A non-destructive testing process for the sealing of sintered tantalum insulators

Abstract

This invention belongs to the field of non-destructive testing technology for high-temperature materials, and discloses a non-destructive testing process for the sealing performance of tantalum insulators after sintering. The process includes: continuously capturing multiple frames of test distortion images containing structured backgrounds using a visible light camera to form an image time series; executing frame-by-frame spatial domain analysis combining machine vision attitude calculation and asymmetric analysis based on actual physical attitude to generate a time series of differential fields; performing time-domain frequency analysis on local thermal anomaly signals in the differential field time series to obtain time-domain frequency characteristics; and comparing the time-domain frequency characteristics with a preset scintillation frequency window to confirm defects. This invention, by introducing time-domain frequency analysis, utilizes the scintillation frequency characteristics of the defect thermal plume to distinguish between real defects and interfering thermal noise, solving the false positive problem of spatial domain analysis. Combined with attitude adaptive compensation, it improves the robustness of the detection.

Description

Technical Field

[0001] This invention relates to a non-destructive testing process for the sealing performance of tantalum insulators after sintering, belonging to the field of non-destructive testing technology for high-temperature materials. Background Technology

[0002] This paper specifically relates to a process method for analyzing the physical properties of electronic components after high-temperature processing using optical principles. In the manufacturing of high-reliability electronic components such as tantalum insulators, sintering is a crucial process determining the mechanical strength and sealing performance of the final product. However, stress release during high-temperature cooling can easily induce penetrating microcracks, leading to sealing failure. Existing technologies have made various attempts to detect sealing defects, but all have inherent limitations. Besides the difficulty in achieving online detection at the physical testing hardware level, the methods for identifying and capturing leakage signal characteristics also have fundamental shortcomings. For example, Chinese invention patent CN110333035A discloses a sealing leak detection device and method for tantalum insulators. This method is based on the helium mass spectrometry leak detection principle, through... Constructing a sealed chamber, filling it with helium, and using a helium mass spectrometer leak detector to determine the leakage rate of the tantalum insulator solves the technical problem that leak detection after capacitor assembly may lead to the scrapping of the entire capacitor, enabling individual leak detection of the tantalum insulator body. However, this improvement is still limited by its physical detection principle. The detection principles of leakage inspection methods such as helium mass spectrometry, such as pressure difference and tracer gas, limit them to offline room temperature vacuum conditions. This makes the detection behavior completely disconnected from the sintering process, which is a critical step, in time and space. This results in a systematic delay between the detection information and the production process. The production line cannot obtain real-time feedback on sintering process parameters. Once a defect is discovered, it may have already caused a batch of quality accidents. This poses a quality control risk for high-reliability applications.

[0003] To achieve online inspection of processes, some technical solutions have attempted to use material physical property analysis methods, such as using infrared thermal imaging technology to directly observe components at high temperatures. However, the deployment cost of high-speed, high-precision infrared thermal imagers is high, and their measurements are easily affected by the uncertainty of the emissivity of the object's surface, making it difficult to reliably use for 100% full inspection. In contrast, although conventional visible light optical analysis methods have low deployment costs, they are considered ineffective in high-temperature inspection scenarios. This is because, on the one hand, the strong thermal radiation of the high-temperature component itself interferes with imaging; on the other hand, the thermally induced refractive index field around the component, commonly known as thermal haze, causes optical distortion, which is generally regarded as a random optical noise that must be eliminated or shielded, hindering precise morphological measurements based on visible light. This technical understanding that regards thermal air distortion as interference limits the application of conventional optical analysis methods in high-temperature online inspection, making online assessment of sealing performance dependent on other detection methods.

[0004] Therefore, the technical problem to be solved by this invention is to re-examine the thermo-induced refractive index field around a high-temperature object, explore how to transform it from optical interference into a signal carrier carrying information about the physical state of the component, and devise a new online analysis method that can stably extract the signal to identify defects under strong thermal field background and environmental airflow interference using low-cost optical components. Summary of the Invention

[0005] This invention provides a non-destructive testing process for the sealing performance of tantalum insulators after sintering. Its main purpose is to solve the problems in the prior art, where the high-temperature thermally induced refractive index field is regarded as optical interference, making it impossible to use low-cost visible light methods for online detection, and the resulting lag in quality control information and limitations in sampling inspection.

[0006] To achieve the above objectives, this invention provides a non-destructive testing process for the sealing performance of tantalum insulators after sintering. The process utilizes a visible light camera to capture a reference image without including the structured background of the tantalum insulator, and continuously captures multiple frames of test distortion images passing through the thermally induced refractive index field of the air medium surrounding the tantalum insulator, which include the structured background, forming an image time series. The process includes:

[0007] The calculation process involves the processor comparing each frame of the test distorted image in the image time series with the reference image to calculate a time series of the absolute distortion field.

[0008] The spatial domain analysis step involves the processor performing frame-by-frame spatial domain analysis on the time series of the absolute distortion field to generate a time series of the difference field. The spatial domain analysis step includes: using machine vision algorithms to calculate the actual physical pose parameters, including translation and rotation parameters, from each frame of the test distorted image in the image time series in real time; based on the actual physical pose parameters, dynamically adjusting the definition of a logical symmetry axis, and performing asymmetry analysis on the time series of the absolute distortion field to calculate and generate the time series of the difference field.

[0009] The time-domain signature identification step involves the processor performing time-domain frequency analysis on the signal composed of local thermal anomalies in the time series of the differential field to obtain time-domain frequency characteristics; a scintillation frequency window based on the fluid dynamics principle of defect thermal plumes is preset.

[0010] The decision-making process involves confirming a local thermal anomaly signal as originating from a sealing defect only when the time-domain frequency characteristics of the signal meet the preset conditions of the flicker frequency window.

[0011] Preferably, the time-domain signature identification step and decision-making step aim to solve the false positive problem in the local thermal anomalies discovered by the spatial domain analysis step, which cannot distinguish between thermal jets originating from real defects and thermal noise originating from interference sources. The time-domain signature identification step specifically includes: the processor, based on the principle of fluid dynamics, sets the scintillation frequency window to a specific frequency range that excludes transient interference signals and low-frequency environmental thermal convection signals, the specific frequency range being 10Hz to 50Hz; the processor performs spectral analysis on the signal intensity variation of the local thermal anomaly in the time series of the differential field over time to calculate the time-domain frequency characteristics; the decision-making step specifically involves: the processor comparing the calculated time-domain frequency characteristics with the specific frequency range, and only performing the operation to confirm the sealing defect when the time-domain frequency characteristics fall within the specific frequency range.

