Double-layer composite quartz crucible surface defect detection method based on machine vision

Abstract

The invention discloses a double-layer composite quartz crucible surface defect detection method based on machine vision, and relates to the technical field of crucible surface defect detection, and the method comprises the following steps: a conveyor belt conveys double-layer composite quartz crucibles to a detection area at a preset speed to ensure that each crucible can move in sequence on a production line; and a detection visual field of a machine vision system is stably entered. According to the method, key features such as the microstructure change index and the crack interval variability index are extracted, so that a machine vision system can identify complex defects more accurately, and misjudgment and missing detection are avoided. According to the method, the detection precision of complex defects is improved by enhancing image processing. And in combination with a machine learning model, the system intelligently evaluates the defect complexity and dynamically adjusts the conveyor belt speed. For simple defects, the high speed is kept, so that the production rhythm is not influenced; for complex defects, the speed of the conveyor belt is reduced, and time is provided for detailed analysis, so that the defect recognition rate is improved, and the product quality is ensured while efficient production is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of crucible surface defect detection, and particularly to a method for detecting surface defects of a double-layer composite quartz crucible based on machine vision. Background Art

[0002] The detection of surface defects of a double-layer composite quartz crucible based on machine vision is a method for automatically identifying and analyzing the defects on the surface of a double-layer composite quartz crucible by using machine vision technology. The double-layer composite quartz crucible is a container commonly used in high-temperature environments, and its surface may have defects such as cracks, bubbles, and scratches due to problems in the manufacturing process or wear and collision during use. The machine vision system performs real-time acquisition and analysis of the crucible surface through a high-speed camera, a light source, and image processing algorithms, and identifies various surface defects. Through image processing techniques such as edge detection, texture analysis, and color difference analysis, the type, size, and location of the defects can be accurately located, and the impact on the performance of the crucible can be evaluated, thereby providing data support for subsequent quality control, repair, or scrapping. The advantage of this method is that it can perform defect detection efficiently, accurately, non-destructively, and automatically, which can greatly improve production efficiency and product quality consistency.

[0003] The prior art has the following deficiencies:

[0004] When detecting the surface defects of a batch of double-layer composite quartz crucibles by machine vision, the crucibles are usually placed on a conveyor belt, and the conveyor belt is used to send the crucibles to the detection area to ensure that each crucible can pass through the field of view of the machine vision system in sequence on the production line, so as to ensure that each crucible is accurately photographed and defect detection is performed. The prior art usually uses a relatively fast conveyor belt speed to quickly transport the crucibles to the detection area to ensure that all crucibles can be detected in a short time in a high-production-line without affecting the overall production rhythm. However, when the surface defects of the double-layer composite quartz crucible are relatively complex, the machine vision system needs more time to process and analyze the images to identify and classify these complex defects. If the conveyor belt continues to operate at a relatively fast speed, serious consequences may occur. Due to insufficient image processing time, the system may make misjudgments. For complex defects (such as cracks with irregular shapes or fine defects), they may be misjudged as normal surfaces, resulting in false negatives; while some normal surface textures or lighting changes may be misjudged as defects, leading to false positives. These misjudgments will not only cause quality problems but also waste production resources, wrongly reject normal crucibles, thereby increasing production costs and affecting the stability of product quality.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, and therefore it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The object of the present invention is to provide a method for detecting surface defects of a double-layer composite quartz crucible based on machine vision. By extracting key features such as the microstructure change index and the crack interval variability index, the machine vision system can more accurately identify complex defects and avoid misjudgment and missed detection. This method improves the detection accuracy of complex defects through enhanced image processing. Combining with a machine learning model, the system intelligently evaluates the complexity of defects and dynamically adjusts the conveyor belt speed. For simple defects, maintain a high speed to not affect the production rhythm; for complex defects, reduce the conveyor belt speed to provide time for detailed analysis, thereby improving the defect recognition rate, ensuring efficient production while guaranteeing product quality, so as to solve the problems in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solution: A method for detecting surface defects of a double-layer composite quartz crucible based on machine vision, comprising the following steps:

[0008] The conveyor belt transports the double-layer composite quartz crucible to the detection area at a preset speed, ensuring that each crucible can move sequentially on the production line and stably enter the detection field of view of the machine vision system;

[0009] When the double-layer composite quartz crucible reaches the detection area, the machine vision system obtains high-resolution image data of the crucible surface in real time through a high-speed camera and an optimized light source setting;

[0010] Preprocess the acquired image information to improve the image quality and enhance the defect features. From the preprocessed image data, extract the key features reflecting the complexity of the defects, and conduct a detailed analysis of the extracted key features under the detection window to quantify the complexity of the surface defects of the current double-layer composite quartz crucible;

[0011] Input the quantified defect features into a pre-trained machine learning model, and intelligently evaluate the complexity of the surface defects of the current double-layer composite quartz crucible through the machine learning model;

[0012] Based on the evaluation results of the machine learning model, classify the current crucible defects into simple defects and complex defects;

[0013] For simple defects, the conveyor belt continues to transport the crucible to the next detection link at a preset speed to ensure that the efficient production rhythm is not affected;

[0014] For complex defects, based on the preset speed, reduce the conveyor belt speed so that the machine vision system has sufficient time for more detailed image processing and defect analysis, thereby improving the recognition rate of complex defects and reducing the risk of misjudgment.

[0015] Preferably, from the preprocessed image data, key features reflecting the complexity of defects are extracted. The extracted features include the change in the microscopic structure of the crucible surface and the change degree of the interval between cracks. Under the detection window, the change in the microscopic structure of the crucible surface and the change degree of the interval between cracks are analyzed, and a microscopic structure change index and a crack interval variability index are respectively generated. The complexity of the surface defects of the current double-layer composite quartz crucible is quantified through the microscopic structure change index and the crack interval variability index.

