A ceramic internal defect detection method, device, equipment and storage medium

Abstract

The application relates to the technical field of ceramic defect detection, and discloses a ceramic internal defect detection method, device, equipment and storage medium, the method comprises the following steps: collecting temperature field time series data of an out-of-furnace ceramic workpiece in a natural cooling process; pre-processing the temperature field time series data to form a temperature field matrix; extracting a temperature gradient amplitude and a cooling rate from the temperature field matrix, and positioning a temperature abnormal area based on a double-threshold condition and a connected region analysis; extracting a multi-dimensional feature vector for representing a defect type from the temperature abnormal area; inputting the multi-dimensional feature vector into a pre-trained support vector machine model to obtain a defect type classification result; and finally outputting a detection result containing a defect type, a position and a severity level. The application effectively solves the problems of single feature extraction, weak defect classification capability and insufficient real-time performance in the prior art, and realizes non-contact, high-precision and high-speed online detection of ceramic internal cracks, bubbles and other defects.

Description

Technical Field

[0001] This invention relates to the field of ceramic defect detection technology, and in particular to a method, apparatus, equipment and storage medium for detecting internal defects in ceramics. Background Technology

[0002] During the sintering and molding process, ceramic products are prone to internal defects such as cracks, bubbles, and inclusions due to factors such as raw materials, processes, and environment. These internal defects can severely degrade the mechanical strength, electrical properties, thermal stability, and service life of ceramic products. Therefore, internal defect detection of ceramic workpieces before they leave the factory is a crucial step in ensuring product quality.

[0003] Traditional methods for detecting internal defects in ceramics mainly include:

[0004] Manual tapping method: Relies on workers' experience to judge by the sound of tapping, which is highly subjective, inefficient, has a high rate of missed detection, and cannot be automated;

[0005] Ultrasonic testing: requires a coupling agent, has requirements on the surface finish and shape of the workpiece, is not effective for workpieces with complex shapes or rough surfaces, and is usually slow.

[0006] X-ray or industrial CT inspection methods: can intuitively present the internal three-dimensional structure and have high detection accuracy, but the equipment is extremely expensive, bulky, poses radiation safety risks, has a slow detection speed, and has high operating and maintenance costs, making it difficult to apply to high-speed production line sites.

[0007] Infrared thermography-based nondestructive testing methods have been widely studied in recent years. The basic principle is that when ceramic workpieces are naturally cooled after exiting a high-temperature kiln, internal defects, due to differences in thermophysical parameters (such as thermal conductivity and heat capacity) between themselves and the base material, affect the surface heat conduction process, thus creating detectable anomalies in the surface temperature field distribution. For example, cracked areas lose heat faster due to the presence of gaps, resulting in a potentially faster cooling rate and a larger temperature gradient in the corresponding surface area; while bubble areas lose heat slowly due to the insulating effect of air, leading to an even slower cooling rate in the corresponding surface area.

[0008] However, existing infrared thermal imaging detection technologies still have significant shortcomings in terms of engineering and practical application, specifically: 1) Limited feature extraction dimensions and sensitivity: Most methods only focus on the absolute temperature difference at a certain moment or a simple static temperature distribution, failing to fully utilize the dynamic information of temperature field evolution over time and spatial gradient information, resulting in insensitivity to minute defects or defects with inconspicuous features; 2) Lack of effective defect classification mechanisms: Existing technologies mostly rely on a binary judgment of "presence or absence of defects," or only use simple threshold segmentation to identify abnormal areas, failing to effectively distinguish the specific type of defect (such as cracks or bubbles), and different types... The defects have varying degrees of impact on product performance, and distinguishing between them is crucial for quality grading and process feedback; 3) Insufficient real-time processing capability, making online application difficult: Infrared thermal imaging generates massive amounts of time-series image data. Traditional software algorithms based on general-purpose processors are slow and cannot meet the requirements of high-speed production lines for real-time detection and response (typically requiring processing cycles within tens of milliseconds); 4) Poor adaptability to industrial environments: The kiln outlet environment is high-temperature and dusty, and ordinary infrared equipment is easily interfered with or even damaged, requiring specialized protection and sampling triggering mechanisms. However, existing research is mostly focused on laboratory environments and lacks systematic design for harsh industrial environments.

[0009] Therefore, there is an urgent need for an online detection technology for internal ceramic defects that can adapt to industrial environments, achieve high-speed and high-precision detection, and intelligently distinguish defect types. Existing technologies still need to be improved and developed. Summary of the Invention

[0010] In order to overcome the shortcomings of the prior art, the present invention aims to provide a method, device, equipment and storage medium for detecting internal defects in ceramics, and to solve the problems of single feature extraction, weak defect classification ability, insufficient real-time performance and poor adaptability to industrial sites in existing methods for detecting internal defects in ceramics.

[0011] The first aspect of this invention provides a method for detecting internal defects in ceramics, comprising the steps of: acquiring time-series data of the surface temperature field of a ceramic workpiece during its natural cooling process after exiting the kiln; performing real-time preprocessing on the acquired time-series temperature field data to obtain a calibrated temperature field matrix; extracting dynamic temperature features from the temperature field matrix, the dynamic temperature features including temperature gradient amplitude and cooling rate; locating temperature anomaly regions on the surface of the ceramic workpiece based on the extracted dynamic temperature features; extracting multidimensional feature vectors from the temperature anomaly regions to characterize defect types; inputting the multidimensional feature vectors into a pre-trained classification model to obtain a defect type classification result corresponding to the temperature anomaly regions; and generating and outputting detection result information containing defect type, location, and severity level based on the defect type classification result and the location information of the temperature anomaly regions.

[0012] Optionally, in a first implementation of the first aspect of the present invention, collecting time-series data of the surface temperature field of a ceramic workpiece during its natural cooling process after exiting the kiln includes the following steps: setting up an infrared imager near the cooling area at the exit of the ceramic kiln in advance; monitoring the workpiece position in real time by an encoder installed on the conveying mechanism of the production line; sending a trigger signal to the infrared imager when the encoder detects that the ceramic workpiece has reached a preset sampling start position; controlling the infrared imager to continuously collect surface infrared images of the workpiece at a sampling frequency of not less than 100Hz and convert them into time-series data of the surface temperature field during the period when the ceramic workpiece passes through the field of view.

