Electronic ceramic product defect recognition method based on machine vision

Abstract

The present application belongs to the technical field of electronic ceramic nondestructive testing, and is used to solve the problem that the prior art cannot utilize the phase information in strain response to realize accurate positioning of the early fatigue region of ferroelectric ceramics, and specifically relates to an electronic ceramic product defect identification method based on machine vision, which comprises the following steps: applying an alternating electric field to stimulate the electronic ceramic to be measured, and synchronously collecting the displacement signal of the surface of the electronic ceramic to be measured and the excitation voltage reference signal; performing frequency scanning according to a preset frequency sequence at a preset low electric field amplitude; automatically determining a healthy reference region based on the statistical distribution of the relaxation spectrum feature vectors of all measurement points or a preset region; and taking the relaxation peak frequency of each candidate fatigue point as a test frequency; the present application realizes accurate positioning of the real fatigue region while eliminating the pseudo defects such as surface stains and speckle decorrelation by constructing a two-dimensional feature vector of the effective piezoelectric coefficient and the phase delay under the preset electric field.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing technology for electronic ceramics, specifically a method for identifying defects in electronic ceramic products based on machine vision. Background Technology

[0002] Ferroelectric ceramic materials, such as lead zirconate titanate piezoelectric ceramics or lead lanthanum titanate electro-optic ceramics, gradually exhibit fatigue degradation during long-term service driven by alternating electric fields due to the reciprocating motion of internal domain walls. In the early stages of this fatigue, domain walls in localized areas are pinned by defects, leading to changes in the piezoelectric response characteristics of that region, while the macroscopic electrical properties often remain within the normal range. For fatigue detection of these materials, the industry currently employs two common methods: one is offline electrical testing, which assesses the degree of fatigue by measuring the overall change in hysteresis loop or piezoelectric coefficient; however, this method only provides average information for the entire device and cannot pinpoint microscopic fatigue areas. The other is surface optical observation, which uses a microscope to directly examine the ceramic surface for microcracks; however, in the early stages of fatigue, domain wall pinning has not yet caused any visible changes in surface morphology, rendering optical observation completely ineffective.

[0003] Some studies have attempted to use laser vibration measurement technology to detect the strain response of piezoelectric ceramics. However, existing methods typically only collect displacement amplitudes at a single frequency or electric field amplitude and compare the amplitude differences between positive and negative excitations. This approach ignores the phase information of the strain response relative to the driving electric field. In fact, domain pinning is primarily manifested as an abnormal increase in phase hysteresis, rather than a significant decrease in amplitude, and the variation of phase hysteresis with driving frequency and electric field amplitude is highly specific. Because existing technologies do not use phase as a detection dimension, early fatigue signals are overwhelmed by system noise and fluctuations in normal regions, making it impossible to achieve high spatial resolution fatigue region localization, let alone distinguish whether fatigue is dominated by reversible domain pinning or irreversible microcrack propagation.

[0004] Therefore, how to utilize the phase information in the strain response to accurately locate the early fatigue region of ferroelectric ceramics is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a machine vision-based method for identifying defects in electronic ceramic products, which solves the problem that existing technologies cannot utilize phase information in strain response to accurately locate the early fatigue region of ferroelectric ceramics.

[0006] The technical problem to be solved by this invention is: how to provide a machine vision-based method for identifying defects in electronic ceramic products that can accurately locate the early fatigue region of ferroelectric ceramics by utilizing phase information in the strain response.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] Machine vision-based defect identification methods for electronic ceramic products include:

[0009] An alternating electric field is applied to excite the electronic ceramic under test, and the displacement signal and excitation voltage reference signal on the surface of the electronic ceramic under test are acquired simultaneously. Synchronous phase-locked demodulation is performed on the displacement signal and excitation voltage reference signal to extract the fundamental amplitude and fundamental phase delay at each test frequency and each test electric field amplitude.

[0010] Under a preset low electric field amplitude, frequency scanning is performed according to a preset frequency sequence. The relaxation spectrum feature vector of each measurement point is extracted based on the fundamental phase delay at each frequency. The relaxation spectrum feature vector includes at least the relaxation peak frequency.

[0011] Based on the statistical distribution of relaxation spectrum feature vectors of all measurement points or preset areas, a healthy reference area is automatically determined, and candidate fatigue points in the measurement points are marked based on the healthy reference area and preset rules.

[0012] For each candidate fatigue point, the relaxation peak frequency of the candidate fatigue point itself is used as the test frequency to perform electric field amplitude scanning and obtain the amplitude curve of the candidate fatigue point under multiple electric field amplitudes.

[0013] The effective piezoelectric coefficient under the preset electric field is extracted from the amplitude curve. The effective piezoelectric coefficient and the phase delay under the preset electric field are used to form a two-dimensional feature vector. The deviation of the two-dimensional feature vector from the two-dimensional feature distribution corresponding to the healthy reference area is calculated. When the deviation exceeds the preset deviation threshold, the candidate fatigue point is determined to be a real fatigue defect area.

[0014] The present invention has the following beneficial effects:

[0015] 1. By using the instantaneous phase of the excitation voltage as the absolute reference for phase-locked demodulation, the fundamental amplitude and fundamental phase delay are synchronously extracted from the displacement signal. The phase delay is used as the core detection dimension, which solves the problem that the early fatigue signal is submerged by noise due to the discard of phase information in the existing technology. It achieves high-sensitivity capture of phase anomalies caused by domain wall pinning. The detection sensitivity is more than an order of magnitude higher than that of the method that only relies on the amplitude.

[0016] 2. By performing low-field frequency scanning on each measurement point and extracting the relaxation peak frequency, half-peak width, and low-frequency limiting phase, and combining the full-field statistical distribution to automatically calibrate the healthy reference area, the influence of batch differences and measurement environment fluctuations on threshold setting is eliminated, achieving adaptive benchmark correction without manual intervention and avoiding over-detection or under-detection caused by fixed thresholds.

[0017] 3. For each candidate fatigue point, the relaxation peak frequency is used as the test frequency. The inflection point electric field is dynamically determined by real-time monitoring of the phase delay increment. Based on this, a non-uniform electric field amplitude sequence is generated and the number of integration cycles is adaptively adjusted to match the measurement parameters with the actual state of the measured point. This solves the problem that fixed parameter scanning cannot take into account the differences in fatigue degree in different regions. The measurement time is shortened by more than 40% while ensuring the detection accuracy.