[0016] Preferably, the specific steps for analyzing the change in the microscopic structure of the crucible surface under the detection window to generate a microscopic structure change index are as follows:

[0017] Image data of the microscopic structure of the surface of the double-layer composite quartz crucible is obtained through a machine vision system. The obtained image data is preprocessed, and a local texture analysis method is applied to extract the key features of the surface microscopic structure. For each local area, a local variability model is used to calculate the change in the microstructure features in the image. The calculation formula is as follows:

[0018]

[0019] , where L local (x, y) is the microscopic structure feature at the position (x, y) in the image, is the gradient of the image, representing the image brightness change rate at the position (x, y), and I(x, y) is the brightness value of the image at the position (x, y);

[0020] Based on the extracted microscopic structure feature L local (x, y), an adaptive image transformation model is used to calculate the change degree of the surface microscopic structure. Through the differential analysis between local features, it is evaluated whether the change in the microscopic structure in the image exceeds the normal range. The calculation expression is as follows:

[0021] ΔΦ mic (x, y) = ∫ R |L local (x, y) - L prev (x, y)| γ dA

[0022] , where R is the detection window range, L prev (x, y) is the local microscopic structure feature value at the point (x, y) in the previous reference image, γ is a regulation factor, dA is the infinitesimal area element of the detection window range R, and ΔΦ mic (x, y) is the change amount of the microscopic structure at the position (x, y);

[0023] Through the change amount of the microscopic structure ΔΦ mic(x, y), combined with the global image data, generates a microstructural change index, and the generation formula is as follows:

[0024]

[0025] , where I mic is the microstructural change index, β is the change sensitivity adjustment factor, and A total is the total area of the image.

[0026] Preferably, the specific steps for analyzing the degree of change in the interval between cracks to generate a crack interval variability index under the detection window are as follows:

[0027] First, extract the crack positions and their corresponding intervals in the image. Through image processing technology, obtain the coordinate positions of the cracks, denoted as P i = {P1, P2, P3,..., P n}), where P i is the center position of the i-th crack, n is the total number of cracks, calculate the distance between adjacent cracks, and the calculation expression is as follows: d i = |P i+1 - P i |, where P i+1 is the center position of the (i + 1)-th crack, that is, the center position of the next crack, and d i is the distance between the i-th crack and the (i + 1)-th crack;

[0028] Analyze the local differences in the crack intervals, adopt a measurement method based on self-similarity and local oscillation, and comprehensively consider the "oscillation degree" of the crack intervals, that is, the amplitude of the crack interval fluctuations. Quantify the variability of the crack intervals through the following formula:

[0029]

[0030] , where d j is the distance between the j-th crack and the (j + 1)-th crack, d j-1 is the distance between the previous pair of cracks, exp(-λ|i - j|) is the exponential decay factor, λ is the decay coefficient, and V i is the local variability value at the i-th crack interval position;

[0031] Combine all local variability values V i to generate a crack interval variability index, and the generation formula is as follows:

[0032]

[0033] , where V k is the local variability value at the k-th crack interval position, and I gapis the crack interval variability index.

[0034] Preferably, the microstructure change index and the crack interval variability index after analysis are input into a pre-trained deep learning model. The defect complexity coefficient is generated by the deep learning model, and the defect complexity coefficient is used to intelligently evaluate the surface defect complexity of the current double-layer composite quartz crucible.

[0035] Preferably, when the defect complexity coefficient generated by intelligently evaluating the surface defect complexity of the current double-layer composite quartz crucible through a pre-trained machine learning model is compared and analyzed with a pre-set defect complexity coefficient reference threshold, the current crucible defects are classified. The classification steps are as follows:

[0036] If the defect complexity coefficient is greater than the pre-set defect complexity coefficient reference threshold, the current crucible defect is classified as a complex defect;

[0037] If the defect complexity coefficient is less than or equal to the pre-set defect complexity coefficient reference threshold, the current crucible defect is classified as a simple defect.

[0038] Preferably, for complex defects, based on a preset speed, the speed of the conveyor belt is reduced to improve the recognition rate of complex defects and reduce the risk of misjudgment. The specific steps are as follows:

[0039] After confirming that the defect is a complex defect, calculate the speed adjustment factor according to the defect complexity to adjust the conveyor belt speed, providing more sufficient time for image analysis. A non-linear function is used to enhance the flexibility and accuracy of the adjustment. The calculation expression is as follows:

[0040]

[0041] , where Adjustment-Factor is the speed adjustment factor, θ is the adjustment coefficient, controlling the amplitude of the speed adjustment factor, D efect is the defect complexity coefficient, D ref is the defect complexity coefficient reference threshold, ω is the non-linear exponential factor, e is the natural base, is the exponential adjustment coefficient;

[0042] According to the calculated speed adjustment factor Adjustment-Factor, adjust the actual speed of the conveyor belt to ensure that the machine vision system has sufficient time to analyze complex defects. The calculation expression is as follows:

[0043]

[0044] , where Adjusted-Speed is the adjusted speed of the conveyor belt, V presetis the preset speed of the conveyor belt, γ speed is the speed adjustment coefficient, which controls the response speed between the defect complexity and the speed change. δ is the exponential amplification coefficient, and η is the minimum response base value, representing the minimum response base value when adjusting the speed.

[0045] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0046] By extracting key features reflecting the complexity of defects, such as the microstructure change index and the crack interval variability index, the machine vision system of the present invention can analyze and identify complex defects on the surface of the double-layer composite quartz crucible more precisely. These key features can describe in detail the fine cracks, irregular flaws or microstructure changes on the surface, thereby helping the system effectively distinguish the normal surface and complex defects during the image processing process. Through this method, the system can better identify complex defects with irregular shapes and blurred boundaries, avoiding misjudgment or missed detection caused by insufficient processing time in traditional detection methods. For example, irregular cracks or tiny flaws can be accurately identified as defects without being misjudged as a normal surface, thereby improving the accuracy of defect detection. In this way, the risks of false positives (misjudging a normal surface as a defect) and false negatives (missing complex defects) are reduced, ensuring that every defect can be accurately detected and guaranteeing the quality of the final product.