[0013] Optionally, in the second implementation of the first aspect of the present invention, the collected temperature field time-series data is preprocessed in real time to obtain a calibrated temperature field matrix, including the following steps: optical distortion correction is performed on each frame of surface temperature field time-series data to eliminate image distortion caused by lens characteristics; outlier cleaning is performed on the corrected surface temperature field time-series data to remove invalid or abrupt temperature points caused by environmental interference; Gaussian filtering is performed on the cleaned surface temperature field time-series data to suppress random noise; and the processed surface temperature field time-series data is organized according to time sequence to form a temperature field matrix T(x, y, t) containing spatial coordinates and time dimensions.

[0014] Optionally, in a third implementation of the first aspect of the present invention, dynamic temperature features are extracted from the temperature field matrix. These dynamic temperature features include temperature gradient magnitude and cooling rate. The steps include: calculating the temperature gradient field in the spatial domain for each time point t in the temperature field matrix T(x, y, t); calculating the temperature change rate of each pixel in the x and y directions using a 3×3 pixel neighborhood difference algorithm; calculating the temperature gradient magnitude G(x, y, t) of each pixel based on the temperature change rate, where the temperature gradient magnitude characterizes the local spatial intensity of the temperature change at that point; setting a time analysis window, calculating the temperature change of each spatial coordinate point (x, y) within the time analysis window; and dividing the temperature change by the duration of the time analysis window to obtain the cooling rate V(x, y) corresponding to that coordinate point.

[0015] Optionally, in a fourth implementation of the first aspect of the present invention, locating the temperature anomaly region on the surface of a ceramic workpiece based on the extracted dynamic temperature features includes the following steps: statistically obtaining the normal temperature gradient amplitude range based on the temperature gradient amplitude data of normal, defect-free samples, and setting a first discrimination threshold; statistically obtaining the normal cooling rate range based on the cooling rate data of normal, defect-free samples, and setting a second discrimination range; traversing each pixel in the temperature field matrix, if its temperature gradient amplitude exceeds the first discrimination threshold and its cooling rate exceeds the second discrimination range, then marking the pixel as a candidate anomaly point; performing connectivity analysis on spatially adjacent candidate anomaly points, and merging them to form a connected temperature anomaly region.

[0016] Optionally, in a fifth implementation of the first aspect of the present invention, the multidimensional feature vector extracted from the temperature anomaly region for characterizing the defect type includes: the maximum temperature gradient value G within the temperature anomaly region. max The deviation rate δ between the average cooling rate in the abnormal temperature region and the average cooling rate in the normal region. avg The physical area S occupied by the temperature anomaly region on the surface of the ceramic workpiece.

[0017] Optionally, in the sixth implementation of the first aspect of the present invention, the pre-trained classification model is a support vector machine model, which is obtained by: collecting a large number of ceramic samples with known defect types, obtaining their temperature field data and extracting multi-dimensional feature vectors to form a labeled sample set; preprocessing and enhancing the labeled sample set, and dividing it into a training set, a validation set and a test set; training the support vector machine model using the training set, and optimizing the model parameters using the validation set; after training, extracting the support vectors and corresponding parameters that determine the classification hyperplane to form a pre-trained classification model.

[0018] A second aspect of the present invention provides a ceramic internal defect detection device, comprising: a data acquisition module for acquiring time-series data of the surface temperature field of a ceramic workpiece during its natural cooling process after exiting the kiln; a preprocessing module for real-time preprocessing of the acquired temperature field time-series data to obtain a calibrated temperature field matrix; a feature extraction module for extracting dynamic temperature features from the temperature field matrix, the dynamic temperature features including temperature gradient amplitude and cooling rate; a positioning module for locating temperature anomaly regions on the surface of the ceramic workpiece based on the extracted dynamic temperature features; a vector extraction module for extracting multidimensional feature vectors characterizing defect types from the temperature anomaly regions; a prediction module for inputting the multidimensional feature vectors into a pre-trained classification model and outputting a defect type classification result corresponding to the temperature anomaly regions; and a result output module for generating and outputting detection result information including defect type, location, and severity level based on the defect type classification result and the location information of the temperature anomaly regions.

[0019] A third aspect of the present invention provides an electronic device comprising: a memory and at least one processor, the memory storing computer-readable instructions, the memory and the at least one processor being interconnected via a circuit; the at least one processor invokes the computer-readable instructions in the memory to cause the electronic device to perform various steps of the ceramic internal defect detection method as described above.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium storing computer-readable instructions that, when executed on a computer, cause the computer to perform the steps of the ceramic internal defect detection method as described above.

[0021] Beneficial effects: Compared with existing technologies, the ceramic internal defect detection method provided by this invention breaks through the limitation of relying solely on a single static temperature value by calculating the dual dynamic temperature features of temperature gradient amplitude (spatial dimension) and cooling rate (temporal dimension). It can more sensitively capture abnormal heat conduction phenomena caused by internal defects, significantly improving the detection rate of minute defects. Furthermore, this invention establishes a precise mapping relationship between a multi-dimensional feature vector of maximum temperature gradient, average cooling rate deviation rate, and abnormal region area and specific defect types such as cracks and bubbles by constructing and training a support vector machine model. This achieves automated and intelligent differentiation of defect types and solves the problem of weak classification ability of traditional methods. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0024] Figure 1 A flowchart of a ceramic internal defect detection method provided in an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of the ceramic internal defect detection device provided by the present invention.

[0026] Figure 3 This is a schematic diagram of the electronic device structure provided by the present invention. Detailed Implementation

[0027] This invention provides a method, apparatus, device, and storage medium for detecting internal defects in ceramics. The terms "first," "second," "third," "fourth," etc. (if applicable) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0028] Please see Figure 1 , Figure 1 A flowchart of a ceramic internal defect detection method provided by the present invention is shown in the figure, which includes the following steps:

[0029] S10. Collect time-series data of surface temperature field of ceramic workpieces during natural cooling process after exiting the kiln;

[0030] In this embodiment, this step aims to obtain complete spatiotemporal information on the surface temperature change of the high-temperature ceramic workpiece during the cooling process. This is the data source for all subsequent analyses, and its completeness, synchronicity, and accuracy directly determine the final detection effect.

[0031] To obtain complete and accurate time-series data of the surface temperature field, the specific steps include:

[0032] S11. An infrared imager is pre-installed near the cooling area at the outlet of the ceramic kiln;

[0033] S12. The workpiece position is monitored in real time by an encoder installed on the production line conveyor mechanism;

[0034] S13. When the encoder detects that the ceramic workpiece has reached the preset sampling start position, it sends a trigger signal to the infrared imager.

[0035] S14. Control the infrared imager to continuously acquire infrared images of the surface of the ceramic workpiece at a sampling frequency of not less than 100Hz during the period when the workpiece passes through the field of view and convert them into time-series data of the surface temperature field.

