A cloud platform-based nondestructive testing and analysis method and system for porosity of refractory materials

Abstract

The application relates to the technical field of material defect detection, in particular to a fire-resistant material porosity nondestructive detection analysis method and system based on a cloud platform, wherein the method provided by the application first screens standard samples meeting quality requirements by using ultrasonic detection and CT scanning technology, then acquires CT scanning data and ultrasonic detection signals of the standard samples, and constructs standard pore characteristic vectors and standard ultrasonic characteristic vectors according to the data, subsequently establishes a standard pore ultrasonic data set according to the standard pore characteristic vectors and the standard ultrasonic characteristic vectors, correlates characteristic data of different sources, establishes a mapping relationship between pore characteristics and ultrasonic characteristics, and thus realizes the prediction of local porosity of a test sample according to a measured ultrasonic characteristic vector. The application combines the advantages of ultrasonic detection and CT scanning technology, calculates the porosity corresponding to the measured ultrasonic signal through the pre-established standard pore ultrasonic data set, and greatly improves the detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of material defect detection technology, specifically to a cloud-based non-destructive testing and analysis method and system for the porosity of refractory materials. Background Technology

[0002] The size and distribution of pores in refractory materials (such as refractory bricks) are key factors affecting thermal conductivity and structural strength. Therefore, the porosity of refractory products is an important indicator for evaluating their quality. Currently, the main methods for measuring the porosity of refractory products include immersion method (Archimedes' drainage method), ultrasonic testing, and X-ray irradiation (industrial CT scanning).

[0003] The immersion method calculates the true porosity of a sample by measuring its dry mass after pulverization and its total mass after immersion. Due to its simple principle, minimal equipment requirements, and generally satisfactory measurement accuracy, it is a basic measurement method recommended by industry standards. However, the measurement process requires pulverizing the sample and immersing it in liquid under vacuum, which is time-consuming and inefficient. Furthermore, it only yields the overall porosity of the sample, failing to obtain the local porosity of the product. This is insufficient for the quality requirements of large-sized refractory products, as local porosity is a significant factor affecting quality.

[0004] Industrial CT scanning is an emerging high-precision scanning technology that uses X-rays to create images by attenuating as they pass through refractory products. The resulting grayscale images are then used to assess the porosity distribution of the product. However, because it cannot directly provide porosity data and relies on image analysis, the process is complex and highly dependent on the operator's technical expertise. Therefore, it cannot meet the requirements of low-cost, high-efficiency inspection methods in daily production processes.

[0005] Ultrasonic testing, as a method that balances accuracy and efficiency, is now widely used in porosity detection. It uses the attenuation and high-frequency echoes produced when ultrasound waves pass through refractory products to determine the presence of large pores, thus assessing the quality of the refractory product. While ultrasonic testing is a more flexible and convenient non-destructive testing method, it only obtains the ultrasonic wave spectrum signal and cannot directly determine the porosity of the refractory product. The entire assessment process still relies heavily on the operator's experience and skill level. Summary of the Invention

[0006] To address the technical problem that existing technologies cannot efficiently and accurately detect the porosity of refractory materials, this application provides a cloud platform-based non-destructive testing and analysis method and system for refractory material porosity, wherein the method includes:

[0007] Standard samples that meet the quality requirements are selected by using ultrasonic testing and CT scanning techniques.

[0008] The standard sample's CT scan data and ultrasonic detection signal are acquired. A standard porosity feature vector is constructed based on the CT scan data, and a standard ultrasonic feature vector is constructed based on the ultrasonic detection signal. Each component of the standard porosity feature vector represents an internal porosity statistical parameter of the standard sample, and each component of the standard ultrasonic feature vector represents a time-domain or frequency-domain feature of the ultrasonic signal.

[0009] A standard stoma ultrasound dataset is constructed based on standard stoma feature vectors and standard ultrasound feature vectors. The standard stoma ultrasound dataset contains multiple standard stoma feature vectors and standard ultrasound feature vectors that correspond one-to-one with each other.

[0010] Obtain the measured ultrasonic feature vector of the sample to be tested, and calculate the local porosity corresponding to the measured ultrasonic feature vector using the standard porosity ultrasonic dataset.

[0011] The cloud-based non-destructive testing system for refractory porosity provided in this application includes:

[0012] An ultrasonic testing device is used to acquire ultrasonic signals from a sample and to obtain ultrasonic feature vectors based on the ultrasonic signals.