[0018] 4. By constructing a two-dimensional feature vector of the effective piezoelectric coefficient and the phase delay under a preset electric field, the Mahalanobis distance between the vector and the distribution of the healthy reference area is calculated. This achieves accurate positioning of the real fatigue area while eliminating pseudo-defects such as surface stains, speckle, and decorrelation. Furthermore, based on the ratio of the electric field increment at the inflection point to the slope, fatigue is classified into mild, moderate, and severe levels. Combined with pre-calibrated failure data, the remaining number of cycles is estimated, providing quantifiable output basis for maintenance decisions and life prediction of ferroelectric ceramic actuators. Attached Figure Description

[0019] 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, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is the main flowchart of Embodiment 1 of the present invention;

[0021] Figure 2 This is a flowchart of the lifetime prediction sub-process of Embodiment 1 of the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In the field of electronic ceramics manufacturing, especially in the quality control of multilayer piezoelectric actuators and microwave dielectric ceramic components, the detection of early fatigue defects is directly related to the long-term reliability of the devices. The physical essence of this type of fatigue is local domain wall pinning and polarization fatigue. Its early signs are not surface cracks or macroscopic deformation, but rather abnormal phase lag of the piezoelectric strain response relative to the driving electric field and drift of the relaxation frequency. However, existing detection methods mainly rely on offline electrical testing or static optical observation. Offline electrical testing can only obtain the average electrical parameters of the entire device and cannot locate microscopic fatigue regions; static optical observation, on the other hand, cannot pinpoint the fatigue area. Initial complete failure – domain wall pinning has not caused any visible changes in surface morphology, and the strain amplitude change in the fatigue region is much smaller than the phase hysteresis change; although some existing technologies have attempted to introduce laser vibration measurement, they only collect displacement amplitude under a single frequency or a single electric field and simply compare the amplitude differences under positive and negative excitation, completely discarding the phase information of the displacement signal relative to the driving electric field; since the instantaneous phase of the excitation signal is not used as the demodulation reference, the amplitude and phase crosstalk each other, causing the weak phase anomaly signal of early fatigue to be submerged by system noise, and it is impossible to obtain spatially resolved phase-frequency spectrum characteristics.

[0024] For example, in the aging test of lead zirconate titanate piezoelectric ceramic actuators, the system only measures the surface displacement amplitude after applying positive and negative electric fields. When no visible cracks have yet formed in the fatigue area, the amplitude change is minimal, and the system misjudges it as healthy. However, the actual phase hysteresis in this area has increased significantly, but its change pattern shifts nonlinearly with the driving frequency, and the relaxation peak frequency drift varies at different locations. The existing scanning method with fixed frequency and fixed electric field amplitude cannot adaptively match the actual relaxation response of each area, resulting in either missed detection of fatigue areas or a large number of false defect alarms due to improper global threshold settings. Furthermore, even if an anomaly is detected in a certain area, the existing solution cannot determine whether the anomaly is caused by reversible domain wall pinning or has developed to the irreversible microcrack stage—the former may be recovered through electric field annealing, while the latter prohibits annealing and requires replacement. If the system misjudges microcracks as domain wall pinning and performs annealing, it will accelerate device failure. Conversely, misjudges reversible pinning as microcracks, leading to unnecessary overall replacement and wasted costs.

[0025] If the above problems are not resolved, the detection system will continue to lose its objective ability to identify early fatigue. Specifically, the loss of phase information will cause the system to rely solely on amplitude thresholds, leading to missed detections or failures in locating fatigue areas. Open-loop scanning with fixed parameters cannot adapt to the dispersion of relaxation characteristics in different regions, resulting in a decrease in detection specificity. Furthermore, the lack of fatigue mechanism attribution analysis will render subsequent maintenance decisions without a scientific basis. Consequently, the inaccuracy of detection results will systematically hinder the development of reliability assessment and life prediction technologies for electronic ceramic devices, affecting the quality control of key components such as piezoelectric actuators and microwave ceramic filters in high-end equipment.

[0026] Example 1: As Figure 1-2 As shown, a machine vision-based defect identification method for electronic ceramic products includes:

[0027] Step S1: Apply an alternating electric field to excite the electronic ceramic under test, and simultaneously acquire the displacement signal and excitation voltage reference signal on the surface of the electronic ceramic under test; perform synchronous phase-locked demodulation on the displacement signal and excitation voltage reference signal, and extract the fundamental amplitude and fundamental phase delay at each test frequency and each test electric field amplitude.

[0028] In this embodiment, the electronic ceramic to be tested is a pre-polarized lead zirconate titanate piezoelectric ceramic sheet with a thickness of 1.0 mm and a coercive field. The voltage is 1.2 kV / mm². The ceramic plate is placed on the measuring platform of a laser vibrometer, with its upper and lower surfaces connected to the output of a programmable bipolar high-voltage amplifier via flexible electrodes. The amplifier outputs a sinusoidal voltage with a frequency range of 0.1 Hz to 200 Hz and an amplitude range of 0 V to 500 V. It also synchronously outputs a TTL trigger signal in phase and frequency with the voltage waveform, as well as an analog signal representing the instantaneous voltage value. The control computer simultaneously acquires three signals via a data acquisition card at a sampling rate of 20 kHz: excitation voltage reference signal... Displacement signal output by laser vibrometer and ambient temperature sensor signal To ensure the phase accuracy of subsequent phase-locked demodulation, the acquisition trigger of all channels is controlled by the same TTL signal, and the zero-crossing time of the voltage reference signal is recorded as the phase reference.

[0029] To eliminate the effects of environmental vibration and thermal drift, the acquired displacement signals were processed. Preprocessing: First, a high-pass filter with a cutoff frequency of 0.1 Hz is used to remove the DC drift component, resulting in... Then based on the temperature signal Change and the coefficient of thermal expansion of ceramic materials (Pre-calibrated to 2.3 × 10) -6 Calculate the thermal expansion displacement compensation using Kelvin, from... Subtracting this compensation amount from the result yields the preprocessed displacement signal. ;

[0030] Specifically: Let the ambient temperature at the start of data collection be... The current temperature is Then the change in temperature Thermal expansion displacement compensation amount The calculation formula is:

[0031] ;

[0032] Where L is the characteristic dimension of the ceramic sheet under test in the measurement direction (in this embodiment, the thickness is taken as 1.0 mm); since the laser vibrometer measures the surface normal displacement, and the thermal expansion in the thickness direction directly affects this displacement, the thickness L is directly used; the preprocessed displacement signal Subtracting this compensation amount yields the temperature-compensated displacement signal. .

[0033] Next, the preprocessed displacement signal With excitation voltage reference signal Perform synchronous phase-locked demodulation to extract the fundamental amplitude and phase delay at the current test frequency f and electric field amplitude E; let the current test frequency be f, in Hertz, and the corresponding angular frequency be... The excitation voltage reference signal can be expressed as: ;

[0034] in Given the amplitude, The initial phase is known (obtained through zero-crossing detection); in order to... To extract the fundamental component with the same frequency as the excitation, two unit amplitude reference sequences need to be generated: an in-phase reference sequence and a non-in-phase reference sequence. and orthogonal reference sequence These two sequences are generated in real time by the control computer, and their frequency f is strictly consistent with the excitation signal.

[0035] The core processing steps of phase-locked demodulation are as follows: Acquire M consecutive complete cycles (M is 10 in this embodiment) Data, number of sampling points ,in For a period of time, Given the sampling rate; calculate the in-phase component X and the quadrature component Y:

[0036] ;

[0037] ;

[0038] In practical digital implementations, the above integral is approximated by summation:

[0039] ;

[0040] ;

[0041] in The displacement value of the nth sampling point. , Through this calculation, the in-phase component X and the quadrature component Y represent the intensity of the components in the displacement signal that are in phase with and quadrature with the excitation voltage, respectively.

[0042] The fundamental amplitude A is calculated by the following formula: ;

[0043] Fundamental phase Calculated by the following formula:

[0044] ;

[0045] in The range of function return values ​​is ;Should This indicates that the fundamental frequency of the displacement signal is relative to the generated in-phase reference sequence. The phase of the excitation voltage; to obtain the true phase delay of the displacement signal relative to the excitation voltage reference signal, the initial phase of the excitation voltage needs to be subtracted. : ;

[0046] Because the result of subtraction may exceed The range requires angle winding correction: if Then subtract ;like Then add This causes the final phase to be delayed. Strictly implement Interval; the unit of this phase delay is radians.