[0036] Specifically, the hardware deployment in this embodiment is as follows: A customized high-resolution infrared imager is installed at an appropriate location behind the ceramic kiln outlet, where the workpiece enters the natural cooling zone. This infrared imager is mounted on a vibration-resistant bracket with an IP65 protection rating and is equipped with a telephoto lens to ensure clear coverage of the entire surface of a single workpiece from a safe distance. The infrared imager is typically equipped with an automatic lens cleaning device driven by compressed air and a water-cooling jacket for equipment cooling to cope with the high-temperature dust environment. For example, a customized high-resolution infrared imager with a resolution of 640×480 and a pixel pitch of 17μm is used. It is mounted on a 6061 aluminum alloy extension bracket (bracket load capacity ≥5kg, vibration resistance rating IP65) 1.5-2m outside the kiln outlet cooling zone and equipped with a 50mm telephoto lens (focal length f=50mm, field of view 12°×9°), capable of completely covering round ceramic workpieces with a diameter ≤50cm or square ceramic workpieces with a diameter of 50cm×50cm.

[0037] Furthermore, to achieve strict synchronization between data acquisition and production line cycle time, and to ensure that the data collected is from the effective workpiece area, an encoder is introduced as a synchronization trigger. For example, the encoder is installed on the shaft of the production line conveyor roller. The system is pre-calibrated as follows: the number of encoder pulses corresponding to when the front end of the ceramic workpiece enters the left edge of the infrared field of view is the workpiece entry threshold; the number of encoder pulses corresponding to when the center of the ceramic workpiece aligns with the center of the field of view is the sampling start threshold.

[0038] As the workpiece moves along the conveyor line, the encoder continuously counts. When the count reaches the sampling start threshold, the encoder sends a trigger signal to the infrared imager via a digital I / O interface, ensuring that each frame of data corresponds to the valid area of ​​the workpiece. Upon receiving the trigger signal, the infrared imager immediately starts, continuously acquiring infrared thermal images of the workpiece at a fixed high frequency (e.g., 100Hz, i.e., one frame every 10 milliseconds), and converts them into temperature field data (each frame is a snapshot of the surface temperature field). This process continues until the workpiece completely moves out of the field of view. The acquired multiple frames of temperature field data constitute the time-series data of the workpiece's surface temperature field. The acquired raw temperature data is transmitted in real time to the subsequent data processing unit via high-speed interfaces such as Gigabit Ethernet. For example, the infrared imager's sampling frequency is set to 100Hz (that is, 1 frame is collected every 10ms), the temperature measurement range is 100-1000℃ (matching the ceramic cooling temperature range of 150-800℃ at the kiln outlet), the temperature measurement accuracy is ±2℃ (capturing the minute temperature difference of 3-5℃ between defective and normal areas), and 16-bit raw temperature data is transmitted via Ethernet, with a single frame data size of 614.4KB (640×480×16bit), and the transmission delay is ≤1ms (far lower than the sampling period, avoiding data accumulation).

[0039] In this embodiment, the above steps achieve precise data acquisition at fixed points, times, and frequencies. Encoder triggering avoids empty or biased acquisition, ensuring that each frame of data corresponds to a valid workpiece area. The high sampling frequency captures the details of rapid temperature changes during the cooling process, laying the foundation for extracting dynamic features. Non-contact measurement avoids damage or interference to the workpiece under inspection.

[0040] S20. Perform real-time preprocessing on the collected temperature field time series data to obtain the calibrated temperature field matrix;

[0041] Because raw temperature field time-series data typically contains various types of noise and distortion, such as image geometric distortion caused by lens optical characteristics, anomalous temperature points (flying points) caused by environmental dust or reflections, and random noise from the sensor itself, this step aims to clean and calibrate the raw temperature field time-series data to provide a high-quality and reliable temperature field data foundation for subsequent accurate analysis. This specifically includes the following steps:

[0042] S21. Perform optical distortion correction on each frame of surface temperature field time-series data to eliminate image distortion caused by lens characteristics.

[0043] S22. Perform outlier cleaning on the corrected surface temperature field time series data, and remove invalid or abrupt temperature points caused by environmental interference.

[0044] S23. Perform Gaussian filtering smoothing on the time series data of the surface temperature field after cleaning to suppress random noise;

[0045] S24. Organize the processed surface temperature field time series data into a temperature field matrix T(x, y, t) containing spatial coordinates and time dimension, where x and y are the coordinates of the original temperature image on the surface of the ceramic workpiece, and t is the cooling time.

[0046] Specifically, infrared lenses, especially wide-angle or some telephoto lenses, may introduce barrel or pincushion distortion, causing a non-linear shift between the pixel positions at the image edges and their actual physical positions. This affects the accuracy of subsequent gradient calculations. The correction process typically uses pre-calibrated lens distortion parameters (k1, k2, p1, p2, etc.) to perform a reverse mapping calculation on the pixel coordinates of each frame of the surface temperature field time-series data, correcting them to the ideal image plane coordinates. The formula can be simplified to: Where (x,y) are the original temperature image coordinates, (x′,y′) are the corrected coordinates, and r 2 =x 2 +y 2 After correction, the geometry of objects in the image is closer to the true proportions.

[0047] In industrial settings, drifting hot dust and momentary reflections from workpiece surfaces can create abnormally bright or dark spots on temperature maps that deviate significantly from the surrounding temperature. These spots do not represent the true surface temperature of the workpiece and must be removed. Common methods for cleaning outliers include:

[0048] Median filtering: By using a sliding window (e.g., 3×3) to traverse the temperature image and replacing the center pixel value with the median of the temperature values ​​within the window, isolated speckle noise can be effectively removed;

[0049] Thresholding method based on neighborhood statistics: Calculate the mean and standard deviation of the temperature of each pixel's surrounding neighborhood (e.g., 5×5). If the difference between the pixel value and the mean exceeds a certain multiple (e.g., 3 times) of the standard deviation, it is identified as an outlier and replaced with the neighborhood mean or interpolation.

[0050] To suppress broader, smaller-amplitude random noise (such as sensor thermal noise), Gaussian filtering is used for smoothing. Gaussian filtering is a linear smoothing filter whose kernel function value follows a two-dimensional Gaussian distribution, with pixels closer to the center having a higher weight. For a pixel (i,j), the filtered value T′(i,j) is calculated using the following formula: In this context, G(m,n) is the weight of the Gaussian kernel function at (m,n), and k is the half-width of the kernel. Gaussian filtering can smooth noise while preserving the edge information (temperature change regions) of the image, which is crucial for subsequent gradient calculation.