[0013] A CT scanning device used to acquire CT images of a sample;

[0014] An image processing device is used to obtain standard pore feature vectors based on CT images of standard samples.

[0015] The data processing device is used to establish a standard stomatal ultrasonic dataset based on the standard stomatal feature vector and the standard ultrasonic feature vector, and to calculate the local porosity corresponding to the measured ultrasonic feature vector based on the standard stomatal ultrasonic dataset.

[0016] Technical effects and advantages of the invention: This application combines the advantages of ultrasound detection and CT scanning technology. It calculates the porosity corresponding to the measured ultrasound signal by using a pre-established standard porosity ultrasound dataset, eliminating the need for a complex CT scan of each test sample and greatly improving detection efficiency. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the detection method provided by the present invention.

[0018] Figure 2 This is a simplified diagram illustrating the principle of ultrasonic testing for refractory materials.

[0019] Figure 3 This is a schematic diagram of the detection system provided by the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the production of refractory materials, porosity measurement plays a crucial role in product quality control. Existing technologies such as immersion testing, ultrasonic testing, and industrial CT scanning have inherent limitations. Immersion testing requires sample pulverization and immersion under vacuum, resulting in a time-consuming process and only providing overall porosity data, failing to reflect the porosity distribution characteristics of localized areas. While industrial CT scanning can generate images of internal structures, its complex operation and high dependence on operator skill make it difficult to meet the demands of low-cost and high-efficiency testing methods in daily production. Ultrasonic testing only acquires the ultrasonic wave spectrum signal, unable to directly quantify porosity values, and its interpretation is significantly influenced by operator experience. Therefore, the problem of efficient and non-destructive testing of local porosity in refractory materials directly affects production efficiency and the accuracy of product quality assessment. The lack of local porosity data, in particular, makes it impossible to meet the requirements for structural strength and thermal conductivity uniformity in larger refractory products.

[0022] For example, on a continuous production line in a large refractory brick manufacturing plant, when processing refractory bricks with dimensions of 800mm × 600mm × 200mm, porosity testing is required in different areas of the product to ensure quality consistency. In practical applications, ultrasonic testing equipment is used to scan the brick surface, but the acquired signals only contain time and frequency domain characteristics. Operators must rely on experience to infer the presence of porosity, leading to deviations in test results between different batches. Simultaneously, to verify local porosity, some samples are extracted for immersion testing, a process that requires interrupting the production line and waiting for several hours, disrupting the production rhythm. In this scenario, the inability to obtain local porosity in real time without damage manifests as production continuity interruption and unreliability in the quality control process. Furthermore, the lack of porosity distribution data makes it difficult for operators to adjust production process parameters in a timely manner, leading to product quality fluctuations and deviations from industry standards.

[0023] To address the technical problem of existing technologies' inefficient and accurate detection of refractory material porosity, this application provides a cloud-based non-destructive testing method for refractory material porosity, referencing... Figure 1 This includes the following steps:

[0024] S1. Standard samples that meet the quality requirements are screened out using ultrasonic testing and CT scanning technology;

[0025] S2. Obtain the CT scan data and ultrasonic detection signal of the standard sample, construct a standard porosity feature vector based on the CT scan data, and construct a standard ultrasonic feature vector based on the ultrasonic detection signal. Each component of the standard porosity feature vector represents an internal porosity statistical parameter of the standard sample, and each component of the standard ultrasonic feature vector represents a time domain or frequency domain feature of the ultrasonic signal.

[0026] S3. Construct a standard stoma ultrasound dataset based on the standard stoma feature vector and the standard ultrasound feature vector. The standard stoma ultrasound dataset contains multiple standard stoma feature vectors and standard ultrasound feature vectors that correspond to each other.

[0027] S4. Obtain the measured ultrasonic feature vector of the sample to be tested, and calculate the local porosity corresponding to the measured ultrasonic feature vector using the standard porosity ultrasonic dataset.

[0028] This application combines the advantages of ultrasound testing and CT scanning technologies. It calculates the porosity corresponding to the measured ultrasound signal by using a pre-established standard porosity ultrasound dataset, eliminating the need for a complex CT scan of each test sample and greatly improving testing efficiency.