[0012] Preferably, the spatial domain analysis step involves performing asymmetric analysis to calculate the time series of the generated difference field, specifically including: constructing a logistic transformation function. ,in Given the coordinates of a point in the image, the logical transformation function, based on the actual physical pose parameters, is used to calculate the point's coordinates. Regarding the coordinates of the mirror point of the real-time physical symmetry axis of the tantalum insulator; by calculating the absolute distortion field at point in each frame of the time series of the absolute distortion field. With point The difference at each point is used to generate a time series of the difference field.

[0013] Preferably, the first calculation step is implemented by executing a digital image correlation algorithm.

[0014] Preferably, the first calculation step is implemented by executing an optical flow algorithm.

[0015] Preferably, in the spatial domain analysis step, the machine vision algorithm is a contour matching algorithm.

[0016] Preferably, in the spatial domain analysis step, the machine vision algorithm is a template matching algorithm.

[0017] Preferably, when the tantalum insulator has a geometrically asymmetric shape, the spatial domain analysis step is replaced by a dispersion differential analysis step. The dispersion differential analysis step aims to solve the technical problem of asymmetric analysis failure caused by the geometric asymmetry of the component. The dispersion differential analysis step includes: illuminating the structured background almost simultaneously using at least two different wavelength light sources; synchronously acquiring multiple frames of test distortion images corresponding to at least two wavelengths using a visible light camera to form multiple image time series; calculating multiple distortion fields corresponding to different wavelengths by a processor; utilizing the dispersion effect of the thermally induced refractive index field on light of different wavelengths, and performing differential processing on multiple distortion fields to eliminate common distortion noise caused by global environmental airflow and amplify the dispersion anomaly signal caused by local thermal anomalies to generate a time series of differential fields.

[0018] Preferably, the process further includes a global baseline dynamic arbitration step, which aims to solve the technical problem that the spatial domain analysis step misjudges environmental noise as local thermal anomalies due to sudden asymmetric turbulence in the production line environment. The global baseline dynamic arbitration step includes: extracting distortion data of the far-field background region physically far from the tantalum insulator from the time series of the absolute distortion field; calculating the spatial variance or temporal variance of the distortion data of the far-field background region to generate a global baseline fluctuation index; setting an environmental stability threshold; only when the global baseline fluctuation index is lower than the environmental stability threshold is the sealing defect confirmation result generated by the decision step confirmed as valid; when the global baseline fluctuation index is higher than the environmental stability threshold, the sealing defect confirmation result is determined to be invalid.

[0019] Preferably, the process further includes a defect quantification step, which includes: offline calibration: establishing a correlation model between the signal characteristics of the local thermal anomaly signal confirmed in the decision-making step and the physical leakage rate measured by the leakage inspection method; online quantification: using the correlation model, estimating the leakage level of the sealing defect based on the signal characteristics of the local thermal anomaly signal detected online.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. The method for which this technology seeks protection establishes a new optical analysis path. It shifts the focus to analyzing the optical refractive index field of the surrounding air medium during the cooling process of a high-temperature component, rather than observing the solid component itself. By calculating the absolute distortion morphology of this field and using the component's own geometric symmetry as a real-time reference, the asymmetric morphology analysis step logically achieves a dual suppression effect: while eliminating the inherent symmetrical thermal field morphology of the component, it simultaneously filters out large-scale symmetrical or uniformly varying environmental airflow disturbances. As a result, the asymmetric thermal plume features caused by localized minor sealing defects, which were previously submerged by the aforementioned two types of strong background distortion, can be clearly identified as high-contrast signals. This technology can further extend optical acquisition from single-frame images to continuous acquisition. Obtaining an image time series introduces a temporal feature analysis dimension on top of the spatial domain distortion morphology analysis described above. Based on the principles of thermophysics and fluid dynamics, the thermal jet formed by the continuous ejection of fixed microcracks exhibits specific quasi-periodic temporal scintillation frequency characteristics when mixed with cold air, while interfering heat sources such as floating hot dust particles exhibit transient or no fixed frequency characteristics. The temporal feature analysis step compares the temporal frequency characteristics of the abnormal signals detected in the spatial domain with a preset scintillation frequency feature window, enabling the system to identify the physical source of the abnormal signals. It can not only locate the anomaly in space but also confirm whether the cause of the anomaly is a sealing defect in the temporal dimension, thereby filtering out false positive interference sources with similar spatial morphology but inconsistent temporal characteristics.

[0022] 2. Addressing the issue of unavoidable random physical posture shifts in test components due to mechanical transmission tolerances during production line operation, this process utilizes the same test distortion image, reusing its visual information to perform real-time posture extraction before or during distortion field calculation. The actual physical posture parameters of the component calculated in this step are used to dynamically adjust the definition of the logical symmetry axis used in subsequent asymmetric analysis steps. This mechanism of dynamically coupling machine vision positioning results with optical distortion field analysis logic in a coordinate system allows the analysis algorithm to actively adapt to the translation and rotation of the component, avoiding differential artifacts and catastrophic misjudgments caused by mechanical positioning errors and logical coordinate system misalignment. This ensures the stability and applicability of the core detection mechanism in real industrial transmission environments; to expand... For the inspection of components whose geometry is asymmetrical, this process provides an alternative analysis path that does not depend on the symmetry of the component itself. For example, by using at least two different wavelengths of light to illuminate the structured background almost simultaneously and acquiring multiple frames of corresponding test distortion images at the same time, the dispersion differential analysis step can be performed. This step utilizes the dispersion effect of the thermally induced refractive index field on light of different wavelengths to acquire a test distortion field and a virtual reference distortion field carrying the exact same global environmental airflow noise within a microsecond timescale. By performing differential processing on the two based on the dispersion model, the highly correlated environmental noise distortion is eliminated, while the abnormal signal with dispersion difference caused by the defect thermal plume is retained, so that the detection capability is free from dependence on the geometry of a specific component.