[0051] Each frame of temperature image, after the above preprocessing and arranged in chronological order, is logically organized into a three-dimensional matrix T(x,y,t), where the coordinates x and y serve as spatial indices (row and column numbers) corresponding to the two-dimensional coordinates of the ceramic surface temperature image; t serves as a time index (frame number) corresponding to different times after the sampling begins. This matrix completely depicts the temperature change process of each point on the workpiece surface over time.

[0052] After processing through the steps in this embodiment, the systematic errors (distortions) and random interferences (noise, outliers) in the original temperature field time series data are effectively suppressed or eliminated, resulting in a temperature field matrix with high signal-to-noise ratio and accurate geometric relationships.

[0053] S30. Extract dynamic temperature features from the temperature field matrix, wherein the dynamic temperature features include temperature gradient magnitude and cooling rate;

[0054] The core of this step is to extract key information that reveals internal defects from the preprocessed data map (temperature field matrix). Static temperature values ​​are easily affected by factors such as the initial temperature uniformity of the workpiece and environmental radiation, while dynamic temperature characteristics such as the drasticness of spatial changes (gradient) and the rate of change over time (rate) more fundamentally reflect the changes in heat conduction paths caused by internal defects, thus indicating the presence of defects more stably and sensitively. Therefore, this embodiment mainly extracts two types of dynamic temperature characteristics from the temperature field matrix: temperature gradient amplitude and cooling rate. The specific steps include:

[0055] S31. For each time point t in the temperature field matrix T(x, y, t), calculate its temperature gradient field in the spatial domain;

[0056] S32. Using a 3×3 pixel neighborhood difference algorithm, calculate the temperature change rate of each pixel in the x and y directions;

[0057] S33. Calculate the temperature gradient magnitude G(x, y, t) of each pixel based on the temperature change rate. The temperature gradient magnitude represents the local spatial severity of the temperature change at that point.

[0058] S34. Set a time analysis window, and for each spatial coordinate point (x, y) in the temperature field matrix, calculate the temperature change within the time analysis window;

[0059] S35. Divide the temperature change by the duration of the time analysis window to obtain the cooling rate V(x, y) corresponding to the coordinate point.

[0060] Specifically, the temperature gradient vector represents the direction and magnitude of the fastest spatial temperature change. Its magnitude, G(x,y,t), reflects the drastic temperature change in a local area. Internal defects (such as cracks) can alter the direction of local heat flow, leading to abnormally steep temperature change zones on the surface. For a temperature field matrix T(x,y,t) at a given frame in time t, the temperature gradient components of pixel (i,j) in the x and y directions are commonly calculated using the Sobel operator or the central difference method. For example, using the central difference method with a 3×3 neighborhood:

[0061] Δx and Δy are the actual physical dimensions of a single pixel on the ceramic surface (which need to be obtained through system calibration, for example, 0.1 mm / pixel).

[0062] For example, in areas with normal and uniform cooling, the temperature gradient is small and gentle (e.g., 1-2℃ / mm). On the surface directly above the crack, due to the rapid longitudinal dissipation of heat along the crack and the obstruction of lateral heat conduction, a narrow low-temperature zone is easily formed, with a large temperature difference on both sides, resulting in a significantly increased calculated gradient amplitude (up to 5-8℃ / mm). In the bubble region, due to its strong thermal insulation, the gradient amplitude is 3-4℃ / mm, slightly higher than the normal region but lower than the crack region, providing spatial feature basis for subsequent classification; by extracting the gradient amplitude, these potential linear defect features can be highlighted from the background.

[0063] In this embodiment, the cooling rate V(x,y) reflects how fast the temperature at a certain point decreases over time. Internal defects, due to their different thermal inertia or thermal resistance compared to the substrate, will affect the cooling rate of their upper surface.

[0064] The cooling rate is a time-dimensional feature, requiring the use of multiple frames of data. To obtain stable values, a time window is typically selected (e.g., an analysis window of 90 seconds, from the 10th to the 100th second after the workpiece enters the field of view); for a point (i,j) on the surface, its corresponding time series T in the temperature field matrix is ​​found. ij (t); Perform linear fitting on the time series within a selected window (or directly calculate the temperature difference between the start and end points) to obtain the rate of temperature change over time, i.e., the cooling rate: The unit is usually ℃ / s. For example, if the analysis window for the cooling stage is set to 90s, the formula for calculating the rate of temperature change is: .

[0065] The cooling rate in the normal area is about 8-10℃ / s; the cooling rate in the bubble area is about 5-7℃ / s (lower than the normal area) due to its strong thermal insulation; and the cooling rate in the crack area is about 11-13℃ / s (higher than the normal area) due to the rapid dissipation of heat. The difference in the rate over time further enhances the defect characteristics.

[0066] For example, in bubble regions, the insulating effect of the internal air hinders heat transfer from the workpiece interior to the surface, resulting in a slower cooling rate on the surface directly above the bubble region compared to normal regions (e.g., 9°C / s for normal regions, 6°C / s for bubble regions). Conversely, crack regions, due to the additional heat dissipation channels they provide, may experience a faster cooling rate on the surface above them than normal regions (e.g., 11°C / s). Therefore, cooling rate characteristics provide a strong temporal dimension for distinguishing between cracks and bubbles.

[0067] This embodiment extracts two key dynamic features—temperature gradient amplitude and cooling rate—from both spatiotemporal dimensions. These features characterize the defect from the perspectives of spatial abruptness and temporal slowness / rapidity, respectively, forming a more comprehensive and profound description of the defect and providing double insurance for subsequent accurate positioning and classification.

[0068] S40. Locate temperature anomaly areas on the surface of ceramic workpieces based on extracted dynamic temperature features;

[0069] After calculating the gradient magnitude and cooling rate of each pixel across the entire field, the goal of this step is to automatically identify suspicious regions where feature values ​​significantly deviate from the normal range, i.e., temperature anomaly regions. This is a crucial step in converting a continuous feature field into discrete, candidate targets to be classified, and it specifically includes:

[0070] S41. Based on the temperature gradient amplitude data of normal defect-free samples, statistically obtain the normal temperature gradient amplitude range and set a first discrimination threshold.

[0071] S42. Based on the cooling rate data of normal, defect-free samples, statistically obtain the normal cooling rate range and set a second discrimination range;

[0072] S43. Traverse each pixel in the temperature field matrix. If its temperature gradient magnitude exceeds the first discrimination threshold and its cooling rate exceeds the second discrimination range, then mark the pixel as a candidate anomaly.

[0073] S44. Perform connectivity analysis on spatially adjacent candidate anomaly points and merge them to form a connected temperature anomaly region.