[0029] Specifically, in step S1, a batch of refractory material samples can undergo preliminary ultrasonic attenuation testing to exclude samples with obvious macroscopic defects. Simultaneously, these samples can be subjected to low-resolution CT scans to roughly assess their internal structural uniformity. Based on preset quality standards, samples with relatively uniform internal structures and representative pore distribution are manually selected as standard specimens. Ideally, the porosity of these specimens should cover a wide range; that is, the standard specimens should preferably come from refractory products with different porosity requirements. This results in a more diverse standard pore ultrasonic dataset, thereby improving the accuracy of the final local porosity measurement.

[0030] Specifically, in some embodiments, the standard stomatal feature vector in step S2 can be obtained through the following steps:

[0031] S21. Perform a CT scan on the standard sample at the location corresponding to the ultrasonic testing point to obtain a CT cross-sectional image of the section where the preset ultrasonic testing point is located, for example... Figure 2 As shown, the ultrasound detection area is cylindrical. Therefore, a CT scan is performed at the center of the ultrasound detection point to obtain a CT image of the cross section at this location, and the CT cross section image is filtered and denoised.

[0032] S22. Use image recognition methods (such as the big law method and the gray-level histogram valley method) to determine an optimal gray-level threshold to distinguish between the material matrix (bright part) and pores (dark part), segment the boundary between the material matrix and pores, and convert the original CT cross-sectional image into a solid pore binary image. The pixel values ​​in the binary image are only 0 and 1, which are used to distinguish between "solid" and "pore".

[0033] S23. Measure and statistically analyze all pores segmented from the binary image of solid pores, calculate the pore statistics data in the CT cross-section image, and arrange the pore statistics data into a one-dimensional array to construct a standard pore feature vector.

[0034] CT scans typically only produce grayscale X-ray images of a cross-section at a specific location. Different shades of gray in the image represent changes in structure or material. Therefore, the analysis of X-ray images requires the integration of image recognition technology.

[0035] Conventional image processing algorithms can automatically or semi-automatically select a grayscale value in the process of segmenting "solids" and "gases". This grayscale value can effectively distinguish the solid parts and pore parts in CT images. For example, Otsu's method or the maximum entropy thresholding method can automatically determine the threshold by analyzing the statistical characteristics of the image grayscale histogram. There are also methods based on region growing or edge detection, which determine the segmentation boundary by analyzing the similarity or discontinuity between pixels.

[0036] After geometrically segmenting the original CT cross-sectional image, geometric topological analysis methods can be used to measure and statistically analyze stomatal features. For example, connected component analysis can identify independent connected regions (i.e., individual pores) in the image, and measurements can be taken for each connected region to calculate stomatal statistics.

[0037] Total porosity: The percentage of the total area of ​​pores to the image area;

[0038] Average equivalent pore diameter: The average diameter after all pore volumes are equivalent to circles;

[0039] Pore ​​size quantiles, such as D10, D50, and D90, represent the equivalent diameter of pores when the cumulative volume distribution of pores reaches 10%, 50%, and 90%, respectively. These effectively describe the concentration and dispersion of pore sizes.

[0040] The average value of stomatal sphericity: measures how close the shape of a stoma is to an ideal sphere (sphericity = 1 is a perfect sphere).

[0041] By arranging and organizing the above statistical data, an N-dimensional standard stomatal feature vector can be constructed.

[0042] To ensure that the porosity structure reflected by the ultrasonic feature vectors obtained from ultrasonic testing is consistent with the porosity structure reflected by the CT cross-sectional image, the obtained CT cross-sectional image should be located at the preset ultrasonic testing points. Since a standard sample can have multiple ultrasonic testing points, multiple standard ultrasonic feature vectors and corresponding standard porosity feature vectors can be obtained from a single standard sample.

[0043] Furthermore, in the process of CT image processing, although an optimal grayscale threshold can be automatically determined when using image recognition algorithms for segmentation, in practice, the grayscale threshold automatically given by the algorithm often fails to effectively segment small-sized pores when dealing with CT image recognition of some dense refractory materials, resulting in a large error in the pore statistical results.

[0044] In response, this application further proposes a method for determining the optimal grayscale threshold. This method, in the step of obtaining the binary image of solid pores, includes the following specific steps:

[0045] S221. Obtain the true porosity of standard samples using the Archimedes displacement method;

[0046] S222. The cross-sectional CT image is segmented using an initial grayscale threshold to form an initial segmentation binary image;

[0047] S223. Perform morphological operations on the initial segmented binary image and calculate porosity using statistical CT.