[0047] For example, assuming the test frequency f = 10 Hz, the electric field amplitude E = 0.2 kV / mm (corresponding to a voltage of 200 V), and the data acquisition is M = 10 cycles, the sampling rate is... If the frequency is kilohertz, then the number of sampling points per cycle is 2000, and the total number of sampling points N = 20000; after calculation, X = 1.2 nanometers and Y = 0.5 nanometers, then the fundamental amplitude is... nanometer; radians; known initial phase of excitation voltage radians, then radians, this value is in Within the range, no correction is needed; if the calculated It is 3.0 radians. radians, then radians, since 2.9 is less than (Approximately 3.1416), still within the range; if If it is 3.1 radians, then radians, still less than Only in extreme cases, such as near and Correction is only required when the result is negative, which will not be elaborated here.

[0048] Through the above phase-locked demodulation process, the corresponding fundamental amplitude can be obtained for each preset frequency f and electric field amplitude E. and fundamental phase delay In subsequent steps, these quantities will be used for low-frequency scanning and Rayleigh electric field scanning to extract relaxation spectrum features and fatigue-related parameters. All thresholds involved in this step (such as M=10 cycles, sampling rate 20 kHz) are exemplary values ​​in this embodiment. In practical applications, they can be adjusted according to the characteristics of ceramic materials and the performance of measuring equipment, as long as the signal-to-noise ratio requirements of phase-locked demodulation are met.

[0049] In this embodiment, the system presets the following key parameters, all of which were obtained through multiple measurements of healthy samples under the same test conditions in preliminary experiments:

[0050] Noise threshold for phase delay measurement The value is 0.00524 radians. This value is the upper limit of the standard deviation of the phase measurement values ​​in the healthy region under static and unexcited conditions. It is used in step S4 to determine whether the number of integration cycles needs to be increased.

[0051] Minimum identifiable peak height The radian value is used in step S2 to determine whether the relaxation peak can be effectively identified.

[0052] The remaining thresholds (such as the relative deviation threshold, Mahalanobis distance threshold, etc.) are set according to the specific values ​​in steps S3 and S5.

[0053] Step S2: Under a preset low electric field amplitude, perform frequency scanning according to a preset frequency sequence, and extract the relaxation spectrum feature vector of each measurement point based on the fundamental phase delay at each frequency. The relaxation spectrum feature vector includes at least the relaxation peak frequency.

[0054] In this embodiment, the value of the low electric field amplitude needs to be determined first. For ferroelectric ceramic materials, the low electric field amplitude should ensure that the domain wall motion is in the linear response region and that irreversible domain flipping does not occur. In this embodiment, the low electric field amplitude is set as follows: ,in The coercive field of the material; the known coercive field kilovolts per millimeter, therefore kV / mm; this value ensures that the piezoelectric strain response is mainly contributed by the intrinsic lattice piezoelectric effect and reversible domain wall vibrations, excluding nonlinear domain flipping interference.

[0055] Preset frequency sequence The relaxation peaks of ferroelectric ceramics are determined based on their dielectric relaxation characteristics. Generally, the relaxation peaks of ferroelectric ceramics appear in the range of several hertz to hundreds of hertz. In this embodiment, the selected frequency sequence is: 1 hertz, 2 hertz, 5 hertz, 10 hertz, 20 hertz, 50 hertz, and 100 hertz, for a total of seven frequency points. The sequence is appropriately densified in the low-frequency band to accurately capture the relaxation peak morphology.

[0056] When performing a frequency scan, for each frequency value in the frequency sequence A sinusoidal alternating electric field of the corresponding frequency is applied according to the method described in step S1, with the electric field amplitude fixed at a certain value. kV / mm; at each frequency, displacement signals for 20 complete cycles (M=20) are continuously acquired, and synchronous phase-locked demodulation is performed to obtain the fundamental phase delay at that frequency. To simplify the notation, let's call it... Seven discrete data points were ultimately obtained. .

[0057] Next, based on these discrete data points, the relaxation spectrum feature vector for each measurement point is extracted. This feature vector includes the relaxation peak frequency. Half-peak width and low-frequency limiting phase The extraction process consists of the following sub-steps:

[0058] First, determine the low-frequency limiting phase. In frequency scanning, the lowest frequency The phase delay at Hertz typically corresponds to the quasi-static response of domain wall motion and can be used as the phase baseline; therefore, we take: If, at the lowest test frequency of 1 Hz, the phase still shows a decreasing trend with decreasing frequency (which can be judged by calculating the slope of the last two frequency points, with a slope greater than 0.01 radians / Hz), it indicates that the relaxation peak may be below the scanning range; at this time, the system automatically extends the lower limit of the frequency scan to 0.5 Hz or 0.2 Hz, or obtains the low-frequency limit phase by fitting the Debye model and extrapolating.

[0059] Then, the discrete phase delay-frequency data points need to be interpolated to obtain a continuous curve for accurately finding the relaxation peak; this embodiment uses the cubic spline interpolation method; specifically, seven data points are known. Cubic spline interpolation constructs a piecewise cubic polynomial function. Satisfy each sub-interval The above is a cubic polynomial with continuous first and second derivatives over the entire interval; the interpolation process is a standard numerical algorithm, which will not be elaborated here; after interpolation, a continuous smooth curve with a frequency range from 1 Hz to 100 Hz is obtained. .

[0060] In the interpolation curve The frequency corresponding to the maximum phase delay is located by peak search, which is the relaxation peak frequency. This embodiment uses the golden section search algorithm to find the global maximum value within the interpolation interval; firstly, the search interval is determined as follows: Calculate the function value within the interval, gradually narrowing the interval until the interval width is less than the preset precision (e.g., 0.1 Hz), and finally obtain the result. satisfy: Let the phase delay at the relaxation peak be denoted as . .

[0061] Next, the phase peak height is calculated. : ;

[0062] Half peak width Defined as the phase delay dropping to half of the peak value (i.e. The width of the frequency range corresponding to the relaxation peak; specifically, to the left of the relaxation peak (frequency less than...). (within the range), find the frequency point. satisfy: ;

[0063] and For the closest The left intersection point; if it lies between discrete points, the exact solution is obtained through interpolation curves; similarly, on the right side of the relaxation peak (frequency greater than... (within the range), find the frequency point. Satisfying the same equation, and being the closest The right intersection point; then the half-peak width is: ;

[0064] To ensure calculation stability, if the peak height If the value is too small (e.g., less than 0.01 radians), the half-peak width is unreliable. In this case, the half-peak width should be set to a large preset value (e.g., 100 Hz) or marked as an invalid point. In this embodiment, a minimum identifiable peak height is set. It is 0.02 radians.

[0065] The following is a specific numerical example; assuming that after the low-field frequency scan in step S1, the following seven data points are obtained: Hertz time radian; Hertz time radian; Hertz time radian; Hertz time radian; Hertz time radian; Hertz time radian; Hertz time radians; then the low-frequency limiting phase. Radius; After performing cubic spline interpolation on the above points, peak search revealed that the maximum phase occurs at approximately 18 Hz, hence the relaxation peak frequency. hertz, radians; phase peak height radians; half-peak phase is Radius; on the interpolation curve, find on the left. Hertz, found on the right Hertz, then half-peak width hertz.