[0047] Based on the intelligent evaluation of the machine learning model, the system of the present invention can dynamically adjust the speed of the conveyor belt according to the complexity of the defects. For simple defects, the system maintains a faster conveyor belt speed to ensure that the high-yield production line can complete the detection in a shorter time without affecting the overall production rhythm. For complex defects, the system reduces the speed of the conveyor belt according to the model evaluation results to ensure that the machine vision system has enough time for detailed image processing and accurate defect analysis. This method can fully improve the recognition rate and processing accuracy of complex defects without sacrificing production efficiency. Through this speed adjustment mechanism, the production line can not only maintain efficient operation but also ensure high-quality detection of complex defects, reducing problems such as rework and scrapping caused by missed detection or misjudgment of defects, thereby reducing production costs and optimizing the overall quality control process. This flexible detection scheme not only improves the intelligence level of the production line but also effectively balances the relationship between production speed and detection accuracy. Description of the Drawings

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0049] Figure 1 This is the method flow chart of the surface defect detection method for the double-layer composite quartz crucible based on machine vision in the present invention. Specific embodiments

[0050] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that this disclosure will be more complete and thorough, and will fully convey the concept of the example embodiments to those skilled in the art.

[0051] The present invention provides a surface defect detection method for a double-layer composite quartz crucible based on machine vision as Figure 1 shown, including the following steps:

[0052] The conveyor belt transports the double-layer composite quartz crucible to the detection area at a preset speed, ensuring that each crucible can move sequentially on the production line and stably enter the detection field of view of the machine vision system;

[0053] The conveyor belt transports the double-layer composite quartz crucible to the detection area at a preset speed means that on the production line, the conveyor belt is set at a pre-determined and stable speed for moving the double-layer composite quartz crucible to the detection area of the machine vision system. The setting of this preset speed ensures that each crucible can pass through the detection area orderly and evenly, avoiding crucible overlap or uneven spacing caused by unstable speed. By maintaining a consistent conveying speed, the machine vision system can accurately capture the surface image of each crucible within a predetermined time window, ensuring the clarity and integrity of the image. This not only improves the efficiency and accuracy of the detection process, but also reduces the risk of misjudgment and missed detection caused by fluctuations in the conveying speed. In addition, the preset speed also helps the production line maintain the overall production rhythm, ensuring stable and efficient quality detection even in the case of high production volume, thus guaranteeing the consistency of product quality and the smoothness of the production process.

[0054] The speed setting of the conveyor belt needs to comprehensively consider the production efficiency and the response ability of the detection system to achieve the best balance point. A high-speed conveyor belt helps to improve the throughput of the overall production line, while stable conveying ensures the consistency and accuracy of image acquisition.

[0055] When the double-layer composite quartz crucible reaches the detection area, the machine vision system obtains high-resolution image data of the crucible surface in real time through a high-speed camera and an optimized light source setting;

[0056] The quality of image acquisition is crucial, and it is necessary to ensure uniform illumination and no reflection interference to clearly present the details of the crucible surface. Image acquisition from multiple angles and multiple light sources can capture surface defects more comprehensively and reduce detection errors caused by changes in perspective or illumination.

[0057] Preprocess the acquired image information to improve the image quality and enhance the defect features. From the preprocessed image data, extract the key features reflecting the complexity of the defects, and conduct a detailed analysis of the extracted key features under the detection window to quantify the complexity of the surface defects of the current double-layer composite quartz crucible;

[0058] The preprocessing step is a crucial link to ensure the quality of the image data. First, the denoising process removes the random noise in the image through filtering algorithms (such as Gaussian filtering, median filtering) to enhance the smoothness of the image. Gray normalization standardizes the gray values of the image to reduce the brightness difference caused by light changes. The light compensation technique is used to uniform the light distribution of the image and eliminate the influence of over-bright or over-dark regions. In addition, image preprocessing may also include geometric correction and color correction to ensure that the geometric shape and color information of the image accurately reflect the actual surface state of the crucible. Through these preprocessing steps, the quality of the image is significantly improved, providing reliable data support for subsequent feature extraction and defect recognition.

[0059] From the preprocessed image data, extract the key features reflecting the complexity of the defects. The extracted features include the change in the microscopic structure of the crucible surface and the change degree of the interval between cracks. Under the detection window, analyze the change in the microscopic structure of the crucible surface and the change degree of the interval between cracks, and generate a microscopic structure change index and a crack interval variability index respectively. Quantify the complexity of the surface defects of the current double-layer composite quartz crucible through the microscopic structure change index and the crack interval variability index.

[0060] The abnormal change in the microscopic structure of the crucible surface usually indicates the existence of complex defects on the surface of the current double-layer composite quartz crucible, especially cracks with irregular shapes or fine defects. The abnormal change in the surface microscopic structure may include phenomena such as uneven distribution of grain size, excessive oxidation in local areas, bubble formation, surface porosity, etc., which may be closely related to the existence of complex defects. Especially at the edges of cracks or defects, the microscopic structure features are often different from those of the normal surface material. For example, when a crack propagates along the surface, it may cause local material micro-deformation, forming a complex grain structure or porosity change. This abnormal change in the microscopic structure not only makes the crack morphology more irregular, but also may lead to a more tortuous and uneven crack propagation path. In addition, fine defects are often difficult to identify through conventional macroscopic visual inspection, and their structural changes at the microscopic level may be significantly different from the normal surface, such as fine depressions on the surface, the embedding of bubbles, or irregular fracture textures. These changes in the microscopic structure provide strong evidence for the complexity of the defects and usually require more refined image processing and analysis methods to accurately identify, so as to avoid missed detection or misjudgment. Therefore, the abnormal change in the microscopic structure is one of the key bases for judging the complexity of the defects.