[0048] S224. Using the relative error between CT-calculated porosity and true porosity as the optimization target, optimize the initial grayscale threshold. When the relative error between CT-calculated porosity and true porosity is less than the minimum porosity error, output the corresponding grayscale threshold as the optimal grayscale threshold.

[0049] By introducing true porosity as the optimization benchmark and adopting an iterative optimization mechanism, the determination of the grayscale threshold no longer depends on subjective experience, but is automatically adjusted through objective data-driven methods. This ensures the accuracy of CT image segmentation, enabling the optimized segmentation results to more accurately reflect the internal pore structure of refractory materials and providing a high-quality data foundation for the subsequent construction of standard pore feature vectors.

[0050] Archimedes' drainage method is a commonly used method for measuring porosity in the refractory materials industry; the specific process will not be described in detail here.

[0051] Furthermore, considering that the porosity features affecting the ultrasound signal are not a two-dimensional planar structure without thickness, but a three-dimensional structure within an approximately cylindrical body corresponding to the ultrasound detection point, this application provides a method for constructing a three-dimensional CT model using multiple CT cross-sectional images, and then using the three-dimensional CT model to obtain standard porosity feature vectors. Specifically, this includes the following steps:

[0052] S24. Perform omnidirectional CT scans on the standard sample from multiple angles and positions to obtain multiple CT scan images, and filter and reduce noise in the CT scan images.

[0053] S25. Use a three-dimensional image registration algorithm to establish a three-dimensional CT model of a standard sample based on multiple CT scan images;

[0054] S26. Using image recognition methods to determine an optimal grayscale threshold, perform image segmentation on the three-dimensional CT model, and distinguish voxels representing the material matrix (high density) from voxels representing pores or air pores (low density), thereby accurately extracting the internal pore (air pore) structure from the three-dimensional model and generating a three-dimensional pore model. The three-dimensional pore model is a binary model with coordinate values ​​of "solid" and "gas".

[0055] S27. Based on the beam diameter and focal length of the ultrasonic probe, define a cylindrical region of interest in the three-dimensional pore model corresponding to the ultrasonic testing point. This region represents the refractory material characteristics actually sampled at the ultrasonic testing point. Statistically analyze the porosity feature data of each region of interest to form a standard porosity feature vector.

[0056] A 3D image registration algorithm is used to integrate scattered 2D CT images into a 3D CT model of a standard sample, thus fully reflecting the grayscale distribution and geometry inside the sample. Then, image processing methods are used to segment the solid and gaseous parts, resulting in an accurate 3D pore model. The porosity feature data within the region of interest obtained from the 3D pore model effectively reflects the influence of the three-dimensional internal structure on the ultrasonic signal, overcoming the inherent limitations of 2D images in capturing 3D pore distribution and significantly improving the accuracy of local porosity calculation.

[0057] The following is a concrete example. First, for a standard specimen, an industrial microfocus CT scanner can be used for omnidirectional scanning. For example, the specimen is placed on a rotating stage and rotated in 0.5-degree or 1-degree increments within a 180-degree or 360-degree range, acquiring one X-ray projection image at each angle, for a total of 360 or 720 projection images. These projection images are then input into 3D reconstruction software, such as VGStudio MAX or Avizo, using filtered backprojection algorithms or iterative reconstruction algorithms to reconstruct a 3D CT model of the standard specimen from these 2D projection images. This model consists of a series of continuous voxels, each with a corresponding grayscale value. Next, to extract porosity information from the 3D CT model, image processing software, such as MATLAB's image processing toolbox or ImageJ, can be used. First, Otsu's method can be used to automatically calculate a globally optimal grayscale threshold, which can effectively distinguish between solid voxels and porosity voxels in the 3D CT model. Then, the 3D CT model is binarized and segmented according to the threshold. Voxels with gray values ​​below the threshold are marked as pores, and those above the threshold are marked as solids, thus obtaining a 3D pore model. Finally, to correlate with the ultrasound detection points, it is assumed that the ultrasound probe has a beam diameter of 30 mm and a focal length of 100 mm. In the 3D pore model, for each preset ultrasound detection point, a cylindrical region of interest with a diameter of 30 mm and a length of 100 mm can be defined centered on that point. Within this cylindrical region, the number of all pore voxels can be counted, and the total pore volume can be calculated. Simultaneously, connected component analysis can be performed on each individual pore within the region to measure its equivalent diameter, sphericity, and other geometric parameters, and to calculate the mean and standard deviation of these parameters. These statistical data, such as the total pore volume, average pore diameter, and number of pores, together constitute the standard pore feature vector for that detection point.