[0074] Specifically, before conducting online inspection, a large number of qualified ceramic workpiece samples confirmed to be defect-free are used. Steps S10-S30 are repeated to calculate the temperature gradient amplitude and cooling rate data throughout the entire cooling process. Statistical analysis is then performed on this massive amount of normal data. Assuming that the calculated temperature gradient amplitude in the normal region conforms to a normal distribution, its mean μG and standard deviation σG are taken. According to statistical principles, the vast majority (approximately 99.7%) of normal data will fall within [μG]. Within the range of [3σG, μG+3σG]. Therefore, the first discrimination threshold X can be set to μG+nσG (for example, n=3, i.e., X=μG+3σG). Points with temperature gradient amplitudes exceeding this threshold are considered to have excessively drastic spatial changes and are suspected of being abnormal.

[0075] Similarly, by analyzing the cooling rate of the normal region, we obtain its mean μV and standard deviation σV, and set a second discrimination range Y, for example, Y=[μV]. Points where the cooling rate falls outside the range of 3σV,μV+3σV are considered to have an abnormal cooling rate and are suspected of being abnormal.

[0076] As an example, assuming statistical analysis yields μG = 1.5℃ / mm and σG = 0.5℃ / mm, then X = 1.5 + 3 0.5 = 3.0℃ / mm. Assuming μV = 9.0℃ / s and σV = 0.8℃ / s, then Y = [9.0...]. 2.4,9.0+2.4]=[6.6,11.4]℃ / s.

[0077] For the current ceramic workpiece under test, iterate through each pixel (i,j) in its temperature field matrix and check the gradient magnitude G(i,j,t) corresponding to that point. mid (The value can be taken as the midpoint or maximum value of the analysis window) and the cooling rate V(i,j); if the point simultaneously satisfies: condition A: G(i,j,t) mid X > 0 and condition B: V(i,j) If Y is selected, the pixel is marked as a candidate outlier.

[0078] The advantage of using a dual-condition judgment logic in this embodiment is that: using the temperature gradient threshold alone may misjudge some edge areas or normal abrupt changes caused by surface roughness as abnormalities; using the cooling rate range alone may be insensitive to some minor defects. Using "AND" logic for dual-condition filtering can significantly improve the accuracy of positioning and reduce false alarms, retaining only those points that show significant anomalies in both spatial and temporal dimensions.

[0079] Furthermore, the previous step yielded a binary image (outliers are represented by 1, and normal points by 0), where outliers may be discrete. This embodiment employs a connected component analysis algorithm to connect spatially adjacent (usually using the 8-neighborhood connectivity criterion) candidate outliers, merging them into independent connected regions. To filter out small-area pseudo-outliers that may be caused by noise, an area threshold can be set (e.g., the physical area is less than the area corresponding to 5 pixels), eliminating those connected regions with excessively small areas. Each remaining connected region is defined as a temperature anomaly region on the surface of the workpiece, which is also a potential defect target that needs further analysis and classification in subsequent steps.

[0080] This embodiment achieves automated and highly reliable preliminary location of potential defect areas from complex temperature fields through statistically based adaptive threshold setting and rigorous dual-condition screening logic. It effectively suppresses misjudgments caused by noise interference and local non-uniformity, providing clear and accurate target input for the next stage of fine classification.

[0081] S50. Extract a multi-dimensional feature vector from the temperature anomaly region to characterize the defect type;

[0082] In this embodiment, after locating the abnormal region, it is necessary to further profile it and extract refined features that can distinguish different types of defects such as cracks and bubbles. This step abstracts each abnormal region into a multi-dimensional feature vector, which will serve as the input fingerprint for the intelligent classification model. In this embodiment, the multi-dimensional feature vector extracted from the temperature anomaly region to characterize the defect type includes: the maximum temperature gradient value G within the temperature anomaly region. max The deviation rate δ between the average cooling rate in the abnormal temperature region and the average cooling rate in the normal region. avg The physical area S occupied by the temperature anomaly region on the surface of the ceramic workpiece.

[0083] Specifically, for each temperature anomaly region identified in the previous step, the following three core features are calculated to form a three-dimensional feature vector:

[0084] Feature 1: Maximum temperature gradient value G within the region max The calculation method is as follows: traverse the temperature gradient magnitude G(x,y,t) of all pixels in the abnormal region in all relevant frames, and find the maximum value among them. This feature captures the extreme degree of spatial temperature change in the region.

[0085] For crack defects: they typically manifest as a narrow, linear region with a very strong temperature contrast at its edges, therefore G max The value is very high (e.g., 6-8℃ / mm);

[0086] For bubble defects: it is usually an approximately circular region with a relatively gentle temperature transition at its edge, therefore G max The value is moderately high (e.g., 2-4℃ / mm).

[0087] Normal region: G max Very low (e.g., less than 1°C / mm);

[0088] Therefore, the maximum temperature gradient value G max It can be considered one of the strongest features of dividing cracks and bubbles.

[0089] Feature 2: Average cooling rate deviation rate δ within the region avg The calculation method is as follows: First, calculate the average value V(x,y) of the cooling rate of all pixels in the abnormal region. region Then, obtain the average cooling rate μV of the normal region calibrated in step S40; finally, calculate the deviation rate according to the formula: This characteristic reflects the degree and direction of deviation of the overall cooling rate of the region from the normal level.

[0090] Regarding bubble defects: thermal insulation leads to slow cooling, V region <μV, therefore δ avg For larger negative values ​​(e.g., -20% to -30%);

[0091] For crack defects: rapid heat dissipation leads to rapid cooling, V region >μV, therefore δ avg For larger positive values ​​(e.g., +20% ~ +40%).

[0092] Therefore, the average cooling rate deviation δ avg It not only provides numerical differences, but also provides a clear physical indication through the positive and negative signs, which is another very strong distinguishing feature.

[0093] Feature 3: The physical area S occupied by the temperature anomaly region on the surface of the ceramic workpiece is calculated as follows: count the total number of pixels N contained in the connected region. pixels According to the physical area A corresponding to a single pixel as defined by the system pixel (For example, 0.01 mm²), calculate the physical area: S = N pixels ×A pixelS Under similar initiation causes (such as small cracks caused by the same stress vs. small bubbles of the same volume), different defects exhibit different anomalous region areas in the surface temperature field.

[0094] For bubble defects: heat forms an approximately circular heat island or cold island above the bubble, and the affected area is relatively diffuse and large.

[0095] For crack defects: heat is dissipated along the crack gap, and the affected area is concentrated near the line and the area is small.