[0066] The output relaxation spectrum eigenvector is =[18 Hz, 38.3 Hz, 0.05 radians]; This feature vector will be used for the automatic determination of the healthy reference area and the marking of candidate fatigue points in subsequent steps; It is worth noting that for each measurement point of the ceramic sheet (i.e., each pixel or each region of interest), the above frequency scanning and feature extraction process needs to be repeated to obtain the feature vector distribution of the entire field; In this embodiment, the laser vibrometer adopts a point-by-point scanning method with a step size of 0.1 mm, covering the entire surface of the ceramic sheet to form a feature map.

[0067] If more than half of the frequency points at a given measurement point fail to be demodulated during frequency scanning (e.g., the amplitude is zero or the phase is invalid), then that point is marked as an invalid measurement point and will not participate in subsequent health reference area calibration and candidate fatigue point screening. Invalid measurement points usually correspond to severe surface contamination, poor electrode contact, or measurement system failure, and their locations should be listed in the report for manual verification.

[0068] Step S3: Based on the statistical distribution of relaxation spectrum feature vectors in all measurement points or preset areas, automatically determine the healthy reference area, and mark the candidate fatigue points in the measurement points based on the healthy reference area and preset rules.

[0069] In this embodiment, the relaxation spectrum feature vector of each measurement point on the ceramic sheet surface has been obtained through step S2. The feature vector of each measurement point contains three components: relaxation peak frequency. (Unit: Hertz) Half-peak width (Unit: Hertz) and low-frequency limiting phase (Unit: radians); Assuming a total of Let there be n measurement points, and denote the feature vector of the i-th measurement point as . The goal of this step is to first automatically identify a healthy area from these measurement points as a reference benchmark, and then screen out candidate points that may be fatigued based on this benchmark.

[0070] First, the automatic determination process of the health reference area is executed; this process does not rely on any subjective point selection, but rather on adaptive regional positioning based on the statistical distribution characteristics of the entire field.

[0071] Calculate the median of the relaxation peak frequencies at all measurement points; then... relaxation peak frequency value Sort in ascending order, if If the number is odd, take the middle value; if the number is even, take the average of the two middle values. This median is denoted as . Let the first preset threshold be... (i.e., 10%); select measurement points that satisfy the following formula to form a candidate reference region. : ;

[0072] This condition requires that the relative deviation of the candidate point from the median be less than 10% in order to exclude points with significant frequency shifts due to fatigue or measurement abnormalities.

[0073] In the candidate reference area Within, select the low-frequency limiting phase. The smallest measurement point is selected as the initial reference point; the index of this point is denoted as . ,Right now: ;

[0074] The reason for choosing the minimum low-frequency limiting phase is that in normal and healthy ferroelectric ceramics, the low-frequency limiting phase is usually small (close to zero), while in fatigue or defect regions, the low-frequency loss increases due to domain wall pinning, and the phase will increase significantly; therefore, the minimum value is taken as the seed point of the healthy region.

[0075] Using the initially selected reference point as the center, a neighborhood window of a preset size is extracted; in this embodiment, the preset window size is 5×5, that is, including the neighborhood window of the selected reference point. The system covers 25 measurement points, extending 2 steps upwards, downwards, left, and right from the center (assuming the measurement points are arranged in a regular rectangular grid). If the initially selected reference point is located at the edge, causing the window to exceed the field of view, then all existing points are used. The arithmetic mean of the relaxation peak frequencies of all measurement points within this window is calculated. and sample standard deviation :

[0076] , ;

[0077] Where K is the number of measurement points within the window.

[0078] Set the second preset threshold as (i.e., 5%); if satisfied If the standard deviation is less than 5% of the mean, then the measurement points within the window are considered to be highly consistent in terms of relaxation peak frequency, which is consistent with the characteristics of healthy materials. The arithmetic mean of the relaxation spectrum eigenvectors of all measurement points within the window is taken as the reference benchmark for the healthy reference area.

[0079] ;

[0080] If the above conditions are not met, it indicates that there are significant fluctuations in the area near the initially selected reference point, which may be in a transition zone or at the edge of a defect. In this case, the neighborhood window size is expanded, for example, by trying 7×7, 9×9, and 11×11 in sequence, and the standard deviation condition is repeatedly calculated until the condition is met or the window is expanded to the maximum preset size (15×15 in this embodiment). If the condition still cannot be met, the pre-stored factory database is called. This database is the average value of the feature vector of the healthy reference area calculated by scanning the entire surface of standard qualified samples using the same testing system before the same batch of ceramics leaves the factory, and is pre-stored in the device memory. At this time, the factory benchmark is used as the reference benchmark for the healthy reference area.

[0081] The method for constructing the factory database is as follows: Before the same batch of ceramics leaves the factory, a full-surface scan of the standard qualified samples is performed using the same testing system as in this embodiment. The relaxation spectrum feature vector of each measurement point is calculated according to the method of steps S2 to S3, and the statistical median or average value of the entire field is taken as the factory reference benchmark for the batch. The database uses the batch number as the index key and stores the following contents: the average relaxation peak frequency, the average half-peak width, the average low-frequency limiting phase, and the reference values ​​of the inflection point electric field and slope ratio. When a stable healthy reference area cannot be formed from the current sample itself during online detection (for example, expanding the window to the maximum preset size still does not meet the standard deviation condition), the system automatically reads the batch number of the ceramic being tested and calls the corresponding factory benchmark as the reference benchmark for the healthy reference area.

[0082] After the above process, the reference benchmark for the health reference zone is finally obtained:

[0083] , , ;

[0084] Next, candidate fatigue points in the measurement points are marked based on the health reference area and preset rules; for each measurement point i, its three relative deviations from the reference benchmark are calculated:

[0085] First relative deviation (relative deviation of relaxation peak frequency) ;

[0086] Second relative deviation (half-width relative deviation) ;

[0087] in For a very small positive number (e.g.) This is used to avoid division by zero errors; when the half-width of the reference is zero, this process ensures the stability of the calculation.

[0088] Third relative deviation (low-frequency limiting phase relative deviation) ,in The preset minimum phase reference value is 0.01 radians.

[0089] Set three deviation thresholds: First deviation threshold Second deviation threshold The third deviation threshold These threshold values ​​are based on statistical analysis of a large number of healthy and fatigued samples, ensuring that most of the normal fluctuations in the healthy region fall within the threshold, while at least one deviation in the truly fatigued region exceeds the threshold.

[0090] Judgment rule: When any of the following conditions are met: , , The measurement point is marked as a candidate fatigue point and added to the candidate fatigue point queue. Otherwise, it is considered a healthy point and will not proceed to subsequent detailed measurements.

[0091] The following is an example using numerical data; assuming the health reference zone baseline obtained in step S2 is: hertz, hertz, radian;

[0092] For a given measurement point, its feature vector is: hertz, hertz, radian;

[0093] Calculate the relative deviations:

[0094] greater than ;

[0095] less than ;

[0096] much larger .

[0097] Since both the first and third relative deviations exceed their corresponding thresholds, the measurement point is marked as a candidate fatigue point. .