[0058] Similarly, in the process of image segmentation of 3D CT models, the true porosity measured by the Archimedes displacement method can be introduced to adjust the optimal grayscale threshold, thereby obtaining a more accurate 3D pore model. This specifically includes the following steps:

[0059] S261. Obtain the true porosity of standard samples using the Archimedes displacement method;

[0060] S262. Use an initial grayscale threshold to segment the three-dimensional CT model to form an initial three-dimensional pore model;

[0061] S263. Calculate the true porosity of each region of interest in the initial three-dimensional pore model using CT.

[0062] S264. Using the relative error between CT-calculated porosity and true porosity as the optimization target, the initial grayscale threshold is optimized. When the relative error between CT-calculated porosity and true porosity is less than the minimum porosity error, the corresponding grayscale threshold is output as the optimal grayscale threshold.

[0063] Specifically, in some embodiments, this application obtains ultrasound feature vectors through the following steps:

[0064] S28. For each standard sample, ultrasonic testing is performed at the preset ultrasonic testing points, and the original ultrasonic echo signal is preprocessed.

[0065] S29. Extract time-domain and frequency-domain features from the preprocessed ultrasonic echo signal, including longitudinal wave velocity, transverse wave velocity, first wave amplitude, center frequency, -6dB bandwidth, low-frequency band energy ratio, high-frequency band energy ratio, wavelet energy variance, etc.

[0066] S30. Perform Z-score standardization or max-min normalization on the time-domain and frequency-domain features to make their mean 0 and variance 1, or scale them to the [0,1] interval. Arrange all the standardized features in a fixed order to form a standard ultrasound feature vector.

[0067] Figure 2 The diagram illustrates the principle of ultrasonic testing. During ultrasonic testing, a series of fixed locations are pre-determined and marked on a standard sample as detection points for ultrasonic wave transmission and reception. These points can be planned according to the sample's geometry, material properties, and testing requirements. For example, they can be determined using regular grid division, dense sampling of specific areas, or scanning along a specific path. Their purpose is to ensure that each test is performed at the same or comparable locations, thereby guaranteeing the consistency and repeatability of data acquisition and providing reliable input for subsequent feature extraction and model training.

[0068] After obtaining the standard ultrasonic feature vector, it is stored in the standard porosity ultrasonic dataset along with the corresponding standard stomatal feature vector. Since a standard sample can have multiple ultrasonic testing points, multiple standard ultrasonic feature vectors and corresponding standard stomatal feature vectors can be obtained from a single standard sample. The entire standard stomatal ultrasonic dataset can contain standard stomatal feature vectors and standard ultrasonic feature vectors from multiple different standard samples. Furthermore, the more standard samples involved in the measurement, the more data is available in the standard stomatal ultrasonic dataset, resulting in a more accurate calculated porosity.

[0069] In the process of acquiring standard pore feature vectors and standard ultrasonic feature vectors, since the Archimedes water displacement method requires the standard sample to be crushed, it is necessary to perform ultrasonic testing and CT scanning on the standard sample before proceeding, so as not to affect the acquisition of standard pore feature vectors and standard ultrasonic feature vectors.

[0070] Specifically, in step S4, the steps for obtaining the measured ultrasound feature vector are the same as those for obtaining the standard ultrasound feature vector, and will not be repeated here.

[0071] After obtaining the measured ultrasonic feature vector, the porosity corresponding to the measured ultrasonic feature vector is calculated through the following steps:

[0072] S41. Calculate the similarity between the measured ultrasonic feature vector and each standard ultrasonic feature vector in the standard porosity ultrasonic dataset.

[0073] S42. The local porosity recorded in the standard porosity feature vector corresponding to the standard ultrasonic feature vector with the smallest similarity is the local porosity corresponding to the measured ultrasonic feature vector.

[0074] Similarity is a metric that measures how close two vectors are to each other, quantifying the relationship between abstract feature vectors. Besides Euclidean distance, similarity can be calculated in several other ways. For example, cosine similarity can be used, measuring directional consistency by calculating the cosine of the angle between two vectors, suitable for scenarios with high-dimensional feature vectors and where direction is more important than absolute distance. Alternatively, Manhattan distance (L1 norm) can be used, which calculates the sum of the absolute differences in each dimension of two vectors; its computational complexity is lower, making it suitable for scenarios with high computational efficiency requirements.