[0061] The specific steps for analyzing the microstructure change on the surface of the crucible under the detection window and generating the microstructure change index are as follows:

[0062] The image data of the microstructure on the surface of the double-layer composite quartz crucible is obtained through a machine vision system. After preprocessing the obtained image data, removing noise, adjusting the contrast, and performing illumination compensation, the local texture analysis method is applied to extract the key features of the surface microstructure. The features mainly include the texture complexity of the local area, grain distribution, porosity, crack distribution, etc. For each local area, the local variability model is used to calculate the change of the microstructure features in the image. The calculation formula is as follows:

[0063]

[0064] , where L local (x, y) is the microstructure feature at the position (x, y) in the image, is the gradient of the image, representing the image brightness change rate at the position (x, y), and I(x, y) is the brightness value of the image at the position (x, y);

[0065] This model can accurately reflect the local changes of the surface microstructure, capture the change degree of the texture and the complexity of the cracks, and provide basic features for the subsequent steps.

[0066] Based on the extracted microstructure feature L local (x, y), the change degree of the surface microstructure is calculated using the adaptive image transformation model. Through the differential analysis between local features, it is evaluated whether the change of the microstructure in the image exceeds the normal range, especially for tiny cracks, defects, or irregular surface textures. The calculation expression is as follows:

[0067] ΔΦ mic (x, y) = ∫ R |L local (x, y) - L prev (x, y)| γ dA

[0068] , where R is the detection window range, L prev (x, y) is the local microstructure feature value of the previous reference image at the point (x, y), γ is the adjustment factor used to non-linearly amplify or reduce the sensitivity of the change amount, dA is the tiny area element of the detection window range R, and ΔΦ mic (x, y) is the microstructure change amount at the position (x, y);

[0069] By calculating the microstructure change amount ΔΦ mic(x, y) can quantify the degree of microstructural changes in the image. A larger change indicates significant microstructural changes on the surface, which may involve complex defects such as crack propagation and flaw generation.

[0070] Through the microstructural change amount ΔΦ mic (x, y), combined with the global image data, a microstructural change index is generated. This index can comprehensively consider the complexity of the surface microstructural changes and finally evaluate the complexity of the current crucible surface defects. The generation formula is as follows:

[0071]

[0072] , where I mic is the microstructural change index, β is the change sensitivity adjustment factor used to non-linearly amplify or weaken the influence of the local change amount on the overall change index, and A total is the total area of the image.

[0073] The role of this index is to comprehensively consider the microstructural changes in all local regions. The larger the index value, the more complex the surface defects are, which may involve cracks with irregular shapes or fine flaws. This index can provide accurate input for the machine learning model to help the system make a clear judgment between complex and simple defects.

[0074] The larger the performance value of the microstructural change index generated after analyzing the microstructural changes on the crucible surface under the detection window, the more significant the microstructural anomalies on the crucible surface are usually, which are usually related to complex defects such as cracks with irregular shapes or fine flaws. For example, the propagation of cracks may lead to the fracture of local grains, an increase in the surface oxidation degree, or the formation of bubbles and pores. These changes will cause large fluctuations in the surface microstructure, thus reflecting the existence of complex defects. On the contrary, if the microstructural change index is small, it indicates that the changes in the surface defects are relatively simple, which may be just small fluctuations in surface smoothness or regular flaws. Such defects are easier to identify by conventional methods and usually do not have a significant impact on the functional performance of the crucible.

[0075] The intervals between cracks show irregular variations, which usually indicate the existence of complex defects on the surface of the current double-layer composite quartz crucible, such as cracks with irregular shapes or minute flaws. This is because the intervals between cracks reflect the non-uniformity and complexity during the defect formation process. Generally speaking, if the intervals between cracks are regular and uniform, it may mean that the generation of defects is driven by a single factor (such as uniform temperature change or stress concentration), and the defects are relatively simple and easy to predict. However, when the intervals between cracks show irregular variations, it usually means that the defect generation process is more complex and may be affected by the interaction of multiple factors, such as local stress concentration, uneven temperature gradient, and inconsistencies in the internal micro-structure of the material. The irregularity of the crack intervals indicates that the generation and propagation of cracks have no fixed pattern, which increases the complexity and unpredictability of the defects. At this time, the defects may be cracks with irregular shapes or minute flaws, and their impacts are more difficult to be easily identified and evaluated by conventional detection means, and may pose greater potential hazards to the structural integrity and high-temperature performance of the crucible. Therefore, the irregularity of the crack intervals is an important sign for judging the complexity of surface defects and can effectively indicate the degree of defect complexity.

[0076] The specific steps for analyzing the degree of change in the intervals between cracks under the detection window to generate the crack interval variability index are as follows:

[0077] First, extract the crack positions and their corresponding intervals in the image. Through image processing techniques (such as edge detection, morphological operations, etc.), obtain the coordinate positions of the cracks, denoted as P i ={P1, P2, P3, ……, P n}, where P i is the center position of the i-th crack, n is the total number of cracks, calculate the distance between adjacent cracks, and the calculation expression is as follows: d i =|P i+1 -P i |, in the formula, P i+1 is the center position of the (i + 1)-th crack, that is, the center position of the next crack, and d i is the distance between the i-th crack and the (i + 1)-th crack;

[0078] The purpose of this step is to ensure the accurate calculation of the interval of each crack by accurately extracting the crack intervals. The crack interval d i reflects the spatial pattern of crack distribution and lays the foundation for subsequent interval difference analysis.