[0098] For another measurement point, its eigenvector is: hertz, hertz, radian;

[0099] Calculated , , Note that the third relative deviation is exactly equal to the threshold. According to the judgment rules, it is only triggered when it is greater than the threshold; therefore, this point is not marked as a candidate fatigue point.

[0100] After all measurement points in the field have undergone the above-mentioned determination, a candidate fatigue point queue is output. , as input for step S4; if If the result is empty, it is determined that there are no signs of early fatigue in the entire ceramic sheet under test, and the process ends; otherwise, proceed to subsequent fine measurement.

[0101] After completing the automatic calibration of the health reference area, the system performs a fine electric field amplitude scan at each measurement point within the reference area using the same method as in step S4 (the test frequency uses the relaxation peak frequency of each point, and the electric field amplitude sequence is the same), and then applies the preset electric field... The effective piezoelectric coefficient and phase delay at each point are obtained by interpolation, thereby constructing the two-dimensional characteristic distribution statistical parameters of the healthy reference area.

[0102] Step S4: For each candidate fatigue point, use the relaxation peak frequency of the candidate fatigue point itself as the test frequency, perform electric field amplitude scanning, and obtain the amplitude curve of the candidate fatigue point under multiple electric field amplitudes.

[0103] In this embodiment, a candidate fatigue point queue is obtained through step S3. For each candidate fatigue point in the queue, an adaptive fine electric field amplitude scan must be performed to obtain the piezoelectric response characteristics of that point under different electric field amplitudes. The core of this step is that the test frequency no longer uses a uniform global reference frequency, but instead adopts the relaxation peak frequency of the candidate fatigue point itself. (This value has been saved in step S2), and the step size of the electric field amplitude and the number of integration cycles for each amplitude are dynamically adjusted according to the real-time measurement results to balance measurement accuracy and efficiency.

[0104] Let the candidate fatigue point currently being processed be numbered p, and its own relaxation peak frequency be... This value originates from the relaxation spectrum eigenvector of the measurement point in step S2; the test frequency is set as follows: ;

[0105] The first step in this process is to determine the inflection point electric field of the candidate fatigue point. The physical meaning of the inflection point electric field is: when the electric field amplitude is below this value, the phase delay is approximately constant (or changes slowly) with the electric field; when the electric field amplitude exceeds this value, the phase delay begins to increase significantly; in order to accurately locate the inflection point, the system performs a pre-scanning process.

[0106] Pre-scan from initial electric field amplitude In the beginning, among them kilovolts per millimeter is the coercive field of the material, therefore kV / mm; using fixed step size Increasing in kilovolts per millimeter; at each electric field amplitude Below, the applied frequency is For a sinusoidal alternating electric field, following the phase-locked demodulation method in step S1, 10 complete cycles are acquired, and the fundamental phase delay under this electric field is calculated. Record the phase delay increments corresponding to two adjacent electric field amplitudes sequentially: ;

[0107] in Starting from k=2 (i.e., the third electric field point), compare the current increment. Compared with the previous increment When the condition is met twice consecutively At this point, it is considered that a significant jump has occurred in the phase delay, and the current electric field amplitude is then... Determined as the inflection point electric field To prevent false triggering due to noise, confirmation is only granted after two consecutive successful conditions are met; the pre-scan continues until the electric field amplitude reaches [value missing]. kV / mm; if no condition is met within this range, then the inflection point electric field is set to .

[0108] For example, suppose the following data is obtained from a pre-scan at a candidate fatigue point:

[0109] kV / mmHm radian;

[0110] kV / mmHm radians, increment ;

[0111] kV / mmHm radians, increment ;

[0112] kV / mmHm radians, increment ;

[0113] kV / mmHm radians, increment .

[0114] at this time The previous increment ,satisfy (Equal to is not considered greater than), continue to observe;

[0115] kV / mmHm radians, increment greater than (Equal to is not considered greater than), and it has not exceeded consecutively;

[0116] kV / mmHm radians, increment No greater than If the condition is not met, but if it exceeds twice in a row in a subsequent period, then it is recorded. To simplify the example, assume that if two consecutive increments exceed twice the previous increment in a higher electric field, then the corresponding electric field is taken as the inflection point.

[0117] In reality, the inflection point electric field of actual fatigue materials may appear at approximately Nearby; for demonstration purposes, this embodiment assumes that no situation meeting the conditions is found during the pre-scan, and the final setting is... kilovolts per millimeter.

[0118] After obtaining the inflection point electric field, the system constructs a non-uniform fine electric field amplitude sequence; let the maximum test electric field... In this embodiment, Multiplying by 1.2 gives 0.864, which exceeds... Therefore, take kilovolts per millimeter; then six voltage levels are generated, determined proportionally based on the inflection point electric field:

[0119] , , , , , ;

[0120] because The calculation yields:

[0121] , , , , , kilovolts per millimeter; these values ​​are all less than This meets the requirements; note that the electric field points near the inflection point (e.g., 0.432 and 0.540) are spaced at 0.108 kV / mm, while the low electric field region (0.108 to 0.324) is spaced at 0.108, which is practically uniform, but the sampling density near the inflection point is still sufficient relative to the absolute field value; if the electric field at the inflection point is very small (e.g., less than 0.05 × 10⁻⁶), the sampling density is sufficient. If the absolute electric field sequence is used instead, it is: 0.05, 0.10, 0.15, 0.20, 0.25, 0.30 times. This is to ensure that there are enough sampling points in the low field.

[0122] For each electric field amplitude in the fine electric field amplitude sequence The system needs to adaptively determine the number of integration cycles under this voltage to ensure the statistical accuracy of the phase delay measurement; the specific process is as follows: the initial number of the first cycle is set to... One complete cycle; the system applied frequency is Amplitude sinusoidal electric field, continuously collected For each cycle, phase-locked demodulation is performed to obtain the phase delay value within that cycle. ; calculate this Sample standard deviation of each phase delay value:

[0123] ;

[0124] in The average value is used; based on the system's preset noise threshold. Degrees (approximately 0.00524 radians when converted); if Then accept the current measurement and take this. The average value of each period is used as the final phase delay. Simultaneously calculate the corresponding average amplitude. (The same periodic average is used).

[0125] like Then increase to the second cycle number For each period, re-collect data and calculate the standard deviation. If it is still greater than Then it increases to the third cycle number. The data is collected and calculated over several cycles; 50 cycles is the maximum number of cycles set in this embodiment. Regardless of whether the standard deviation meets the condition, the average value of 50 cycles is used as the final result. This adaptive mechanism ensures fast measurement when the signal quality is good, and improves the signal-to-noise ratio by increasing the integration time when the noise is high. When the maximum integration cycle number is reached... Even under these conditions, the standard deviation of the phase delay still exceeds the noise threshold. At that time, the phase delay measurement at this voltage is marked as low confidence and assigned a weighting factor. In the subsequent Mahalanobis distance calculation, the two-dimensional feature vector of this candidate fatigue point... The standardized value will be multiplied by a weighting factor, i.e.:

[0126] , ;

[0127] Weighting factors The value range is (0,1], and in this embodiment it is 0.5; if two or more consecutive electric field amplitude points are marked as low confidence, the candidate fatigue point is directly judged as a false defect and will no longer participate in the subsequent fatigue judgment.