[0075] The advantage of similarity-based calculation methods is that they can provide local porosity assessment results for test samples quickly and accurately with a small amount of computation, thereby improving the efficiency of non-destructive testing of refractory porosity. They can also achieve good accuracy when there is a large amount of data in the standard pore ultrasonic dataset. Furthermore, the small amount of data computation is suitable for local deployment, and the application of this invention can be realized without purchasing high-performance data processors.

[0076] However, when the standard pore ultrasound dataset has limited data, the above-mentioned vector similarity-based calculation method may suffer from insufficient accuracy.

[0077] In some embodiments, the local porosity can also be obtained through the following steps:

[0078] S43. Using all the standard ultrasound feature vectors in the dataset as features (X), and using the value of each component in the standard stomatal feature vector corresponding to the standard ultrasound feature vector as the target label (y), a regression model is trained using supervised learning algorithms (such as linear regression, support vector regression SVR, random forest, neural network, etc.). This regression model reflects the functional relationship between the standard ultrasound feature vector and each component in the standard stomatal feature vector.

[0079] S44. Inputting the measured ultrasonic vector into the trained regression model can directly give the predicted stomatal feature vector, which includes the feature of local porosity.

[0080] As a specific implementation method, a multilayer perceptron (MLP) neural network can be used when training the regression model. This neural network can contain an input layer, multiple hidden layers, and an output layer. The input layer receives a standard ultrasound feature vector; for example, if the standard ultrasound feature vector contains 10 time-domain and frequency-domain features (such as peak amplitude, energy, center frequency, attenuation coefficient, etc.), the input layer can have 10 neurons. Each hidden layer can contain tens to hundreds of neurons and uses activation functions such as ReLU (Rectified Linear Unit) to introduce nonlinearity. The output layer corresponds to each component of the standard stomatal feature vector; for example, if it is necessary to predict three stomatal statistical parameters—local porosity, average stomatal diameter, and stomatal density—the output layer can have 3 neurons. During training, mean squared error (MSE) can be used as the loss function, and the Adam optimizer can be used for iterative updates of the model parameters. The training data comes from the aforementioned standard stomatal ultrasound dataset. By dividing the dataset into training, validation, and test sets, the generalization ability of the model can be effectively evaluated, and overfitting can be prevented. Once the model is trained, it can be deployed on cloud platform computing nodes, for example, by encapsulating it as a microservice using Docker containerization technology. When a sample needs to be tested, the raw ultrasonic signal acquired by the ultrasonic testing device is preprocessed and feature extracted to form a measured ultrasonic feature vector. This measured ultrasonic feature vector is sent to the regression model server deployed on the cloud platform via an API interface. Upon receiving the measured ultrasonic feature vector, the model server immediately performs forward propagation calculations and outputs a predicted stomatal feature vector. For example, this predicted stomatal feature vector may contain a value representing the local porosity of the current detection point as X%, another value representing the average stomatal diameter as Y micrometers, and the stomatal density as Z pores / millimeter cubic meter. These prediction results can then be returned to the user interface or stored in a database for further analysis and display.

[0081] Through the above technical solution, this application effectively solves the problems of insufficient accuracy, low efficiency, and inability to fully capture the complex nonlinear relationship between ultrasonic features and porosity characteristics in traditional methods for calculating the local porosity of refractory materials. Specifically, by introducing a supervised learning algorithm to train a regression model, the model can learn from a large amount of standard data and establish a more accurate and complex mapping relationship between ultrasonic signal features and the internal porosity characteristics of the material, thereby significantly improving the prediction accuracy of local porosity. In daily production, simply inputting the measured ultrasonic vector into the trained model allows the model to quickly provide prediction results, greatly improving detection efficiency and reducing reliance on manual experience and complex calculations. This method not only provides quantitative local porosity data, but also allows operators to evaluate the overall internal structure of the entire refractory material product based on the local porosity of multiple ultrasonic testing points, thus providing a more reliable and efficient non-destructive testing method for the quality control and performance evaluation of refractory materials.

[0082] refer to Figure 3 The present invention also provides a cloud platform non-destructive testing and analysis system for refractory material porosity, used to implement the above-mentioned testing method, including:

[0083] An ultrasonic testing device is used to acquire ultrasonic signals from a sample and to obtain ultrasonic feature vectors based on the ultrasonic signals.