[0096] For defects of similar size (crack length, bubble diameter), the area S of the abnormal region caused by the bubble is usually 3 to 5 times that of the crack. Therefore, the physical area S occupied by the temperature abnormal region on the surface of the ceramic workpiece provides indirect information about the geometry of the defect and can serve as an effective supplement to the first two features, especially important in distinguishing between cracks and bubbles of similar size.

[0097] Through the steps of this embodiment, each blurred abnormal region is precisely quantized into a three-dimensional feature vector [G]. max ,δ avg The vector [,S] provides a unique and quantifiable identity description for defect types from three orthogonal dimensions: local maximum mutation intensity, overall temporal evolution deviation, and spatial influence range, providing perfect input data for machine learning classifiers.

[0098] S60. Input the multidimensional feature vector into the pre-trained classification model to obtain the defect type classification result corresponding to the temperature anomaly region;

[0099] In this embodiment, this step is the core of intelligent defect classification. It utilizes a machine learning model to automatically learn and master the complex mapping relationship between feature vectors and defect categories, thereby enabling rapid and accurate type identification of newly detected unknown abnormal regions. The pre-trained classification model is a support vector machine (SVM) model, obtained through the following methods: collecting a large number of ceramic samples with known defect types, acquiring their temperature field data, and extracting multi-dimensional feature vectors to form a labeled sample set; preprocessing and enhancing the labeled sample set, and dividing it into a training set, a validation set, and a test set; training the SVM model using the training set, and optimizing the model parameters using the validation set; after training, extracting the support vectors and corresponding parameters that determine the classification hyperplane to form the pre-trained classification model.

[0100] Specifically, Support Vector Machines (SVMs) excel at small-sample, high-dimensional, and non-linear classification problems. The goal of SVM is to find an optimal hyperplane (a plane in three-dimensional space) that separates two classes of samples (such as cracks and bubbles) as much as possible, maximizing the margin between the points closest to this hyperplane in each class. These closest points are the support vectors. In this embodiment, the extracted three-dimensional feature vectors have a moderate dimension, and sample acquisition is costly (requiring manual destructive verification of the true defect type), making SVM very suitable. Furthermore, the final decision function of the SVM model depends on only a few support vectors, resulting in a relatively simple model structure that is easy to implement and accelerate in hardware, meeting real-time requirements.

[0101] As an example, the training process of the classification model is as follows:

[0102] Sample library construction: Collect a large number of ceramic workpiece samples with known internal defect types (which can be confirmed by industrial CT, destructive dissection, etc.); perform steps S10-S50 above on each sample to extract feature vectors [G] for each real defect region. max ,δ avg[S], and label them (e.g., cracks are labeled as "+1", bubbles are labeled as "-1"). This forms a labeled sample set, for example, containing 5000 crack samples and 5000 bubble samples, covering defects of different sizes (cracks 0.1-5mm, bubbles 0.5-10mm) and different locations (center / edge / corner). Each sample set contains a 3D feature vector and manually labeled tags (crack = 1, bubble = -1).

[0103] Data preprocessing and augmentation: Removing outlier samples (such as G) max >10℃ / mm), through slight perturbation characteristic values ​​(such as G) max To generate more training samples (fluctuating within ±10%), thus enhancing the model's robustness;

[0104] Model training and optimization:

[0105] a. Divide the dataset: Divide the sample set into training set, validation set and test set in proportion (e.g. 7:2:1); b. Select kernel function: Since the features may be linearly separable or approximately linearly separable, in order to balance accuracy and hardware implementation complexity, a linear kernel function is preferred. The formula is K(x1,x2)=x1x2+b, where b is the bias term.

[0106] c. Parameter tuning: The main adjustment is to the penalty parameter C (to control the tolerance for misclassified samples). The performance of C∈[0.01,0.1,1,10,100] is tested by 5-fold cross-validation to determine the optimal parameter of C.

[0107] d. Training: Train the SVM model using the training set and selected parameters;

[0108] Model solidification: After training, all support vectors (e.g., 523 support vectors are extracted, including 268 crack support vectors and 255 bubble support vectors, and subsequent inference only needs to use support vectors to reduce hardware resource consumption), the corresponding Lagrange multiplier coefficients, and bias terms, etc. These parameters will be solidified into hardware or software for online inference.

[0109] Finally, the multidimensional feature vector of the temperature anomaly region in the ceramic workpiece under test is input into the pre-trained classification model to obtain the defect type classification result corresponding to the temperature anomaly region. The defect discrimination principle is: SVM finds the optimal classification hyperplane to maximize the distance between crack and bubble samples and the hyperplane. The hyperplane equation is w. x + b = 0 (w is the weight vector). For the new input three-dimensional feature vector x (i.e., [G... max ,δ avg ,S]), if w If x+b>0, it is determined to be a crack; if w If x+b<0, it is determined to be a bubble. The hyperplane can maximize the classification margin between the two classes of samples and improve the generalization ability; the model can also output the confidence of the classification.

[0110] Through the steps of this embodiment, the system acquires the ability to intelligently identify cracks and bubbles. It is no longer a simple alarm for abnormalities, but can provide clear information on the type of defect. This is of great value for quality grading and tracing the source of process problems (for example, cracks may be caused by stress, and bubbles may be caused by poor venting). The high accuracy and fast reasoning capability of the SVM model ensure the feasibility of online application.

[0111] S70. Based on the defect type classification results and the location information of the temperature anomaly area, generate and output detection result information including defect type, location and severity level.

[0112] This embodiment is the final output of the detection process. It encapsulates all the structured information obtained from the previous analysis and outputs it in a standardized format to drive the downstream production execution system (such as automatic rejection) and the upper-level information management system.

[0113] Specifically, for each identified defect, a complete inspection record is generated, typically including: Defect type: such as "crack" or "bubble"; Defect location: the coordinates of the center point of the defect area (x0, y0) with reference to the workpiece coordinate system; Optional: the outer rectangular boundary (x1, y1, x2, y2) of the defect area for visualization or precise location and removal; Severity level: based on the defect size (e.g., crack length, bubble diameter) and characteristic intensity (e.g., G). max The size of the crack is classified into several levels (e.g., minor, moderate, severe). For example, a crack length greater than 2 mm or a bubble diameter greater than 5 mm can be classified as severe. Auxiliary information includes: inspection timestamp, workpiece unique ID (which can be generated by encoder pulse count), classification confidence level, etc.