[0128] After completing measurements at all six electric field amplitudes, the system obtains a fine amplitude curve for the candidate fatigue point: a series of discrete points. and the corresponding phase delay curve The amplitude curve will be used in the subsequent step S5 to extract the effective piezoelectric coefficient; the phase delay curve will be used to extract the inflection point electric field and slope ratio (see the supplementary processing before step S5 for details).

[0129] It is worth noting that, for the same candidate fatigue point, the amplitude data under each electric field needs to be recorded simultaneously during the fine electric field scanning process described above; the amplitude calculation is completed synchronously with the phase-locked demodulation: after calculating the in-phase component X and the quadrature component Y, the amplitude... Then, it is obtained by averaging using the same periodicity as the phase. ; final output This is the amplitude curve of the candidate fatigue point under multiple electric field amplitudes; this curve will be passed to step S5 to extract the effective piezoelectric coefficient.

[0130] Step S5: Extract the effective piezoelectric coefficient under the preset electric field from the amplitude curve, construct a two-dimensional feature vector from the effective piezoelectric coefficient and the phase delay under the preset electric field, calculate the degree of deviation between the two-dimensional feature vector and the corresponding two-dimensional feature distribution of the healthy reference area, and determine the candidate fatigue point as a real fatigue defect area when the degree of deviation exceeds the preset deviation threshold.

[0131] In this embodiment, for each candidate fatigue point, the value of that point in the fine electric field amplitude sequence has already been obtained in step S4. to The amplitude curve below and the corresponding phase delay curve This step first requires extracting the inflection point electric field from the phase delay curve. and slope ratio These two parameters will be used for subsequent fatigue grading and will also provide a reference for the two-dimensional feature distribution of the healthy reference area. Then, the effective piezoelectric coefficient under the preset electric field is extracted from the amplitude curve and a two-dimensional feature vector is formed with the phase delay under the same preset electric field. Finally, the Mahalanobis distance between the two-dimensional feature vector and the corresponding two-dimensional feature distribution of the healthy reference area is calculated, and a threshold comparison is used to determine whether it is a real fatigue defect.

[0132] First, the extraction process of the inflection point electric field and slope ratio is described; based on the physical properties of ferroelectric ceramics in the Rayleigh region, the phase delay... The relationship with the electric field amplitude E exhibits a piecewise linear pattern before and after the inflection point; for the current candidate fatigue point, the phase delay curve obtained from its fine measurement contains six discrete points. ,in Arrange these points in ascending order of electric field amplitude; the electric field sequence is denoted as... The corresponding phase sequence is The system uses a piecewise linear regression method to automatically find the segmentation point (i.e., candidate inflection point).

[0133] Specifically, for each possible splitting position k (where k = 2, 3, 4, 5, i.e., the splitting point is located at...), and (between), dividing the data points into low electric field segments. and high electric field section Perform univariate linear regression on the data points (at least two points) in the low electric field segment to fit a straight line. To obtain a low field slope Perform univariate linear regression on the data points in the high electric field segment to fit a straight line. The high-field slope is obtained. Calculate the sum of squared residuals from the two regressions, and select the split point that minimizes the total sum of squared residuals as the optimal split point; the maximum electric field value in the low-electric-field segment corresponding to this split point. Minimum electric field value in the high electric field range Between these two values, the actual inflection point electric field is taken as the average of the two values, i.e.: ;

[0134] If there are few data points or the linearity is poor, the corresponding data points can be directly taken. or In this embodiment, the average value is taken.

[0135] Then calculate the slope ratio: ;

[0136] in For a very small positive number (e.g.) This avoids a zero denominator when the low-field slope is zero; for healthy materials, the low-field slope is small but not zero, and the high-field slope is close to the low-field slope, therefore... For fatigue materials, the slope at low field approaches zero, while the slope at high field increases significantly, leading to... Significantly greater than 1.

[0137] Next, the effective piezoelectric coefficient under the preset electric field is extracted; the preset electric field in this embodiment is taken as [value missing]. kilovolts per millimeter; the inflection point electric field at the candidate fatigue point Less than 0.2 At that time, the preset electric field Adjusted to (like Alternatively, the maximum value of the fine scan sequence can be taken; simultaneously, the system will output a prompt in this case: "The preset electric field has been adjusted, and the fatigue characteristics may be in the early stage. It is recommended to combine other indicators for comprehensive judgment." Note that this value may not be at the precise point of the fine electric field sequence and needs to be obtained from the amplitude curve through linear interpolation; since the fine electric field sequence contains two adjacent points, 0.216 and 0.324, and 0.24 is between the two, the linear interpolation formula is used: ;

[0138] in , , and These represent the average amplitude under the corresponding electric field; the effective piezoelectric coefficient is defined as: ;

[0139] The unit of this value is nanometers / kilovolts / millimeter, or picometers / volt, reflecting the piezoelectric strain capability of the material under a preset electric field; the phase delay under the same preset electric field. It can also be obtained through interpolation: the phase delay curve under the same preset electric field. Perform linear interpolation at the point to obtain .

[0140] In step S3, the system has constructed a two-dimensional feature distribution of the health reference region; specifically, the health reference region contains K measurement points, and for each health point, a preset electric field is applied beforehand. Its effective piezoelectric coefficient was calculated at kilovolts per millimeter (via fine scanning or initial calibration in step S4). and phase delay ,in The system calculates the first-order statistical characteristics of these health points:

[0141] Arithmetic mean of effective piezoelectric coefficient ;

[0142] Sample standard deviation of effective piezoelectric coefficient ;

[0143] Arithmetic mean of phase delay ;

[0144] Sample standard deviation of phase delay ;

[0145] Correlation coefficient between effective piezoelectric coefficient and phase delay: ;

[0146] Based on the above statistics, construct the covariance matrix. ;

[0147] For the current candidate fatigue point, its two-dimensional feature vector is: First, perform a standardization transformation to obtain the standardized value: , ;

[0148] Then calculate the Mahalanobis distance. Its definition takes into account the correlation between features: ;

[0149] in Let be the inverse of the covariance matrix; for the two-dimensional case, the inverse matrix can be expressed analytically; let ,but: ;

[0150] Substituting into the calculation, we get: When the calculated correlation coefficient When the value is greater than 0.999, ρ will be forcibly limited to ±0.999 to avoid the Mahalanobis distance value overflowing due to an excessively small denominator. At the same time, if the covariance matrix is ​​close to singular, the system will output a warning indicating that the features of this batch of samples are too strongly correlated, and it is recommended to re-evaluate whether the selection of the preset electric field is reasonable.

[0151] In this embodiment, a preset deviation threshold is set as follows: When the calculated Mahalanobis distance is greater than 3, the candidate fatigue point is determined to be a real fatigue defect region; otherwise, it is determined to be a false defect (such as surface stains, speckle decorrelation, or false markings caused by system noise) and is not included in the fatigue region output.

[0152] In this embodiment, for each measurement point identified as a true fatigue defect region, the system further performs a fatigue grading step to quantify the fatigue severity of that region; this grading is based on the inflection point electric field of the true fatigue defect region. and slope ratio The difference between the reference value and the corresponding reference value in the health reference area.