[0084] A CT scanning device used to acquire CT image data of standard samples;

[0085] An image processing device is used to obtain standard pore feature vectors based on CT image data of standard samples;

[0086] The data processing device is used to establish a standard stomatal ultrasonic dataset based on the standard stomatal feature vector and the standard ultrasonic feature vector, and to calculate the local porosity corresponding to the measured ultrasonic feature vector based on the standard stomatal ultrasonic dataset.

[0087] Compared with existing technologies, the above system has the following advantages: by combining the ultrasonic feature vectors acquired by the ultrasonic testing device with the CT images acquired by the CT scanning device in a data-driven manner through image processing and data processing devices, a mapping relationship between porosity features and ultrasonic features is established. This solves the problem that existing technologies cannot efficiently and non-destructively detect the local porosity of refractory materials. At the same time, it avoids dependence on operator experience and complex operations, achieving the effect of automated and quantitative detection of local porosity.

[0088] Furthermore, the system may also include a true porosity measuring device for measuring the true porosity of the sample, so as to improve the accuracy of pore segmentation when processing CT image data.

[0089] Specifically, in some embodiments, the steps for obtaining a standard stomatal ultrasound dataset include:

[0090] S100. Use an ultrasonic testing device to perform ultrasonic testing on the preset ultrasonic testing points on the standard sample, and obtain the standard ultrasonic feature vector based on the ultrasonic signal.

[0091] S200. The standard sample is scanned using a CT scanning device to obtain CT scan data at the position corresponding to the ultrasonic detection point. The image processing device establishes a standard pore feature vector based on the CT scan data.

[0092] S300: The data processing device establishes a standard pore ultrasonic dataset based on the standard pore feature vector and the standard ultrasonic feature vector.

[0093] Specifically, in daily applications, the image processing and data processing devices in the aforementioned system can be deployed locally, meaning the image processing and data processing processes are performed by a local workstation, or they can be deployed on a cloud platform. For example, a server cluster can be used to store and process large-scale data, equipped with a database management system and data analysis software. The standard porosity ultrasonic dataset is a structured data collection containing porosity feature vectors and their corresponding ultrasonic feature vectors from multiple standard samples. This dataset can be stored in the form of a relational database, NoSQL database, or file system (such as HDF5, CSV files), and contains unique identifiers for associating porosity features and ultrasonic features. During application, IoT technology is used to upload the ultrasonic feature vectors acquired by the ultrasonic testing device and the CT image data acquired by the CT scanning device to the cloud platform. The cloud platform then performs the local porosity calculation step. The entire system can correspond to multiple production lines or testing units, not only reducing the cost of local deployment but also facilitating the collection of testing data. The data in the standard porosity ultrasonic dataset is continuously supplemented during daily testing, thus enabling the entire system to "get stronger with use."

[0094] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-destructive testing method for the porosity of refractory materials based on a cloud platform, characterized in that, Includes the following steps: Standard samples that meet the quality requirements are selected by using ultrasonic testing and CT scanning techniques. The standard sample's CT scan data and ultrasonic detection signal are acquired. A standard porosity feature vector is constructed based on the CT scan data, and a standard ultrasonic feature vector is constructed based on the ultrasonic detection signal. Each component of the standard porosity feature vector represents an internal porosity statistical parameter of the standard sample, and each component of the standard ultrasonic feature vector represents a time-domain or frequency-domain feature of the ultrasonic signal. A standard stoma ultrasound dataset is constructed based on standard stoma feature vectors and standard ultrasound feature vectors. The standard stoma ultrasound dataset contains multiple standard stoma feature vectors and standard ultrasound feature vectors that correspond one-to-one with each other. Obtain the measured ultrasonic feature vector of the sample to be tested, and calculate the local porosity corresponding to the measured ultrasonic feature vector using the standard porosity ultrasonic dataset.

2. The method according to claim 1, characterized in that, The standard stomatal feature vector is obtained through the following steps: A CT scan is performed on the standard sample to obtain a CT cross-sectional image of the section where the preset ultrasonic testing points are located; The optimal grayscale threshold was determined using image recognition methods, and the original CT cross-sectional image was converted into a binary image of solid pores using a segmentation method. All pores in the binary image of solid pores are measured and statistically analyzed to calculate the pore feature data corresponding to the CT cross-section image, and a standard pore feature vector is constructed.