[0023] 3. To address sudden, asymmetric local turbulence interference in the production line environment, this process can also execute a global baseline dynamic arbitration step in parallel. This step extracts and statistically analyzes the distortion data of the far-field background region, which is physically far from the component in the absolute distortion field, as an information byproduct that only characterizes the global environmental airflow state. By calculating the spatial or temporal variance of the distortion data in this far-field region, a global baseline fluctuation index is generated. The system can sense the environmental stability at the current detection moment in real time. When the index is higher than a preset environmental stability threshold, the decision arbitration module can determine that the defect identification result generated by the core detection mechanism is invalid or affected by environmental interference. This design uses the accompanying information in the image to reverse verify whether the operating premise of the core detection logic is valid, avoiding misjudging asymmetric strong noise turbulence as a defect signal when there are sudden changes in the harsh environment. Attached Figure Description

[0024] Figure 1 This is a flowchart of the detection process of the present invention that integrates temporal signature recognition and global baseline arbitration;

[0025] Figure 2 This is a comparison diagram of the power spectral density characteristics of the defect signal and the environmental noise of this invention;

[0026] Figure 3This is a diagram showing the system module composition and data flow architecture of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining this invention and are not intended to limit the scope of protection of this invention.

[0028] This invention provides a non-destructive testing process for the sealing performance of tantalum insulators after sintering. Its core operation begins with the optical acquisition stage, where a visible light camera continuously captures multiple frames of test distortion images, including a structured background, passing through the air medium surrounding the target (high-temperature tantalum insulator), forming an image time series. This image time series is transmitted to a processor, which sequentially executes calculation steps, spatial domain analysis steps, temporal signature identification steps, and a final decision step. The calculation step aims to generate an absolute distortion field time series by comparing each frame of the test distortion image with a pre-acquired reference image that does not contain the structured background of the tantalum insulator. The spatial domain analysis step processes this absolute distortion field time series frame by frame to generate a differential field time series. The temporal signature identification step then performs temporal frequency analysis on the local thermal anomaly signals in the differential field time series to obtain temporal frequency characteristics. Ultimately, the decision-making process compares the time-domain frequency characteristics with a preset scintillation frequency window based on fluid dynamics principles. Only when preset conditions are met is the local thermal anomaly signal confirmed as originating from a sealing defect. In a specific engineering implementation, this process targets the field of application where, after the sintering process of high-reliability electronic components such as tantalum insulators, online full inspection of their sealing performance is performed using optical metrology. The technical challenge it faces is that the thermally induced refractive index field (TIRF) around high-temperature components is often considered optical noise, hindering the application of low-cost visible light methods. This process design, by translating the thermally induced refractive index field into a signal carrier carrying the physical information of the defect, provides an analysis procedure that can operate stably in the production line environment. This procedure begins with optical acquisition, and the enabling environment requires a visible light camera, for example, with a global shutter and a frame rate of not less than [missing information]. The system uses a high-frequency industrial CMOS camera and a fixed structured background (e.g., random speckle pattern or checkerboard pattern) with high spatial frequency characteristics. Initially, when the detection station is unobstructed, the system first uses the visible light camera to capture and store a reference image of the structured background. Subsequently, when the tantalum insulator, freshly sintered and in a high-temperature cooling state, enters the optical path of the inspection station, the visible light camera is triggered, sampling at a preset frequency (e.g., [missing information]). (Hz) Continuously captures multiple frames of test distortion images passing through the thermally induced refractive index field around the tantalum insulator. This forms an image time series.

[0029] Upon receiving the image time series, the processor immediately initiates a computation step. The purpose of this step is to transform the pixel brightness information of the original image into a vector field representing the physical light deflection. The application-level obstacle it faces is the need to calculate the minute pixel displacements caused by TIRF with high accuracy and efficiency. To address this challenge, this computation step is configured to execute a Digital Image Correlation (DIC) algorithm, which uses a reference image... Using this as a baseline, each frame of the image time series is used as a test distortion image. As input, through Define dense computational subregions above, and in Within the corresponding search domain, the most relevant matching position is found, thereby calculating the sub-pixel displacement vector of the center point of each sub-region. These vectors together constitute the absolute distortion field characterizing the thermally induced refractive index field. In another implementation, to meet the computational efficiency requirements of high-speed production lines, this calculation step can also be configured to execute an optical flow algorithm, such as the Lucas-Kanade optical flow method, to estimate more quickly. Compared to The pixel displacement; the value of this step lies in quantitatively translating the invisible refractive index field changes into a high-resolution time series of absolute distortion fields that can be processed by subsequent algorithms. Immediately afterwards, the processor... The time series analysis begins with a frame-by-frame spatial domain analysis step. The core objective of this step is to analyze the spatial domain of the data, which incorporates a mixture of component-symmetric thermal fields, ambient airflow noise, and defect-asymmetric thermal plumes. In this study, signals related only to defects are extracted. The key challenge in the field of metrology is that mechanical transfer tolerances in the production line cause random shifts in the physical orientation (translation and rotation) of tantalum insulators in each frame of the image. This makes any symmetry analysis method based on a fixed coordinate system ineffective due to alignment artifacts.

[0030] To address this challenge, the spatial domain analysis step was designed as an adaptive attitude solution-asymmetric analysis cascade procedure. First, the processor utilizes the test distortion image... Visual information is obtained in real time from machine vision algorithms, such as a contour matching algorithm based on part contours or a template matching algorithm based on part features. The solution includes translation parameters. , ) and rotation parameters ( The processor then dynamically adjusts the definition of the logical symmetry axis based on the actual physical attitude parameters calculated in real time, and adjusts the absolute distortion field of the current frame. Performing asymmetric analysis; in econometric practice, this asymmetric analysis is specifically implemented by constructing a logical transformation function. ,in Coordinates of points in the image Logical transformation function by( , , () is the input parameter used to calculate the point. Regarding the mirror point coordinates of the real-time physical symmetry axis of the tantalum insulator The processor then performs calculations. At point With point The distortion vector difference at that point, i.e. To generate difference fields The time series; the value of this mechanism that dynamically couples machine vision positioning results with optical distortion field analysis logic in a coordinate system is that it uses the geometric symmetry of the component itself as a real-time reference, which not only eliminates the inherent symmetrical thermal field shape of the component body, but also filters out large-scale symmetrical environmental airflow disturbances. At the same time, it avoids artifacts caused by mechanical positioning errors and logical coordinate system misalignment, so that the asymmetric thermal plume features caused by local small sealing defects, which were originally submerged by strong background distortion, can be identified as high-contrast signals.