[0114] Furthermore, the aforementioned detection results are output through a result output module. This module transmits instructions containing defect location information to the production line's programmable logic controller (PLC) in real time via industrial fieldbuses such as RS-485 / Modbus RTU. Based on the location information, the PLC controls the pneumatic push rod or flip-plate mechanism at the corresponding workstation, accurately removing the workpiece from the qualified product line when it reaches that workstation. The result output module also uploads a more complete detection data packet (including the original temperature frame, preprocessed temperature field matrix, feature vector, classification results, equipment status, etc.) to the factory's manufacturing execution system or industrial database via Ethernet / TCP-IP protocol. This enables model optimization: accumulating labeled samples for SVM incremental training; quality traceability: querying historical detection data via workpiece ID; and equipment diagnostics: monitoring temperature data stability, and triggering maintenance alarms if continuous abnormal data is detected, indicating an imager malfunction.

[0115] This embodiment completes a closed loop from data analysis to physical execution and digital management. It not only achieves the ultimate goal of inspection—eliminating defective products—but also builds valuable digital quality assets, providing data support for continuous process improvement and product quality enhancement.

[0116] The method for detecting internal defects in ceramics in the embodiments of the present invention has been described above. The following describes the device for detecting internal defects in ceramics in the embodiments of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the ceramic internal defect detection device of the present invention includes:

[0117] Data acquisition module 10 acquires time-series data of surface temperature field of ceramic workpieces during natural cooling process after exiting the kiln;

[0118] Preprocessing module 20 is used to perform real-time preprocessing on the acquired temperature field time series data to obtain the calibrated temperature field matrix;

[0119] Feature extraction module 30 extracts dynamic temperature features from the temperature field matrix, the dynamic temperature features including temperature gradient amplitude and cooling rate;

[0120] The positioning module 40 is used to locate temperature anomaly areas on the surface of a ceramic workpiece based on the extracted dynamic temperature features.

[0121] Vector extraction module 50 is used to extract multidimensional feature vectors from the temperature anomaly region to characterize the defect type;

[0122] The prediction module 60 is used to input the multidimensional feature vector into a pre-trained classification model and output the defect type classification result corresponding to the temperature anomaly region.

[0123] The result output module 70 is used to generate and output detection result information containing defect type, location and severity level based on the defect type classification result and the location information of the temperature anomaly area.

[0124] In this invention, the preprocessing module 20, feature extraction module 30, localization module 40, vector extraction module 50, prediction module 60, and result output module 70 are all integrated into a Xilinx ZYNQ data chip, preferably a Xilinx Zynq UltraScale+ series chip such as the XCZU9EG chip. This chip integrates a dual-core ARM processor (PS side) and FPGA logic (PL side), achieving high-speed data interaction between the PS side and PL side via the AXI4 bus. It also possesses large-capacity logic units, BlockRAM, multipliers, and other resources. Each module adopts a pipelined + parallel computing architecture. Each module caches data through FIFO (First-In-First-Out) to achieve pipelined operation. At the same time, the same calculation step (such as gradient calculation of a 3×3 neighborhood) is processed in parallel. The gradient calculation of each pixel is completed by an independent arithmetic logic unit (ALU), which greatly improves the computational efficiency.

[0125] Based on the same ideas as the methods in the above embodiments, the device provided by the present invention can implement the methods in the above embodiments. For ease of explanation, the structural schematic diagram of the device embodiment only shows the parts related to the embodiments of the present invention. Those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer modules than illustrated, or combine certain modules, or have different module arrangements.

[0126] Figure 2 The ceramic internal defect detection device in this embodiment of the invention is described in detail from the perspective of modular functional entities. The ceramic internal defect detection device in this embodiment of the invention is described in detail below from the perspective of hardware processing.

[0127] Figure 3This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 100 can vary significantly due to differences in configuration or performance. It may include one or more central processing units (CPUs) 111 (e.g., one or more processors) and a memory 121, and one or more storage media 130 (e.g., one or more mass storage devices) for storing application programs 133 or data 132. The memory 121 and storage media 130 may be temporary or persistent storage. The program stored in the storage media 130 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 100. Furthermore, the processor 111 may be configured to communicate with the storage media 130 and execute the series of instruction operations in the storage media 130 on the electronic device 100.

[0128] Electronic device 100 may also include one or more power supplies 141, one or more wired or wireless network interfaces 151, one or more input / output interfaces 161, and / or one or more operating systems 131, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The device structure shown does not constitute a limitation on the electronic device 100, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0129] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a ceramic internal defect detection method.

[0130] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0131] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] In summary, existing infrared detection technologies either focus only on static temperature or simply use a single dynamic feature (such as only the cooling curve). This invention, however, creatively extracts two complementary dynamic core features—temperature gradient amplitude and cooling rate—simultaneously in the spatiotemporal dimensions, and further constructs a multidimensional discriminative feature vector including maximum gradient, average rate deviation rate, and region area. Furthermore, it introduces and optimizes a Support Vector Machine (SVM) model for the first time, specifically for infrared thermographic classification of internal defects in ceramics, establishing a reliable and learnable mapping relationship from complex thermophysical phenomena to specific defect categories. This holistic approach of multidimensional dynamic features and SVM intelligent classification fundamentally solves the core problems of limited feature extraction and weak classification capabilities in traditional methods.

[0133] To address the stringent real-time requirements of production lines, this invention goes beyond simply optimizing algorithms; it employs a collaborative hardware and software design. For computationally intensive and rule-based tasks such as data preprocessing, gradient calculation, and region analysis, an innovative FPGA-based pipeline + parallel processing hardware acceleration architecture is designed and embedded into the programmable logic of the ZYNQ chip. Simultaneously, control logic, such as the SVM classifier, is deployed on the processing system side. This architecture fully leverages the advantages of heterogeneous computing, achieving end-to-end millisecond-level latency from data acquisition to result output at 100Hz high-frequency sampling, creatively solving the engineering challenge of insufficient real-time performance in high-speed online infrared thermal imaging detection scenarios.

[0134] Addressing the harsh industrial environment at kiln exits, this invention proposes a systematic engineering solution. Through an encoder position synchronization triggering mechanism, it ensures seamless integration of data acquisition with production line cycle time and data validity. A triple-protection design—water-cooled jacket, protective shell, and compressed air purging—guarantees the long-term stable operation of core sensing equipment in high-temperature, dusty environments. These are not simply a collection of components, but a creative system integration closely integrated with the core detection algorithm to ensure the quality of the data source, enabling advanced detection algorithms to reliably transition from the laboratory to the industrial field.