[0153] First, the system needs to obtain the inflection point electric field reference value of the healthy reference region. Slope Ratio Reference Value In step S3, the system has determined a healthy reference region containing K healthy measurement points. In steps S4 and S5, the system performs a fine electric field amplitude scan on each measurement point within these healthy reference regions and calculates the inflection point electric field and slope ratio for each healthy point. The arithmetic mean of the inflection point electric fields of all measurement points within these healthy reference regions is taken as... Take the arithmetic mean of the slope ratios as... For ideally healthy ferroelectric ceramic materials, the inflection point electric field should be close to zero, and the slope ratio should be close to 1; let the reference value obtained through the above calculations be:

[0154] kilovolts per millimeter ;

[0155] For the current actual fatigue defect region, its inflection point electric field and slope ratio Extraction has been completed in step S5; calculate the increment of the electric field at the inflection point. and the relative value of the slope ratio Then, With material coercive field (In this embodiment) By comparing the values ​​in kilovolts per millimeter (kV / mm), the dimensionless inflection point electric field increment ratio is obtained: ;

[0156] Based on prior knowledge of the fatigue evolution of ferroelectric ceramics, this embodiment sets the following grading thresholds:

[0157] First threshold (upper limit of mild fatigue): ;

[0158] Second threshold (upper limit of slope ratio for mild fatigue): ;

[0159] The third threshold (the upper limit of moderate fatigue; exceeding this value indicates severe fatigue): ;

[0160] Fourth threshold (upper limit of slope ratio for moderate fatigue): .

[0161] Note the first threshold. Second threshold Used to determine the upper boundary of mild fatigue; third threshold and the fourth threshold Used to distinguish between moderate and severe cases.

[0162] The judgment rules are as follows:

[0163] Mild fatigue: when and hour;

[0164] Moderate fatigue: when ,or (Note that if either of these conditions is met, it is considered moderate).

[0165] Severe fatigue: when or hour.

[0166] The following three numerical examples illustrate the grading process.

[0167] Example 1 (Mild Fatigue): Suppose the inflection point electric field of a real fatigue defect region. kV / mm, slope ratio Health Reference Area kilovolts per millimeter ;but:

[0168] kilovolts per millimeter ;

[0169] ;

[0170] because and The fatigue level is classified as mild; this level of fatigue is in the early stage, and monitoring or electric field annealing repair is recommended.

[0171] Example 2 (Moderate Fatigue): Let... kilovolts per millimeter ;but:

[0172] kilovolts per millimeter ;

[0173] ;

[0174] Here Meets the moderate condition, and at the same time The condition is met, therefore it is judged as moderate fatigue; this level indicates that fatigue has obviously developed, and the use conditions should be restricted and replacement should be planned.

[0175] Example 3 (Severe Fatigue): Let... kilovolts per millimeter ;but:

[0176] ;

[0177] ;

[0178] because and Both exceed the severe threshold and are judged as severe fatigue; this level indicates that the material is close to failure and should be stopped and replaced immediately.

[0179] It should be noted that the above thresholds (0.2, 0.4, 1.5, 2.5) are determined based on the accelerated aging experimental statistical data of the lead zirconate titanate piezoelectric ceramic material used in this embodiment. For ferroelectric ceramics with different compositions or different processes, the thresholds can be calibrated in advance using the same method and stored in the equipment parameter library. After the grading determination is completed, the system outputs the grading results (mild, moderate or severe) of each real fatigue defect area as a pseudo-color image superimposed on the optical image of the ceramic surface, which is convenient for operators to interpret intuitively. The grading results also serve as the basic input for subsequent life prediction steps.

[0180] In this embodiment, for each measurement point identified as a real fatigue defect area, after completing the fatigue classification step, the system further performs a life prediction step to estimate the remaining number of cycles for the fatigue area under continued service conditions. This prediction is based on the pre-stored relationship between the electric field increment at the failure inflection point of the same batch of materials and the total number of failure cycles, and is estimated using a linear proportional model.

[0181] First, the system needs to acquire the pre-stored failure parameters of this batch of materials. These parameters are obtained through accelerated aging tests on ferroelectric ceramic samples from the same batch; the specific calibration process is as follows: multiple standard samples from the same batch are taken and tested under the rated operating electric field. (In this embodiment) kilovolts per millimeter, therefore (kilovolts / mm) and rated frequency Cyclic loading is performed at Hertz, and the full-field detection of this patent is performed on the sample every certain number of cycles (e.g., every 100,000 cycles) to track the inflection point electric field change at each measurement point; when the inflection point electric field increment at a certain measurement point reaches the preset failure threshold (set to in this embodiment), the failure is considered complete. At a voltage of kilovolts per millimeter (kV / mm), record the total number of cycles experienced at that measurement point, which is taken as the failure cycle number of the sample; average the failure cycle numbers of multiple samples to obtain the failure cycle number of the entire batch of materials. Simultaneously, the electric field increment at this inflection point is taken as the electric field increment at the failure inflection point. Storage; In this embodiment, it is assumed that the result is obtained through calibration:

[0182] kilovolts per millimeter Ten thousand cycles;

[0183] In actual testing, for the current real fatigue defect region, its inflection point electric field and the inflection point electric field of the healthy reference zone The electric field increment at the current inflection point has already been obtained in step S5 and the fatigue grading step; calculate the electric field increment at the current inflection point:

[0184] ;

[0185] in The arithmetic mean of the electric field at the inflection point of the healthy reference region (in this embodiment) (kilovolts / mm); Note that when Exceed When the point is reached or exceeded, it is considered that the failure state has been reached or exceeded, and the remaining number of cycles is zero.

[0186] The relative fatigue ratio is defined as the ratio of the current inflection point electric field increment to the failure inflection point electric field increment. ;

[0187] When this ratio is between 0 and 1, it represents the linear progression of fatigue from healthy to complete failure.

[0188] When relative fatigue ratio At that time, the system uses a linear extrapolation model to calculate. ;in The function ensures that the remaining number of iterations is non-negative; when hour, .

[0189] When relative fatigue ratio In some cases, the linear model may be unreliable, and the system will automatically switch to a quadratic or exponential model.

[0190] (0.7 < α ≤ 0.85) or (α > 0.85);

[0191] Where c is calibrated to c=2.3 through accelerated aging experiments. For new materials that have not been calibrated, a conservative value in the range of c=2.0 to 2.5 can be taken by default (2.3 is taken in this embodiment). This value range is representative in the statistics of fatigue end-stage evolution rate of common ferroelectric ceramics. At the same time, the output results are marked "nonlinear extrapolation, high uncertainty".

[0192] The following numerical example illustrates the calculation process of the lifetime prediction step; assuming a real fatigue defect region has an inflection point electric field obtained in step S5. kV / mm, Health Reference Zone kV / mm, then: kV / mm, electric field increment at failure inflection point kV / mm, number of failure cycles Ten thousand cycles; relative fatigue ratio:

[0193] ;

[0194] Remaining loop count:

[0195] Ten thousand times;

[0196] The estimated value is approximately 3.7 million cycles; this predicted value is output to the test report and combined with the fatigue grading results (e.g., this point may be judged as mild or moderate fatigue) for user reference.

[0197] If the electric field at the inflection point of another fatigue region kilovolts per millimeter, then kV / mm (subtract 0.02 to get 1.08), , A result of 10,000 reads indicates that the point has become completely invalid.