[0031] However, identifying local thermal anomalies solely in the spatial domain still faces the technical problem of false positives, namely, the inability to distinguish between thermal jets originating from genuine defects and thermal noise from interference sources (such as floating hot dust particles), as the two may have similar spatial morphology in a single frame. To address this measurement specificity problem, this process introduces a time-domain signature identification step. This step no longer analyzes single-frame images but utilizes the time series of the difference field. Processor to The signal is composed of local thermal anomalies, for example, at a certain fixed coordinate. At that location, its signal strength Over time The system performs time-domain frequency analysis on the changes, which is achieved using standard spectral analysis algorithms (such as Fast Fourier Transform, FFT) to calculate the time-domain frequency characteristics (e.g., its power spectral density distribution or dominant frequency) of the local thermal anomaly signal. Simultaneously, the system presets a scintillation frequency window based on the principle of defect thermal plume hydrodynamics. This principle states that a thermal jet formed by continuous ejection from a fixed microcrack exhibits specific, quasi-periodic time-domain scintillation frequency characteristics when mixed with cold air, while interfering heat sources such as floating dust exhibit transient (high-frequency) or no fixed frequency characteristics, and low-frequency environmental thermal convection exhibits extremely low frequencies. To achieve accurate identification of the metering signal, the scintillation frequency window is set to a specific frequency range that excludes transient interference signals and low-frequency environmental thermal convection signals. In a preferred embodiment, this specific frequency range is determined as follows: Hz to Hz; the subsequent decision-making step is to execute a strict signature identification logic: the processor compares the calculated time-domain frequency characteristics with a specific frequency range ( ) Hz to The comparison is performed at Hz, and the operation to confirm the sealing defect is only performed when the dominant frequency of the time domain frequency feature falls into the specific frequency range. The value of this serial procedure of spatial domain positioning and time domain signature identification lies in improving the detection system's ability to identify the physical source of abnormal signals and confirm whether the cause of the abnormality is due to a sealing defect in the time dimension.

[0032] In another embodiment, when the tantalum insulator has a geometrically asymmetric shape, the aforementioned symmetry-based spatial domain analysis step will fail. In this case, the step is replaced by a dispersive differential analysis step; the enabling environment for this step is adjusted to utilize at least two different wavelengths of light sources (e.g., wavelength 1). Red LEDs and wavelengths of The blue LEDs) almost simultaneously (e.g., in Alternating pulses within milliseconds illuminate the structured background; a visible light camera simultaneously acquires multiple frames of test distortion images corresponding to at least two wavelengths, forming multiple image time series. and The processor then uses the aforementioned calculation steps (such as DIC) to calculate multiple distortion fields corresponding to different wavelengths. and The core of this step is to utilize the dispersion effect of the thermally induced refractive index field on light of different wavelengths, because... and The acquisition interval is on the order of microseconds, and both carry the exact same global environmental airflow distortion noise, but the dispersion effect caused by local thermal anomalies (high-temperature thermal plumes) differs; therefore, the processor performs differential processing on multiple distortion fields, for example, calculating... ,in To eliminate common distortion noise caused by global environmental airflow and amplify the dispersion anomaly signal caused by local thermal anomalies, a time series of differential fields is generated based on the constants calibrated by the dispersion model. This dispersion differential path allows the detection capability of this process to be independent of the geometry of specific components. To further improve the metrological robustness of the system in the production line environment, this process may also include a global baseline dynamic arbitration step. This step aims to solve the technical problem that the aforementioned spatial domain analysis step misjudges environmental noise as local thermal anomalies due to sudden asymmetric turbulence in the production line environment, such as gusts of wind caused by personnel movement or door opening. The arbitration step is implemented by: from the absolute distortion field... In the time series, distortion data from far-field background regions physically far from the tantalum insulator (e.g., the four corners of the image) is extracted; the processor calculates the spatial or temporal variance of the distortion data from the far-field background regions to generate a global baseline fluctuation index. The system presets an environmental stability threshold (e.g., based on the stable operation of the production line). (Statistical value setting); When making a decision, the sealing defect confirmation result generated by the decision step (e.g., time domain signature identification) is only confirmed as valid when the global baseline fluctuation index is lower than the environmental stability threshold; when the global baseline fluctuation index is higher than the environmental stability threshold, it indicates that there is asymmetric turbulence interference in the current environment, and the system will determine the confirmation result as invalid and may trigger an alarm or retest.

[0033] Finally, to correlate the results of this optical inspection process with physical standards in the field of leakage inspection, the process may also include a defect quantification step. This step includes an offline calibration stage and an online quantization stage. During offline calibration, a group of samples with different physical leakage rates are selected, and the process of this invention is used for detection. The signal characteristics of their local thermal anomaly signals are extracted, such as the signal integral intensity and the dominant amplitude of the flicker frequency, thereby establishing a correlation model between the signal characteristics and the physical leakage rate, such as a polynomial fitting curve or a neural network model. During online quantization, after a defect is confirmed, the system uses the established correlation model to estimate the leakage level of the sealing defect based on the signal characteristics of the local thermal anomaly signals detected online in real time. For example, the output is... This provides a traceable metrological standard for new material testing. To ensure clear engineering reproducibility of key parameters in this technical solution, the lower and upper limits of the flicker frequency window used in the time-domain signature identification step are determined through the following standardized engineering procedures: The procedure for setting the lower limit frequency includes continuously acquiring an image time series without containing the test piece or containing only a confirmed sealed test piece under standard production line operating conditions using an optical acquisition module; performing calculation steps on this sequence to obtain the absolute distortion field; extracting distortion data from the far-field background region physically far from the test piece and performing time-domain frequency analysis; obtaining the environmental noise spectrum under this operating condition; and setting the lower limit frequency to be higher than the dominant frequency of this environmental noise spectrum, such as... A specific value, for example, set to The procedure for setting the upper limit frequency includes using a set of defect samples with known leakage rates at the lower limit of detection (i.e., the weakest signal), acquiring their image time series at a standard detection temperature, performing calculation steps, spatial domain analysis steps, and spectral analysis, statistically analyzing the dominant frequency distribution of the defect thermal plume signal of this set of samples, and setting the upper limit frequency to a frequency value that can cover a specific percentage point (e.g., 95th percentile) of this distribution. Simultaneously, the sampling frequency of the visible light camera in the optical acquisition step... It is set to be no less than twice the upper limit frequency, i.e. .

[0034] The procedure for determining the environmental stability threshold upon which the global baseline dynamic arbitration step relies includes: after deployment on the production line, under acceptable stable operating conditions (i.e., all conventional equipment is running, but there is no sudden asymmetric turbulence), continuously running the detection system for an extended period of time, and continuously calculating the global baseline fluctuation index of the far-field background distortion data. The time series, for this Perform statistical analysis on the time series and calculate its average value. with standard deviation The environmental stability threshold Therefore, it was set as ,in It is an engineering safety factor determined based on the actual interference level of the production line, for example. A value of 3 to 6 can be used; this is the constant used in the dispersion difference analysis step to eliminate common distortion noise. The calibration preferably employs an online adaptive procedure, which utilizes distorted data from the far-field background region (this data contains only common ambient airflow noise and no defect signals). The processor is configured to periodically (e.g., every detection cycle or every few seconds) process the acquired far-field distorted data corresponding to different wavelengths. and Perform a least squares fit or correlation analysis to calculate in real time the result that makes the difference signal... The current best is to minimize the variance. Value, and use the real-time updated value. Dispersion difference processing is performed to compensate for the air refractive index drift caused by slow changes in ambient temperature or atmospheric pressure.