[0079] Analyze the local differences in the crack intervals, adopt a measurement method based on self-similarity and local oscillation, and comprehensively consider the "oscillation degree" of the crack intervals, that is, the amplitude of crack interval fluctuations. Quantify the variability of the crack intervals through the following formula:

[0080]

[0081] , where d j is the distance between the j-th crack and the (j + 1)-th crack, d j-1 is the distance between the previous pair of cracks, exp(-λ|i - j|) is an exponential decay factor used to adjust the contribution of positions far from the i-th crack interval to V i , λ is the decay coefficient that controls the influence of intervals far from i on the calculation of the current crack interval variability, and V i is the local variability value at the i-th crack interval position;

[0082] By calculating the differences and fluctuations of the crack intervals, this step can effectively capture the irregularity and variation trend of the crack intervals. The exponential decay function makes the contribution of cracks in adjacent regions to the interval variability more important, which helps to detect complex defects with large local variations.

[0083] Combining all local variability values V i , a crack interval variability index is generated, and the generation formula is as follows:

[0084]

[0085] , where V k is the local variability value at the k-th crack interval position, and I gap is the crack interval variability index.

[0086] When calculating the global variability, this step considers the weights of each local variability and introduces a distance decay function to adjust larger crack intervals to ensure that the index can more accurately reflect the complexity of the crack intervals.

[0087] The final crack interval variability index I gap combines the local variability and the complexity of the global crack interval distribution. When the crack distribution is irregular, the index value is high, and vice versa. This index can effectively quantify the fluctuations of the crack intervals and reflect the complexity level of the crucible surface defects, especially having good discrimination for cases with irregular shapes or intertwined multiple cracks.

[0088] The larger the performance value of the crack interval variability index generated after analyzing the degree of change in the interval between cracks under the detection window, the more likely it indicates the presence of complex defects on the surface of the current double-layer composite quartz crucible, such as cracks with irregular shapes or fine flaws. The crack interval variability index quantifies the non-uniformity of crack distribution by analyzing the degree of change in the interval between cracks. When the change in crack interval is large, it means that the cracks are irregularly distributed on the crucible surface, which may be caused by various factors, such as uneven stress, inconsistent material microstructure, sudden temperature changes, etc. Such irregular crack propagation patterns usually have higher complexity and are difficult to accurately identify and evaluate by conventional detection methods, thus increasing the potential risk to the structural performance of the crucible. On the contrary, when the crack interval variability index is low, it indicates that the intervals between cracks are relatively uniform and regular, usually indicating simpler defects, such as common surface cracks or flaws, with more regular defect morphologies, which are easy to detect and have less impact on the crucible.

[0089] Input the quantified defect features into a pre-trained machine learning model, and use the machine learning model to intelligently evaluate the complexity of the surface defects of the current double-layer composite quartz crucible;

[0090] Input the analyzed microstructure change index and crack interval variability index into a pre-learned deep learning model, generate a defect complexity coefficient through the deep learning model, and use the defect complexity coefficient to intelligently evaluate the complexity of the surface defects of the current double-layer composite quartz crucible.

[0091] A pre-learned machine learning model refers to a model that uses a large amount of historical data to train machine learning or deep learning algorithms before actual production or application scenarios, so as to obtain a model that can intelligently process and predict input data. This process usually includes steps such as data collection, feature engineering, model training, and validation. By training on a large amount of labeled data, the machine learning model can learn the potential patterns or rules in the data, and then automatically identify its features and make predictions when facing new, unseen input data. In the scenario of surface defect detection of double-layer composite quartz crucibles, the pre-learned model can efficiently and accurately evaluate new data during the detection stage by learning different types of defect features (such as microstructure changes, crack intervals, etc.) and the relationships between these features and defect complexity.

[0092] The dataset used to train the model is usually based on a large amount of historical detection data, which includes different types of defect samples and corresponding complexity labels. By processing this data with deep learning models (such as convolutional neural networks, recurrent neural networks, etc.), the model can extract high-level features that are helpful for defect recognition and complexity assessment, and learn to map these features to complexity coefficients. In practical applications, when the surface defect image of a new crucible is input into the trained model, the model will automatically calculate its microstructure change index and crack interval variability index, and combine these feature values to generate a defect complexity coefficient through existing knowledge. This coefficient can intelligently evaluate the complexity of the current defect, thereby guiding subsequent production decisions and defect repair processes, reducing human intervention, and improving the accuracy and efficiency of detection.

[0093] The machine learning model is not limited here, and any machine learning model that can achieve comprehensive analysis of the microstructure change index I mic and the crack interval variability index I gap to generate a defect complexity coefficient D efect is acceptable. To implement the technical solution of the present invention, the present invention provides a specific implementation method;

[0094] The defect complexity coefficient D efect is generated by the following formula:

[0095]

[0096] , where f1 and f2 are respectively the preset proportionality coefficients of the microstructure change index I mic and the crack interval variability index I gap , and both f1 and f2 are greater than 0.

[0097] The "preset proportionality coefficient" here refers to the numerical coefficient used to balance the weights of different parameters in generating the defect complexity coefficient D efect , that is, f1 and f2 in the formula. Since the microstructure change index I mic and the crack interval variability index I gap may be different in numerical range, dimension, or the degree of influence on defect complexity, it is necessary to weight these two indices through the preset proportionality coefficient to ensure that their contributions in calculating defect complexity are coordinated with each other. The preset proportionality coefficient is usually obtained through statistical learning or optimization methods of a large number of sample data in experiments or actual data analysis, aiming to make the generated defect complexity coefficient more accurately reflect the true complexity of the defect.