3. The method according to claim 2, characterized in that, In the process of obtaining the binary image of solid pores, the optimal grayscale threshold is determined through the following steps: The true porosity of standard samples was obtained using the Archimedes displacement method. The CT cross-sectional image is segmented using an initial grayscale threshold to form an initial segmentation binary image; Morphological operations were performed on the initial segmented binary image, and porosity was calculated using statistical CT. The relative error between CT-calculated porosity and true porosity is used as the optimization target. The initial grayscale threshold is optimized. When the relative error between CT-calculated porosity and true porosity is less than the minimum porosity error, the corresponding grayscale threshold is output as the optimal grayscale threshold.

4. The method according to claim 1, characterized in that, The standard stomatal feature vector is obtained through the following steps: A comprehensive CT scan was performed on the standard sample to obtain multiple CT scan images; A three-dimensional CT model of a standard specimen was established based on multiple CT scan images using a three-dimensional image registration algorithm. An optimal grayscale threshold is determined using image recognition methods, and the three-dimensional CT model is segmented to obtain a three-dimensional pore model. Based on the beam diameter and focal length of the ultrasonic probe, a cylindrical region of interest is defined at each position corresponding to the ultrasonic detection point in the three-dimensional pore model. The porosity feature data of each region of interest are statistically analyzed to form a standard porosity feature vector.

5. The method according to claim 4, characterized in that, In the process of obtaining the three-dimensional pore model, the optimal grayscale threshold is determined through the following steps: The true porosity of standard samples was obtained using the Archimedes displacement method. An initial grayscale threshold is used to segment the three-dimensional CT model to form an initial three-dimensional pore model. Calculate the true porosity using CT for each region of interest in the initial 3D pore model. The relative error between CT-calculated porosity and true porosity is used as the optimization target. The initial grayscale threshold is optimized. When the relative error between CT-calculated porosity and true porosity is less than the minimum porosity error, the corresponding grayscale threshold is output as the optimal grayscale threshold.

6. The method according to claim 1, characterized in that, The standard ultrasound feature vector is obtained through the following steps: For each standard sample, ultrasonic testing is performed at preset ultrasonic testing points, and the original ultrasonic echo signal is preprocessed. Time-domain and frequency-domain features were extracted from the preprocessed ultrasonic echo signal; The time-domain and frequency-domain features are normalized and sorted to form ultrasound feature vectors.

7. The method according to claim 1, characterized in that, The local porosity corresponding to the measured ultrasonic feature vector is calculated using the following steps: Calculate the similarity between the measured ultrasonic feature vector and each standard ultrasonic feature vector in the standard porosimeter ultrasonic dataset; The local porosity recorded in the standard porosity feature vector corresponding to the standard ultrasonic feature vector with the lowest similarity is the local porosity corresponding to the measured ultrasonic feature vector.

8. The method according to claim 1, characterized in that, The porosity corresponding to the measured ultrasonic eigenvector is calculated using the following steps: Using all the standard ultrasound feature vectors in the dataset as features, and taking the value of each component in the standard stomatal feature vector corresponding to the standard ultrasound feature vector as the target label, a regression model is trained using a supervised learning algorithm. This regression model reflects the functional relationship between the standard ultrasound feature vector and each component of the standard stomatal feature vector. By inputting the measured ultrasonic vector into the trained regression model, a predicted stomatal feature vector can be directly generated, which records the local porosity.

9. A cloud-based non-destructive testing system for the porosity of refractory materials, used to implement the testing method according to any one of claims 1-8, characterized in that, include: An ultrasonic testing device is used to acquire ultrasonic signals from a sample and to obtain ultrasonic feature vectors based on the ultrasonic signals. A CT scanning device used to acquire CT images of a sample; An image processing device is used to obtain standard pore feature vectors based on CT images of standard samples. The data processing device is used to establish a standard stomatal ultrasonic dataset based on the standard stomatal feature vector and the standard ultrasonic feature vector, and to calculate the local porosity corresponding to the measured ultrasonic feature vector based on the standard stomatal ultrasonic dataset.

10. The system according to claim 9, characterized in that, The standard stomatal ultrasound dataset was obtained through the following steps: An ultrasonic testing device is used to perform ultrasonic testing on preset ultrasonic testing points on a standard sample, and the standard ultrasonic feature vector is obtained based on the ultrasonic signal. A standard sample is scanned using a CT scanning device to obtain CT scan data at the location corresponding to the ultrasonic testing point. An image processing device then establishes a standard pore feature vector based on the CT scan data. The data processing device establishes a standard pore ultrasound dataset based on the standard pore feature vector and the standard ultrasound feature vector.

Images