[0035] Example 1: In a specific industrial metrology scenario, the process claimed in this invention is deployed at the end of a continuously operating tantalum insulator sintering production line for online quality inspection. This production line station faces typical challenges in two fields: firstly, the components are carried by a high-speed conveyor belt, and mechanical vibration and positioning tolerances cause random deviations in the physical posture of each tantalum insulator to be tested when it arrives at the inspection station. For example, in... direction Pixels, in direction Pixels, and The first issue is the rotational offset; the second is the presence of a large forced cooling fan next to the testing station, which causes a persistent main frequency in the environment. The system experiences strong, asymmetric, low-frequency environmental airflow interference near Hz. Under these conditions, a tantalum insulator, fresh from the sintering furnace and in a high-temperature cooling state, carrying a penetrating microcrack, enters the testing station. This microcrack is continuously ejecting a stream of air with specific hydrodynamic characteristics, whose dominant frequency is near Hz. A high-frequency scintillation heat plume near Hz; simultaneously, a piece of incandescent slag dust, acting as a transient heat source without fixed-frequency interference, happens to drift across the detection area; the system starts detection according to procedure, and the visible light camera... The image time series is continuously captured at a sampling frequency of Hz, and the processor then executes calculation steps to generate a time series of absolute distortion field. ,Should At this point, it is a complex distorted field that combines four signals, the main components of which include the component's own symmetrical laminar thermal field, Low-frequency environmental turbulence at Hz, transient slag dust signals, and Hz defective thermal plume signal.

[0036] The processor then performs a frame-by-frame spatial domain analysis step, which demonstrates the first layer of collaborative mechanism: machine vision algorithms, such as a contour matching algorithm, first calculate in real time from each frame of the test distortion image the components, such as... Pixel translation and The actual physical attitude parameters of the degree rotation; these attitude parameters are immediately used to dynamically adjust the definition of the logical symmetry axis used in the asymmetry analysis, and the processor then uses logical transformation functions. Calculate the time series of the generated difference field Under this differential operation, the symmetrical components of signal 1 (component symmetrical thermal field) and signal 2 (low-frequency environmental turbulence) are significantly eliminated even when the component undergoes physical attitude shift. However, signals 3 (slag dust) and 4 (defect plume), as locally asymmetrical anomalous signals, are retained. At this point, the system still faces the dilemma of false positives in the spatial domain. The system then initiates a time-domain signature identification step, which demonstrates a second collaborative mechanism: the cascaded identification of spatial and time-domain analysis. The processor no longer... Instead of making decisions based on the single-frame amplitude, this study performs time-domain frequency analysis on the signal strength changes over time in the regions containing signals 3 and 4, for example, by performing a Fast Fourier Transform (FFT). The analysis results show that the spectrum of signal 3 (slag dust) is diffuse and has no fixed dominant frequency, which does not conform to the characteristics of a hydrodynamic jet; while the spectrum of signal 4 (defect plume) is more... A narrow-band peak is observed at Hz; finally, the decision-making step is initiated, and the processor compares the time-domain frequency characteristics of these two signals with a preset value. Hz to The flicker frequency window of Hz is compared. Signal 3 is judged as interference and filtered out because its frequency characteristics do not fall into the window. Signal 4 is filtered out because... The Hz main frequency falls into this window and is finally identified as originating from a sealing defect. This operating mode first uses spatial domain asymmetry analysis calibrated by machine vision to suppress strong background noise of symmetry, and then uses frequency characteristic analysis in the time domain to identify the physical source of the asymmetric signal, thereby achieving effective identification of signals originating from sealing defects under the dual interference of mechanical vibration and environmental thermal noise.

[0037] Example 2: To objectively verify the effectiveness and robustness of the process of the present invention in a simulated industrial environment, especially its ability to distinguish signals originating from sealing defects from common interference sources, the following experiment was designed and executed; the experiment aimed to verify the system's ability to detect known minor leakage defects and suppress defect-free samples and transient thermal interference when using the complete technical process; the test platform was built on an optical vibration isolation platform to simulate a production line inspection station, which included: a temperature-controlled heating stage for heating the tantalum insulator sample to be tested to the initial cooling temperature after sintering. A two-dimensional precision displacement stage holds the sample to simulate random position and orientation shifts introduced by a conveyor belt, with the shift range set in the X direction. Pixel, Y direction Pixels, rotation The optical system consists of a visible light industrial camera (CMOS sensor, 1280x1024 pixel resolution, global shutter, set frame rate). The sample consists of a random speckle structured background plate (Hz) fixed in the direction directly opposite the camera's optical axis; the environmental interference simulation part includes an adjustable-speed axial flow fan placed to the side of the sample, generating a frequency of approximately [frequency missing]. A non-uniform airflow field of Hz, and a particle dispersing device, for introducing a diameter of approximately Hz into the region near the sample in a specific test group. micrometers, preheated to approximately The graphite particles simulate transient incandescent dust interference; data processing is performed by a computer equipped with a standard computing processor.