[0098] Specifically, the magnitudes of the proportionality coefficients f1 and f2 reflect the importance of each index to the final defect complexity. If f1 is larger, it means that the microstructure change index Imic It has a higher weight in the evaluation of defect complexity and is applicable to scenarios where microstructural changes have a significant impact on defect complexity; while a larger f2 indicates the crack spacing variability index I gap has a greater weight in the evaluation. By reasonably setting f1 and f2, it can better adapt to the complexity evaluation requirements of different types of defects, thereby improving the accuracy and applicability of the model evaluation.

[0099] From the defect complexity coefficient, the larger the performance value of the microstructure change index generated after analyzing the microstructural changes on the surface of the crucible under the detection window, and the larger the performance value of the crack spacing variability index generated after analyzing the degree of change in the spacing between cracks under the detection window, the larger the performance value of the defect complexity coefficient generated when the intelligent evaluation of the surface defect complexity of the current double-layer composite quartz crucible is performed by a pre-trained machine learning model. This indicates that the probability of the surface defects of the current double-layer composite quartz crucible being more complex is greater. Conversely, it indicates that the probability of the surface defects of the current double-layer composite quartz crucible being more complex is smaller.

[0100] Based on the evaluation results of the machine learning model, the current crucible defects are classified into simple defects and complex defects;

[0101] Compare and analyze the defect complexity coefficient generated when the intelligent evaluation of the surface defect complexity of the current double-layer composite quartz crucible is performed by a pre-trained machine learning model with the pre-set reference threshold of the defect complexity coefficient to classify the current crucible defects. The classification steps are as follows:

[0102] If the defect complexity coefficient is greater than the pre-set reference threshold of the defect complexity coefficient, then classify the current crucible defects as complex defects;

[0103] If the defect complexity coefficient is less than or equal to the pre-set reference threshold of the defect complexity coefficient, then classify the current crucible defects as simple defects.

[0104] Simple defects usually refer to those with regular shapes, large sizes, easy to identify, and have little impact on the production process or product performance. These defects have obvious boundaries and uniform characteristics and usually do not have too many complex changes. Complex defects refer to those with irregular shapes, small sizes, blurred boundaries, and may have a more serious impact on the crucible performance. Such defects may present cracks, tiny flaws, internal defects, or abnormal surface textures, etc.

[0105] For simple defects, the conveyor belt continues to transport the crucible to the next detection link at a preset speed to ensure that the efficient production rhythm is not affected;

[0106] For simple defects, the purpose of the conveyor belt continuing to transport the crucible to the next inspection step at a preset speed is to quickly identify and handle these defects that have less impact on product quality while ensuring the efficient operation of the production line. Simple defects usually have obvious characteristics and relatively large sizes, and are easily identified and classified through standard inspection methods of machine vision systems. This approach can maximize production efficiency, avoid wasting time and resources due to over-analyzing simple defects, and ensure that the production rhythm is not affected. At the same time, quickly handling these defects can also free up more inspection resources for dealing with complex defects, thereby optimizing the overall production process and achieving a balance between high output and high-quality control.

[0107] For complex defects, based on the preset speed, the speed of the conveyor belt is reduced so that the machine vision system has sufficient time for more detailed image processing and defect analysis, thereby improving the recognition rate of complex defects and reducing the risk of misjudgment;

[0108] For complex defects, based on the preset speed, the steps to reduce the speed of the conveyor belt to improve the recognition rate of complex defects and reduce the risk of misjudgment are as follows:

[0109] After confirming that the defect is a complex defect, a speed adjustment factor is calculated according to the defect complexity to adjust the conveyor belt speed, providing more sufficient time for image analysis. The calculation of the adjustment factor takes into account the defect complexity coefficient and the system's response ability, and a non-linear function is used to enhance the flexibility and accuracy of the adjustment. The calculation expression is as follows:

[0110]

[0111] , where Adjustment-Factor is the speed adjustment factor, θ is the adjustment coefficient, controlling the amplitude of the speed adjustment factor, that is, the preliminary ratio of the conveyor belt speed adjustment, D efect is the defect complexity coefficient, D ref is the reference threshold of the defect complexity coefficient, ω is the non-linear exponential factor, used to enhance the response sensitivity of the system to changes in defect complexity, especially when the defect complexity far exceeds the reference threshold, how to accelerate or slow down the adjustment of the conveyor belt speed, e is the natural base, is the exponential adjustment coefficient, controlling the exponential acceleration effect of the conveyor belt speed adjustment as the defect complexity increases with the defect complexity coefficient D efect increases;

[0112] This step combines non-linear and exponential adjustment methods to further reduce the speed of the conveyor belt when the defect complexity is high, thereby providing more processing time for the machine vision system and ensuring that complex defects can be analyzed more accurately.

[0113] Adjust the actual speed of the conveyor belt according to the calculated speed adjustment factor Adjustment-Factor to ensure that the machine vision system has sufficient time to analyze complex defects. The adjustment formula introduces dynamic weights and changes in the preset speed of the conveyor belt, taking into account the flexibility of the system and the different response time requirements for complex defects. The calculation expression is as follows:

[0114]

[0115] , where Adjusted_Speed is the adjusted speed of the conveyor belt, used to dynamically adjust the running speed of the conveyor belt in the case of dealing with complex defects, V preset is the preset speed of the conveyor belt, and γ speed is the speed adjustment coefficient, which controls the response speed between defect complexity and speed change. The higher the value, the greater the impact of complex defects on the conveyor belt speed. δ is the exponential amplification coefficient, used to adjust the non-linear amplification effect of the defect complexity difference on the speed adjustment range. η is the minimum response base value, indicating the minimum response base value when adjusting the speed, used to prevent the conveyor belt speed from completely stopping due to too low complex defects.

[0116] Through this step, the conveyor belt speed can be flexibly adjusted according to the complexity of the defects, ensuring that complex defects can be analyzed more precisely, while ensuring the reasonable adjustment and efficient operation of the production rhythm.

