", "several" is at least two, such as two, three or more, etc., unless otherwise explicitly specifically limited.

[0045] While the present specification has shown and described a number of embodiments of the present application, it is to be understood that such embodiments are merely illustrative of the many possible embodiments thereof. Numerous modifications, adaptations, and variations, will be apparent to those skilled in the art in view of this description, without departing from the spirit and scope of the present application. It should be understood that various alternatives to the embodiments of the application described herein can be employed in practicing the application.

Claims

1. A method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing, characterized in that, include: Infrared thermographic images of the molten pool of a vacuum self-consuming electric arc furnace at current and multiple historical moments are acquired, and the infrared thermographic images are denoised to obtain smooth images. Based on the smoothed image, a stable contour is extracted. Specifically, the extraction of the stable contour involves: calculating the gradient magnitude and local image entropy of each pixel in the smoothed image; weighting and fusing the gradient magnitude and local image entropy to obtain the corrected gray value of each pixel; constructing a feature map based on the corrected gray value; performing adaptive threshold segmentation on the feature map to obtain a binary image; and using the boundary of the target region in the binary image as the stable contour. Calculating the local image entropy includes: selecting a local neighborhood window centered on any pixel in the smoothed image; counting the grayscale values ​​of all pixels within the local neighborhood window, and calculating the probability of each grayscale value appearing in the local neighborhood window; pixel Local image entropy for , grayscale value The probability of it appearing in the local neighborhood window This represents the total number of different grayscale values ​​that appear. Calculate the multi-scale morphological entropy and geometric fluctuation degree of the stable profile; the multi-scale morphological entropy is the ratio of the change in the area of ​​the stable profile after morphological operations at each scale to the sum of the changes in the area of ​​the stable profile after morphological operations at each scale. Trend analysis is used to obtain the trend values ​​of the multi-scale morphological entropy and geometric fluctuation degree respectively; when any trend value exceeds its corresponding preset upper limit, an early warning signal is issued.

2. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, The weight values ​​for the weighted fusion are in the range of [0,1].

3. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, It also includes performing a closing operation on the binary image.

4. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, Current moment Geometric fluctuation of the lower stable profile for: , and Each represents the current time. The center point and area of ​​a stable profile; and At the current time The average center point and average area of ​​the stable profile at multiple historical moments; These are preset hyperparameters; The length of the diagonal of the smoothed image.

5. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, The adaptive threshold segmentation uses the local Otsu algorithm.

6. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, The trend analysis method is a double exponential weighted moving average algorithm.

7. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, The target region is the region corresponding to pixels whose corrected feature values ​​are greater than a set threshold.

8. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, The infrared thermal image is denoised using a median filtering algorithm.

9. The method for monitoring the state of the molten pool in a vacuum consumable arc furnace based on image processing according to claim 1, characterized in that, The issuance of the warning signal also includes: When the trend values ​​of the multi-scale morphological entropy and geometric fluctuation degree simultaneously exceed their corresponding preset upper limits, a level two warning signal is issued.

Images