[0038] The experiment used three groups of tantalum insulator samples: the first group was the test group, which included... The component is pre-drilled with micropores (approximately [diameter missing]). (micrometers) and confirmed by standard helium mass spectrometry leak detector calibration. The first group consisted of defective samples with leakage rates on the order of magnitude; the second group was the control group 1, containing... The first group consisted of samples that were confirmed to be properly sealed by the same helium mass spectrometry leak detection; the second group was the control group 2. The sample was sealed, but the aforementioned graphite particle interference was introduced during the test; the test procedure was repeated for each sample: the sample was heated to... The device is then placed on a displacement stage, and a random attitude offset within a preset range is set; the axial flow fan is activated to create ambient airflow; the camera... Continuous acquisition at Hz frame rate seconds (total) (Frames) of image data, forming an image time series; for the sample in control group 2, approximately [number] frames were collected during the acquisition process. The particle dispersal device is triggered at the second; the acquired image time series is then processed by the processor; the camera frame rate is set to [value missing]. The technical consideration for Hz is that it needs to satisfy the Nyquist sampling theorem to effectively capture the upper limit of the expected defect signal flicker frequency. Hz, while also taking into account data processing load; the acquisition duration is set to The time interval is chosen to ensure that the lowest expected flicker frequency can be captured. Sufficient number of cycles per Hz ( (number of cycles) to perform reliable spectrum analysis; the scintillation frequency window is set to [number of cycles]. Hz to Hz, this range is based on fluid dynamics analysis of the vortex shedding frequency during the cooling and mixing of tiny hot jets in air, combined with the low-frequency airflow disturbances commonly found in production lines (typically below 1 Hz). The absolute distortion field is determined by excluding the spectral characteristics of (Hz) and high-frequency random noise; the processor first performs calculation steps on each frame of the acquired image, using a digital image correlation algorithm to obtain the time series of the absolute distortion field. Subsequently, regarding The spatial domain analysis step is performed frame by frame. The machine vision algorithm uses contour matching to calculate the sample pose parameters in real time and uses them to dynamically adjust the logical symmetry axis of the asymmetric analysis, thereby calculating the time series of the difference field. Next, the system in The system automatically identifies regions of interest (ROIs) where the signal strength exceeds a preset background noise threshold; finally, it analyzes the signal strength variation over time within each ROI. The time-domain signature identification step involves performing a Fast Fourier Transform (FFT) to obtain its spectrum, followed by a decision step that compares its dominant frequency with a preset value. Hz to The Hz flicker frequency window was compared; to present the core results of the experiment in a concentrated manner, one representative test data from each of the three groups of samples was selected for explanation, see Table 1.

[0039] Table 1: Comparison of representative test results for different sample groups.

[0040]

[0041] Data analysis shows (see Table 1) that for the experimental group (defective samples), despite the presence of attitude shift and Despite the ambient airflow interference of Hz, after attitude-adaptive spatial domain analysis, a peak amplitude of Hz was still obtained near the defect location. The local thermal anomaly signal of the pixel, and its subsequent time-domain frequency analysis, shows that the dominant frequency is... Hz, falling Hz to Within a flicker frequency window of Hz, the system ultimately determined that a defect was detected; for control group 1 (intact sample), under the same environmental airflow, the peak value of the residual noise signal after spatial domain analysis was only [missing value]. Pixels, and its main frequency is The frequency was Hz, outside the flicker frequency window, and the system determined it to be defect-free; for control group 2 (intact sample with particle interference), spatial domain analysis also detected a peak amplitude caused by graphite particles. The pixel exhibits localized thermal anomalies, but its time-domain frequency analysis reveals a chaotic spectrum with no stable dominant frequency peak falling within a certain range. Hz to Within the Hz window, the system identifies it as interference and filters it out, ultimately determining it to be defect-free. Experimental results show that the process of this invention combines attitude-adaptive spatial domain differential analysis with time-domain signature recognition based on fluid dynamics characteristics. It can detect minute sealing defects in simulated industrial scenarios containing attitude shifts, ambient airflow, and transient thermal noise, and has the ability to suppress non-defect samples and interference signals.

[0042] Example 3: This example combines Figures 1 to 3 This document describes a non-destructive testing process for the sealing performance of tantalum insulators after sintering. Figure 1 As shown, the process begins with the optical acquisition step, which involves continuously capturing multiple frames of test distortion images containing a structured background. This is followed by a computational step, such as using Digital Image Correlation (DIC) or optical flow algorithms, to compare the test distortion images with a reference image and calculate pixel displacements to generate a time series of the absolute distortion field. This absolute distortion field time series is used in two parallel analysis paths: one path enters the spatial domain analysis step, which performs attitude calculation and physical attitude-based asymmetric analysis to generate a time series of the difference field; the other path is used by the global baseline dynamic arbitration step to extract far-field background distortion data and calculate the global baseline waveform. The dynamic index and differential field time series are then processed by the time-domain signature identification step. Frequency analysis is performed on the local thermal anomaly signal to obtain the time-domain frequency characteristics, and it is determined whether the frequency characteristics meet the preset flicker frequency window. The final decision depends on double confirmation, that is, when the time-domain signature identification step initially confirms the defect (yes), and the global baseline dynamic arbitration step determines that the environmental stability baseline index is below the threshold, the sealing defect confirmation is performed. If there is environmental turbulence interference (no), the confirmation result is determined to be invalid. The confirmed defect signal can also enter the defect quantification step, and its physical leakage level is estimated using the correlation model.

[0043] like Figure 2 As shown, this spectrum illustrates the power spectral density distribution of two typical signals in the frequency domain (X-axis, Y-axis per Hz). The thermal plume originating from a real defect exhibits a defect signal spectrum (solid line) with a specific dominant frequency, such as 25Hz in the figure. Its energy is concentrated within a scintillation frequency window, for example, 10Hz to 50Hz. The ambient noise spectrum (dashed line) represents a typical low-frequency interference, with its energy concentrated near 0Hz and rapidly attenuating as the frequency increases. The two are distinguishable in their frequency domain characteristics. Figure 3As shown, the system includes an optical acquisition module (production line station), a data processing and decision-making unit (industrial control computer), and an offline storage and calibration system (database). The optical acquisition module includes a structured background board / light source and a visible light camera industrial CMOS, used to continuously capture distorted images of the tantalum insulator under test at a high frame rate under high temperature cooling conditions, forming an image time series. This image time series is transmitted to the data processing and decision-making unit. The distortion field calculation module in this unit generates an absolute distortion field time series based on reference images obtained from the reference image database of the offline storage and calibration system. This series is then processed by the spatial domain analysis module to generate a differential field time series. The temporal signature identification module performs spectral analysis and scintillation frequency window comparison on the differential field series to output a local thermal anomaly signal. At the same time, the far-field distortion data extracted from the distortion field calculation module is sent to the global baseline dynamic arbitration module to monitor environmental stability and output an arbitration signal for environmental stability. Finally, the defect quantification and output module combines the local thermal anomaly signal, the arbitration signal, and the quantification correlation model data obtained from the defect quantification correlation model of the offline system to estimate the physical leakage level and provide production line feedback.