[0117] For complex defects, based on the preset speed, the step of reducing the conveyor belt speed mainly aims to ensure that the machine vision system can obtain sufficient time to perform more detailed and comprehensive image processing and analysis on the complex defects on the crucible surface. The role of this step is crucial because complex defects usually have irregular shapes, tiny sizes, and blurred boundaries, and are often difficult to accurately identify through conventional image processing techniques or rapid detection methods. For example, the detection of irregular cracks, tiny defects, or surface micro-cracks may lead to misjudgment or missed judgment due to interference from surface textures or the subtlety of the cracks.

[0118] By reducing the conveyor belt speed, the machine vision system can obtain more time to perform multi-level processing on the images of each crucible. This not only helps to improve the image quality and reduce image distortion caused by motion blur, but also enables the system to use more advanced algorithms for refined defect classification and recognition, enhancing the ability to capture details. Specifically, the slower conveying speed enables the machine vision system to perform high-precision analysis on the images, including more accurate edge detection, texture analysis, calculation of crack complexity, etc., thereby improving the recognition rate of complex defects.

[0119] In addition, reducing the conveyor belt speed can also significantly reduce the risk of misjudgment. For complex defects, rapid processing may result in insufficient extraction of some defect features, making it prone to false negatives (undetected defects) or false positives (misjudging a normal surface as a defect). By slowing down the conveyor speed, the system can analyze image features more deeply, accurately identify complex defects, and ensure the accuracy of the detection results. This approach not only improves the precision of quality control but also reduces the waste of production resources caused by misjudgment, avoiding unnecessary production costs due to the incorrect rejection of normal crucibles. Therefore, by adjusting the conveyor belt speed to adapt to the identification of complex defects, not only is the overall efficiency of the detection system enhanced, but also the high-efficiency operation of the production line and the stability of product quality are ensured.

[0120] Through the extraction of key features reflecting the complexity of defects, such as the microstructure change index and the crack interval variability index, the machine vision system of the present invention can more precisely analyze and identify complex defects on the surface of double-layer composite quartz crucibles. These key features can describe in detail the fine cracks, irregular flaws, or microstructure changes on the surface, thereby helping the system effectively distinguish between normal surfaces and complex defects during the image processing process. By this method, the system can better identify complex defects with irregular shapes and blurred boundaries, avoiding misjudgments or undetected defects caused by insufficient processing time in traditional detection methods. For example, irregular cracks or minute flaws can be accurately identified as defects without being misjudged as normal surfaces, thereby improving the accuracy of defect detection. In this way, the risks of false positives (misjudging a normal surface as a defect) and false negatives (undetected complex defects) are reduced, ensuring that every defect can be accurately detected and guaranteeing the quality of the final product.

[0121] Based on the intelligent evaluation of the machine learning model, the system of the present invention can dynamically adjust the conveyor belt speed according to the complexity of the defects. For simple defects, the system maintains a relatively fast conveyor belt speed to ensure that a high-production-line can complete the detection in a short time without affecting the overall production rhythm. For complex defects, the system reduces the conveyor belt speed according to the model evaluation results to ensure that the machine vision system has sufficient time for detailed image processing and precise defect analysis. This method can fully improve the recognition rate and processing precision of complex defects without sacrificing production efficiency. Through this speed adjustment mechanism, the production line can not only maintain efficient operation but also ensure high-quality detection of complex defects, reducing problems such as rework and scrapping caused by undetected defects or misjudgments, thereby reducing production costs and optimizing the overall quality control process. This flexible detection scheme not only enhances the intelligence level of the production line but also effectively balances the relationship between production speed and detection precision.

[0122] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulations to get a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0123] Only some exemplary embodiments of the present invention have been described above by way of illustration. Without doubt, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0124] It should be noted that in this text, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0125] It should be understood that in various embodiments of the present application, the magnitude of the sequence numbers of the above processes does not mean the order of execution is prior or subsequent. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0126] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this text can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0127] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0128] The unit described as a separation component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing unit, can also be physically present separately for each unit, or two or more units can be integrated in one unit.

[0130] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0131] Only some exemplary embodiments of the present invention have been described above by way of illustration. Without doubt, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

Claims

1. A method for detecting surface defects of a double-layer composite quartz crucible based on machine vision, characterized in that, It includes the following steps: The conveyor belt transports the double-layer composite quartz crucible to the detection area at a preset speed, ensuring that each crucible can move sequentially on the production line and stably enter the detection field of view of the machine vision system; When the double-layer composite quartz crucible reaches the detection area, the machine vision system obtains high-resolution image data of the crucible surface in real time through a high-speed camera and an optimized light source setting; Preprocess the acquired image information to improve the image quality and enhance the defect features. Extract the key features reflecting the complexity of the defects from the preprocessed image data, and conduct a detailed analysis of the extracted key features under the detection window to quantify the complexity of the surface defects of the current double-layer composite quartz crucible; Input the quantified defect features into a pre-trained machine learning model, and use the machine learning model to intelligently evaluate the complexity of the surface defects of the current double-layer composite quartz crucible; Based on the evaluation results of the machine learning model, classify the current crucible defects into simple defects and complex defects; For simple defects, the conveyor belt continues to transport the crucible to the next detection link at a preset speed to ensure that the efficient production rhythm is not affected; For complex defects, based on the preset speed, reduce the speed of the conveyor belt so that the machine vision system has sufficient time for more detailed image processing and defect analysis, thereby improving the recognition rate of complex defects and reducing the risk of misjudgment.