[0135] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting internal defects in ceramics, characterized in that, Including the following steps: Collect time-series data of surface temperature field of ceramic workpieces during natural cooling process after exiting the kiln; The collected time-series temperature field data is preprocessed in real time to obtain the calibrated temperature field matrix. Dynamic temperature features are extracted from the temperature field matrix. These dynamic temperature features include temperature gradient magnitude and cooling rate. Specifically, this includes: calculating the temperature gradient field in the spatial domain for each time point t in the temperature field matrix T(x, y, t); calculating the temperature change rate of each pixel in the x and y directions using a 3×3 pixel neighborhood difference algorithm; calculating the temperature gradient magnitude G(x, y, t) of each pixel based on the temperature change rate, where the temperature gradient magnitude characterizes the local spatial intensity of the temperature change at that point; setting a time analysis window, calculating the temperature change of each spatial coordinate point (x, y) within the time analysis window; and dividing the temperature change by the duration of the time analysis window to obtain the cooling rate V(x, y) corresponding to that coordinate point. The temperature anomaly region on the surface of the ceramic workpiece is located based on the extracted dynamic temperature features. A multidimensional feature vector for characterizing the defect type is extracted from the temperature anomaly region. The multidimensional feature vector includes: the maximum temperature gradient value G within the temperature anomaly region. max The deviation rate δ between the average cooling rate in the abnormal temperature region and the average cooling rate in the normal region. avg The physical area S occupied by the temperature anomaly region on the surface of the ceramic workpiece; The multidimensional feature vector is input into a pre-trained classification model to obtain the defect type classification result corresponding to the temperature anomaly region; Based on the defect type classification results and the location information of the temperature anomaly area, detection result information including defect type, location, and severity level is generated and output.

2. The method for detecting internal defects in ceramics according to claim 1, characterized in that, Collecting time-series data of the surface temperature field of ceramic workpieces during the natural cooling process after exiting the kiln includes the following steps: Infrared imagers are pre-positioned near the cooling zone at the outlet of the ceramic kiln; The workpiece position is monitored in real time by an encoder installed on the production line conveyor mechanism; When the encoder detects that the ceramic workpiece has reached the preset sampling start position, it sends a trigger signal to the infrared imager; The infrared imager is controlled to continuously acquire infrared images of the ceramic workpiece surface at a sampling frequency of not less than 100Hz during the period when the workpiece passes through the field of view and convert them into time-series data of the surface temperature field.

3. The method for detecting internal defects in ceramics according to claim 1, characterized in that, The acquired time-series temperature field data is preprocessed in real time to obtain the calibrated temperature field matrix, including the following steps: Optical distortion correction is performed on each frame of surface temperature field time-series data to eliminate image distortion caused by lens characteristics; Outlier cleaning is performed on the corrected surface temperature field time series data to remove invalid or abrupt temperature points caused by environmental interference. Gaussian filtering was applied to the time-series data of the surface temperature field after cleaning to suppress random noise. The processed surface temperature field time series data are organized according to time series to form a temperature field matrix T(x, y, t) containing spatial coordinates and time dimension, where x and y are the coordinates of the original temperature image on the surface of the ceramic workpiece, and t is the cooling time.

4. The method for detecting internal defects in ceramics according to claim 1, characterized in that, The method for locating temperature anomaly regions on the surface of ceramic workpieces based on extracted dynamic temperature features includes the following steps: Based on the temperature gradient amplitude data of normal, defect-free samples, the normal temperature gradient amplitude range is statistically obtained, and a first discrimination threshold is set. Based on the cooling rate data of normal, defect-free samples, the normal cooling rate range was statistically obtained, and a second discrimination range was set. Traverse each pixel in the temperature field matrix. If its temperature gradient magnitude exceeds the first discrimination threshold and its cooling rate exceeds the second discrimination range, then mark the pixel as a candidate anomaly. Connectivity analysis is performed on spatially adjacent candidate anomalies, and they are merged to form connected temperature anomaly regions.

5. The method for detecting internal defects in ceramics according to claim 2, characterized in that, The pre-trained classification model is a support vector machine model, which is obtained in the following way: A large number of ceramic samples with known defect types were collected, their temperature field data were obtained, and multi-dimensional feature vectors were extracted to form a labeled sample set; The labeled sample set is preprocessed and enhanced, and then divided into a training set, a validation set, and a test set. The support vector machine model is trained using the training set, and the model parameters are optimized using the validation set. After training, the support vectors and corresponding parameters that determine the classification hyperplane are extracted to form a pre-trained classification model.

6. A ceramic internal defect detection device, characterized in that, include: The data acquisition module collects time-series data of the surface temperature field of ceramic workpieces during the natural cooling process after exiting the kiln; The preprocessing module is used to perform real-time preprocessing on the acquired temperature field time series data to obtain the calibrated temperature field matrix; The feature extraction module is used to extract dynamic temperature features from the temperature field matrix. These dynamic temperature features include temperature gradient magnitude and cooling rate. Specifically, it includes: calculating the temperature gradient field in the spatial domain for each time point t in the temperature field matrix T(x, y, t); calculating the temperature change rate of each pixel in the x and y directions using a 3×3 pixel neighborhood difference algorithm; calculating the temperature gradient magnitude G(x, y, t) of each pixel based on the temperature change rate, where the temperature gradient magnitude characterizes the local spatial intensity of the temperature change at that point; setting a time analysis window, calculating the temperature change of each spatial coordinate point (x, y) within the time analysis window; and dividing the temperature change by the duration of the time analysis window to obtain the cooling rate V(x, y) corresponding to that coordinate point. The positioning module is used to locate temperature anomaly areas on the surface of ceramic workpieces based on extracted dynamic temperature features. The vector extraction module is used to extract a multidimensional feature vector from the temperature anomaly region to characterize the defect type. The multidimensional feature vector includes: the maximum temperature gradient value G within the temperature anomaly region. max The deviation rate δ between the average cooling rate in the abnormal temperature region and the average cooling rate in the normal region. avg The physical area S occupied by the temperature anomaly region on the surface of the ceramic workpiece; The prediction module is used to input the multidimensional feature vector into a pre-trained classification model and output the defect type classification result corresponding to the temperature anomaly region. The result output module is used to generate and output detection result information containing defect type, location and severity level based on the defect type classification result and the location information of the temperature anomaly area.

7. An electronic device, characterized in that, It includes a memory and at least one processor, wherein the memory stores computer-readable instructions; The at least one processor invokes the computer-readable instructions in the memory to perform the steps of the ceramic internal defect detection method as described in any one of claims 1-5.

8. A computer-readable storage medium storing computer-readable instructions thereon, characterized in that, When the computer-readable instructions are executed by a processor, they implement the various steps of the ceramic internal defect detection method as described in any one of claims 1-5.

Images