[0044] Example 4: To further illustrate the necessity of the temporal signature identification step of the present invention in distinguishing between real defects and interference sources, the following comparative experiment was conducted; this comparative experiment used the same experimental platform, sample grouping (experimental group, control group 1, control group 2), environmental conditions (including attitude displacement, ambient airflow, and graphite particle interference) and initial processing steps as Example 2, namely optical acquisition, calculation steps, and spatial domain analysis steps including attitude adaptive compensation, to generate a time series of differential fields. The only difference is that this comparative experiment omits the time-domain signature identification step and instead directly generates the signature based on the spatial domain analysis step. The decision is based on the peak signal amplitude of the localized thermal anomaly region (ROI). Specifically, an amplitude decision threshold is set, which is based on the peak background noise level of control group 1 (intact sample, containing only ambient airflow). The pixel value is determined by considering a certain margin, and set to [pixel value]. For pixels, when the peak signal amplitude of the ROI exceeds this threshold, it is determined that there is a defect.

[0045] Using this comparison method that omits the time-domain signature identification step, the three representative samples in Example 2 were processed, and the results are as follows: For the experimental group (samples with micropore defects), the peak signal amplitude of its ROI is... Pixels, exceeding The pixel threshold was therefore determined to be a defect detection, a result consistent with the physical state of the sample; for control group 1 (well-sealed sample), its ROI peak signal amplitude was... The pixels, not exceeding the threshold, were judged to be defect-free, a result consistent with the physical state of the sample; however, for control group 2 (a well-sealed sample with introduced graphite particle interference), its ROI peak signal amplitude was... Pixels, also exceeding The threshold of pixels led to the sample being incorrectly identified as a defect. This comparative test result shows that if the judgment is based solely on the signal amplitude obtained from spatial domain analysis, even if the analysis includes attitude adaptive compensation, it is still impossible to distinguish between thermal plumes originating from real defects and transient thermal interference with similar spatial signal amplitudes, thus leading to false positives.

[0046] Example 5: To ensure the detection accuracy and quantitative reliability of the process of this invention in specific industrial deployments, a standardized offline calibration procedure needs to be executed before the system goes online to determine key operating parameters and build a quantitative model; the first procedure aims to determine the scintillation frequency window applicable to a specific detection object, namely a tantalum insulator of a specific specification. Its initial state definition includes preparing a set of... The samples consist of tantalum insulators with known micropore defects of different sizes, covering a range of pore diameters. micrometers to The leakage rate of these defects, measured in micrometers, has been pre-calibrated. A calibration platform with the same optical configuration as the online inspection station—including camera, background plate, and geometry—but with controlled airflow, such as within a quiescent chamber or using laminar flow purging, has been constructed. The calibration process includes sequentially heating each defective sample to the standard inspection temperature. It is placed at the center of the calibration platform; the frame rate used is detected online using a visible light camera. Hz continuous acquisition time The image time series is processed in seconds; computational and spatial domain analysis steps are performed on the acquired sequence. At this time, no attitude compensation is required because the sample position is fixed. The differential field time series of the defect region is extracted. ;right Spectral analysis was performed on the signal strength over time, such as using Fast Fourier Transform (FFT), to record its dominant frequency. This process was repeated for all defective samples, and all measured dominant frequency values ​​were statistically analyzed. Based on frequency distribution characteristics, such as removing specific percentile data at both ends of the statistical distribution, the upper and lower limits of the frequency range were determined. Combined with the elimination of known environmental noise frequencies, such as the noise dominant frequency obtained through spectral analysis of background distortion without samples, a flicker frequency window covering the dominant frequency range of typical defective signals in this batch and avoiding the main noise frequency band was finally determined. In this calibration, this window was identified as [missing information]. Hz to Hz, and this window parameter is fixed in the decision-making steps of the online detection system.

[0047] The second procedure aims to construct the correlation model required for the defect quantification step, and its initial state definition includes preparing another set of quantities. More than 100 pieces, defect leakage rate calibrated and covering the target quantification range to The tantalum insulator samples and the testing system calibrated using the first procedure were used. The calibration process included: acquiring and processing image time series data for each sample on the calibration platform in the same manner as in the first procedure, until the differential field time series of the defect region of each sample was obtained after stabilization. ;from Extract one or more key characteristic parameters representing the signal, such as the root mean square (RMS) value of the signal amplitude within the ROI, the signal integral intensity, or the amplitude of its dominant frequency peak within the flicker frequency window; subsequently, the measured physical leakage rate is the dependent variable. One or more extracted optical signal feature parameters are used as independent variables. Statistical regression analysis methods, such as least squares method for polynomial fitting or training artificial neural networks, are used to establish the mathematical relationship between them, i.e., the correlation model. The model's prediction accuracy is evaluated through methods such as cross-validation until the model's prediction accuracy is determined, such as the coefficient of determination. Value greater than The model meets the preset engineering requirements; the resulting correlation model is stored in the online detection system and used in the online quantification stage to estimate the physical leakage level of the defect based on the optical signal characteristic parameters measured in real time.

[0048] Example 6: In a scenario simulating a sudden airflow disturbance in a production line environment, specifically during the testing process of a confirmed properly sealed tantalum insulator sample, at the [missing information] stage of the image acquisition process... seconds, brief (lasting approximately) (Seconds) An auxiliary air blowing device above the detection station is activated to simulate local asymmetric turbulence caused by a worker walking quickly past or nearby equipment starting up; the system operates according to a procedure that includes a global baseline dynamic arbitration step, and the processor obtains the differential field time series by performing a spatial domain analysis step. Simultaneously, far-field information extraction and baseline stability analysis are performed in parallel, i.e., the absolute distortion field far from the sample region is continuously calculated. The spatial variance is used to obtain the global baseline volatility index. Data shows that in Instant Within a time interval of seconds, The value is as shown by the previous average. Rise to peak Exceeded the environmental stability threshold (For example ),at the same time, The introduction of this asymmetric turbulence in the vicinity of the sample also resulted in amplitudes exceeding the ROI detection threshold. (For example The system detects a false local thermal anomaly signal (pixel), and the time-domain frequency characteristics of this signal may accidentally fall within the flicker frequency window; however, due to the intervention of the decision arbitration logic, the system detects... It has exceeded This indicates that the current detection result is affected by environmental interference. Therefore, even if subsequent steps (time-domain signature identification and decision-making) may arrive at a preliminary criterion for defect detection, the global baseline dynamic arbitration step will ultimately determine the detection result as invalid or mark it as affected by environmental interference, thereby avoiding misreporting asymmetric environmental noise as a real defect.