2. The method for detecting surface defects of a double-layer composite quartz crucible based on machine vision according to claim 1, characterized in that From the preprocessed image data, extract the key features reflecting the complexity of the defects. The extracted features include the change in the microscopic structure of the crucible surface and the change in the interval between cracks. Under the detection window, analyze the change in the microscopic structure of the crucible surface and the change in the interval between cracks, and generate a microscopic structure change index and a crack interval variability index respectively. Quantify the complexity of the surface defects of the current double-layer composite quartz crucible through the microscopic structure change index and the crack interval variability index.

3. The method for detecting surface defects of a double-layer composite quartz crucible based on machine vision according to claim 2, wherein, The specific steps for analyzing the change in the microscopic structure of the crucible surface under the detection window to generate a microscopic structure change index are as follows: Obtain the image data of the microscopic structure of the double-layer composite quartz crucible surface through the machine vision system, preprocess the acquired image data, and apply the local texture analysis method to extract the key features of the surface microscopic structure. For each local area, use the local variability model to calculate the change in the microscopic structure features in the image. The calculation formula is as follows: where L local (x, y) is the microstructure feature at the position (x, y) in the image, is the gradient of the image, representing the image brightness change rate at the position (x, y), and I(x, y) is the brightness value of the image at the position (x, y); Based on the extracted microscopic structure feature L local (x, y), the degree of change of the surface microscopic structure is calculated using an adaptive image transformation model. By analyzing the differences between local features, it is evaluated whether the change of the microscopic structure in the image exceeds the normal range. The calculation expression is as follows: Δφ mic (x, y) = ∫ R |L local (x, y) - L prev (x, y)| γ dA where R is the detection window range, L prev (x, y) is the local microstructure feature value of the previous reference image at the point (x, y), γ is the adjustment factor, dA is the infinitesimal area element of the detection window range R, ΔΦ mic (x, y) is the amount of microstructure change at the position (x, y); Through the microscopic structure change amount Δφ mic (x, y), combined with the global image data, generate a microscopic structure change index, and the generation formula is as follows: Where, I mic is the microstructure change index, β is the change sensitivity adjustment factor, and A total is the total area of the image.

4. The method for detecting surface defects of a double-layer composite quartz crucible based on machine vision according to claim 2, characterized in that, The specific steps for analyzing the change in the interval between cracks under the detection window to generate a crack interval variability index are as follows: First, extract the crack positions and their corresponding intervals in the image. Through image processing techniques, obtain the coordinate positions of the cracks, denoted as P i ={P1, P2, P3, ……, P n}, where P i is the center position of the i-th crack, n is the total number of cracks. Calculate the distance between adjacent cracks. The calculation expression is as follows: d i =|P i+1 -P i |. In the formula, P i+1 is the center position of the (i + 1)-th crack, that is, the center position of the next crack, and d i is the distance between the i-th crack and the (i + 1)-th crack; Analyze the local differences in the crack intervals, adopt a measurement method based on self-similarity and local oscillation, and comprehensively consider the "oscillation degree" of the crack intervals, that is, the amplitude of the crack interval fluctuations. Quantify the variability of the crack intervals through the following formula: where d j is the distance between the j-th crack and the (j + 1)-th crack, d j-1 is the distance between the previous pair of cracks, exp(−λ|i − j|) is the exponential decay factor, λ is the decay coefficient, and V i is the local variability value at the i-th crack spacing position; Combining all local variability values V i , a crack spacing variability index is generated, and the generation formula is as follows: where V k is the local variability value at the k-th crack interval position, and I gap is the crack interval variability index.

5. The method for detecting surface defects of a double-layer composite quartz crucible based on machine vision according to claim 2, wherein, Input the analyzed microscopic structure change index and crack interval variability index into a pre-learned deep learning model, generate a defect complexity coefficient through the deep learning model, and use the defect complexity coefficient to intelligently evaluate the complexity of the surface defects of the current double-layer composite quartz crucible.

6. The method for detecting surface defects of a double-layer composite quartz crucible based on machine vision according to claim 5, wherein, When the defect complexity coefficient generated by the pre-trained machine learning model for the intelligent evaluation of the surface defect complexity of the current double-layer composite quartz crucible is compared with the pre-set reference threshold of the defect complexity coefficient, the current crucible defects are classified. The classification steps are as follows: If the defect complexity coefficient is greater than the pre-set reference threshold of the defect complexity coefficient, the current crucible defect is classified as a complex defect; If the defect complexity coefficient is less than or equal to the pre-set reference threshold of the defect complexity coefficient, the current crucible defect is classified as a simple defect.

7. The method for detecting surface defects of a double-layer composite quartz crucible based on machine vision according to claim 6, wherein, For complex defects, based on the pre-set speed, the speed of the conveyor belt is reduced to improve the recognition rate of complex defects and reduce the risk of misjudgment. The specific steps are as follows: After confirming that the defect is a complex defect, calculate the speed adjustment factor according to the defect complexity to adjust the conveyor belt speed, so as to provide more sufficient time for image analysis. A non-linear function is used to enhance the flexibility and accuracy of the adjustment. The calculation expression is as follows: Wherein, Adjustment_Factor is the speed adjustment factor, θ is the adjustment coefficient, controlling the amplitude of the speed adjustment factor, D efect is the defect complexity coefficient, D ref is the reference threshold of the defect complexity coefficient, ω is the non-linear exponential factor, e is the natural base, is the exponential adjustment coefficient; According to the calculated speed adjustment factor Adjustment_Factor, adjust the actual speed of the conveyor belt to ensure that the machine vision system has sufficient time to analyze complex defects. The calculation expression is as follows: Where Adjustment_Speed is the speed of the conveyor belt after adjustment, V preset is the preset speed of the conveyor belt, γ speed is the speed adjustment coefficient, which controls the response speed between the defect complexity and the speed change. δ is the exponential amplification coefficient, and η is the lowest response base value, representing the lowest response base value when adjusting the speed.

Images