[0049] Example 7: When the object under test is a tantalum insulator with asymmetrical geometry (e.g., a pin structure with a special shape at one end), the dispersion differential analysis procedure as described in the specific implementation is used for detection; the optical acquisition part of the system is adjusted to use a red LED (center wavelength) nm) and blue LED (center wavelength) A dual-wavelength pulsed light source (nm) is used to... The frame rate alternates rapidly at Hz (e.g., pulse intervals are less than 100 Hz). A structured background is illuminated in milliseconds, and a visible light camera simultaneously acquires images, resulting in two time series images. and The processor performs computational steps (e.g., using the DIC algorithm) on these two sequences respectively to obtain the distortion fields corresponding to the red and blue wavelengths. and A test was conducted on an asymmetric sample with a known micropore defect. The thermal plume (TIRF) generated by this defect caused different deflection angles of light at different wavelengths due to air dispersion. Simultaneously, global ambient airflow noise (e.g., caused by a fan) was highly correlated with the distortions caused by the two wavelengths. The processor performed dispersion differential processing to calculate... , where constant Based on the air's response to the current temperature and pressure and The refractive index ratio is pre-calibrated (e.g.) ); the calculation results show that, In the middle, caused by global environmental airflow, in and The common distortion patterns present in both regions were effectively eliminated (because) ), while the wavelength-dependent dispersive anomaly signal caused by defective thermal plumes (due to The signal was preserved and relatively amplified, forming a clear Local Thermal Anomaly (ROI) signal in the region where it was located; subsequent analysis of this signal... The ROI signal in the process performs time-domain signature identification and decision-making steps, assuming its flicker frequency still falls within the range. Hz to The presence of defects can be confirmed by using a Hz window, indicating that this dispersion difference method is suitable for the detection of geometrically asymmetric components.

[0050] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0051] Finally, it should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A non-destructive testing process for the sealing performance of tantalum insulators after sintering, comprising: capturing a reference image without the structured background of the tantalum insulator using a visible light camera; and continuously capturing multiple frames of test distortion images containing the structured background, passing through the thermo-induced refractive index field of the air medium surrounding the tantalum insulator, to form an image time series; characterized in that the process... include: The calculation process involves the processor comparing each frame of the test distorted image in the image time series with the reference image to calculate a time series of the absolute distortion field. The spatial domain analysis step involves the processor performing frame-by-frame spatial domain analysis on the time series of the absolute distortion field to generate a time series of the difference field. The spatial domain analysis steps include: using machine vision algorithms to calculate the actual physical pose parameters, including translation and rotation parameters, from each frame of the test distortion image in the image time series in real time; based on the actual physical pose parameters, dynamically adjusting the definition of a logical symmetry axis, and performing asymmetry analysis on the time series of the absolute distortion field to calculate the time series of the generated difference field. The time-domain signature identification step involves the processor performing time-domain frequency analysis on the signal composed of local thermal anomalies in the time series of the differential field to obtain time-domain frequency characteristics; a scintillation frequency window based on the fluid dynamics principle of defect thermal plumes is preset. The decision-making process involves confirming a local thermal anomaly signal as originating from a sealing defect only when the time-domain frequency characteristics of the signal meet the preset conditions of the flicker frequency window.

2. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, The temporal signature identification and decision steps aim to solve the false positive problem in local thermal anomalies discovered by the spatial domain analysis steps, which cannot distinguish between thermal jets originating from real defects and thermal noise originating from interference sources. The time-domain signature identification steps specifically include: based on the principle of fluid dynamics, the processor sets the scintillation frequency window to a specific frequency range that excludes transient interference signals and low-frequency environmental thermal convection signals, with the specific frequency range being 10Hz to 50Hz; the processor performs spectral analysis on the changes in signal intensity characterizing local thermal anomalies over time in the time series of the differential field to calculate the time-domain frequency characteristics; the decision-making steps specifically include: the processor compares the calculated time-domain frequency characteristics with the specific frequency range, and only performs the operation of confirming the sealing defect when the time-domain frequency characteristics fall within the specific frequency range.

3. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, The spatial domain analysis step involves performing asymmetric analysis to compute the time series of the generated difference field, specifically including: constructing a logistic transformation function. ,in Given the coordinates of a point in the image, the logical transformation function, based on the actual physical pose parameters, is used to calculate the point's coordinates. Regarding the coordinates of the mirror point of the real-time physical symmetry axis of the tantalum insulator; by calculating the absolute distortion field at point in each frame of the time series of the absolute distortion field. With point The difference at each point is used to generate a time series of the difference field.

4. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, The first calculation step is achieved by executing a digital image correlation algorithm.

5. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, The first calculation step is achieved by executing an optical flow algorithm.

6. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, In the spatial domain analysis step, the machine vision algorithm is a contour matching algorithm.

7. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, In the spatial domain analysis step, the machine vision algorithm is a template matching algorithm.

8. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, When the tantalum insulator has a geometrically asymmetric shape, the spatial domain analysis step is replaced by a dispersion differential analysis step. The dispersion differential analysis step aims to solve the technical problem of asymmetric analysis failure caused by the geometric asymmetry of the component. The dispersion differential analysis step includes: illuminating the structured background almost simultaneously using at least two different wavelength light sources; synchronously acquiring multiple frames of test distortion images corresponding to at least two wavelengths using a visible light camera to form multiple image time series; calculating multiple distortion fields corresponding to different wavelengths by a processor; utilizing the dispersion effect of the thermally induced refractive index field on light of different wavelengths, and performing differential processing on multiple distortion fields to eliminate common distortion noise caused by global environmental airflow and amplify the dispersion anomaly signal caused by local thermal anomalies to generate a time series of differential fields.

9. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, The process also includes a global baseline dynamic arbitration step, which is designed to address the technical problem that the spatial domain analysis step misjudges environmental noise as local thermal anomalies due to sudden asymmetric turbulence in the production line environment. The global baseline dynamic arbitration step includes: extracting distortion data from the far-field background region physically far from the tantalum insulator from the time series of the absolute distortion field; calculating the spatial or temporal variance of the distortion data in the far-field background region to generate a global baseline fluctuation index; setting an environmental stability threshold; confirming the sealing defect confirmation result generated by the decision step as valid only when the global baseline fluctuation index is lower than the environmental stability threshold; and determining the sealing defect confirmation result as invalid when the global baseline fluctuation index is higher than the environmental stability threshold.

10. The non-destructive testing process for the sealing performance of tantalum insulators after sintering according to claim 1, characterized in that, The process also includes a defect quantification step, which includes: offline calibration: establishing a correlation model between the signal characteristics of the local thermal anomaly signals identified in the decision-making step and the physical leakage rate measured by the leakage inspection method; online quantification: using the correlation model, estimating the leakage level of the sealing defect based on the signal characteristics of the local thermal anomaly signals detected online.

Images