[0198] It should be noted that the above linear model is based on the assumption that fatigue evolution and the inflection point electric field increment are approximately linearly related. This assumption has good engineering accuracy when the inflection point electric field increment does not exceed 0.8 × the failure threshold. For more accurate predictions, a piecewise exponential model or a nonlinear model based on machine learning can be used, but the linear model used in this embodiment already meets the industrial testing requirements. While outputting the estimated number of remaining cycles, the system also outputs a 95% confidence interval, which is obtained by repeatedly measuring the healthy reference area to obtain the standard deviation of the inflection point electric field. And calculated according to the error propagation formula; specifically, the confidence interval width is approximately In this embodiment, if If the voltage is kilovolts per millimeter, then the confidence interval width is approximately... 10,000 times.

[0199] The output of the above life prediction steps will be presented to the user as part of the inspection report, along with the fatigue classification results.

[0200] Building upon Example 1, this scheme uses the instantaneous phase of the excitation voltage as the absolute reference for synchronous phase-locked demodulation, simultaneously extracting the fundamental amplitude and phase delay from the displacement signal, and using the phase delay as the core detection dimension. This overcomes the fundamental flaw of existing technologies that rely solely on amplitude information and miss early fatigue phase anomalies. By performing low-field frequency scanning at each measurement point and extracting the relaxation peak frequency, full width at half maximum (FWHM), and low-frequency limiting phase, and automatically calibrating the healthy reference area based on the full-field statistical distribution, the influence of batch differences on threshold setting is eliminated. Furthermore, for each candidate fatigue point, the electric field is dynamically adjusted using its own relaxation peak frequency as the test frequency. A fine scanning of the amplitude step size and integration cycle number is performed to adapt the measurement parameters to the actual state of the measured point. Finally, by constructing a two-dimensional feature vector of the effective piezoelectric coefficient and the phase delay under the preset electric field, the Mahalanobis distance between it and the distribution of the healthy reference area is calculated. This method can accurately locate the real fatigue area while eliminating false defects. Based on the ratio of the electric field increment to the slope at the inflection point, the fatigue is classified into mild, moderate and severe levels. The remaining number of cycles is estimated by combining the failure test data. This method transforms the domain dynamics response of ferroelectric ceramics into quantifiable and spatially mappable phase spectrum features, providing an engineeringable technical path for early fatigue location and life prediction.

[0201] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0202] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0203] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for defect identification in electronic ceramic products based on machine vision, characterized in that, include: An alternating electric field is applied to excite the electronic ceramic under test, and the displacement signal and excitation voltage reference signal on the surface of the electronic ceramic under test are collected simultaneously. Synchronous phase-locked demodulation is performed on the displacement signal and the excitation voltage reference signal to extract the fundamental amplitude and fundamental phase delay at each test frequency and each test electric field amplitude. Under a preset low electric field amplitude, frequency scanning is performed according to a preset frequency sequence. The relaxation spectrum feature vector of each measurement point is extracted based on the fundamental phase delay at each frequency. The relaxation spectrum feature vector includes at least the relaxation peak frequency. Based on the statistical distribution of relaxation spectrum feature vectors of all measurement points or preset areas, a healthy reference area is automatically determined, and candidate fatigue points in the measurement points are marked based on the healthy reference area and preset rules. For each candidate fatigue point, the relaxation peak frequency of the candidate fatigue point itself is used as the test frequency to perform electric field amplitude scanning and obtain the amplitude curve of the candidate fatigue point under multiple electric field amplitudes. The effective piezoelectric coefficient under the preset electric field is extracted from the amplitude curve. The effective piezoelectric coefficient and the phase delay under the preset electric field are used to form a two-dimensional feature vector. The deviation of the two-dimensional feature vector from the two-dimensional feature distribution corresponding to the healthy reference area is calculated. When the deviation exceeds the preset deviation threshold, the candidate fatigue point is determined to be a real fatigue defect area.

2. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, Synchronous phase-locked demodulation specifically includes: generating an in-phase reference sequence and a quadrature reference sequence with the same frequency as the excitation voltage reference signal; calculating the integral average of the displacement signal and the in-phase reference sequence over multiple consecutive periods as the in-phase component, and calculating the integral average of the displacement signal and the quadrature reference sequence over multiple consecutive periods as the quadrature component; performing a square root operation on the sum of squares of the in-phase component and the quadrature component to obtain the fundamental amplitude, and performing an arctangent operation on the in-phase component and the quadrature component to obtain the fundamental phase; subtracting the initial phase of the excitation voltage reference signal from the fundamental phase, and performing angle winding correction on the difference to make it fall into the negative. Zhizheng Within the range, the fundamental phase delay is obtained.

3. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, The relaxation spectrum feature vector also includes the half-peak width and the low-frequency limiting phase. The specific steps for extracting the relaxation spectrum feature vector include: taking the fundamental phase delay at the lowest test frequency as the low-frequency limiting phase; interpolating the fundamental phase delay data points obtained by frequency scanning, and locating the frequency corresponding to the maximum phase delay on the interpolation curve as the relaxation peak frequency; calculating the difference between the phase delay at the relaxation peak frequency and the low-frequency limiting phase as the phase peak height; finding frequency points on the left and right sides of the relaxation peak frequency where the phase delay is equal to the low-frequency limiting phase plus half the phase peak height, and calculating the difference between the right frequency point and the left frequency point as the half-peak width.

4. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, The automatic determination of the health reference area specifically includes: calculating the median of the relaxation peak frequencies of all measurement points; selecting measurement points whose relative deviation from the median is less than a first preset threshold to form a candidate reference area; selecting the measurement point with the smallest low-frequency limiting phase within the candidate reference area as an initial reference point; extracting a neighborhood window of a preset size centered on the initial reference point; calculating the arithmetic mean and standard deviation of the relaxation peak frequencies within the window; when the standard deviation is less than a second preset threshold of the arithmetic mean, using the arithmetic mean of the relaxation spectrum feature vectors of all measurement points within the window as the reference benchmark for the health reference area; otherwise, expanding the size of the neighborhood window until the conditions are met or calling the pre-stored factory database.

5. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, Marking candidate fatigue points in measurement points based on a healthy reference region and preset rules specifically includes: calculating a first relative deviation between the relaxation peak frequency of each measurement point and the relaxation peak frequency of the healthy reference region; calculating a second relative deviation between the half-width at half-maximum (WWHM) of each measurement point and the WWHM of the healthy reference region; and calculating a third relative deviation between the low-frequency limiting phase of each measurement point and the low-frequency limiting phase of the healthy reference region. When the first relative deviation is greater than a first deviation threshold, or the second relative deviation is greater than a second deviation threshold, or the third relative deviation is greater than a third deviation threshold, the measurement point is marked as a candidate fatigue point.

6. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, Performing an electric field amplitude scan includes dynamically adjusting the electric field amplitude step size and the number of integration cycles: starting from the initial electric field amplitude, the electric field is increased by a fixed step size, and the increment of the fundamental phase delay under adjacent electric field amplitudes is calculated in real time; when the increment exceeds a preset multiple of the previous increment twice in a row, the current electric field amplitude is determined as the inflection point electric field of the candidate fatigue point; A non-uniform electric field amplitude sequence is generated based on the inflection point electric field, wherein the electric field amplitude interval near the inflection point electric field is smaller than the interval in the region far from the inflection point electric field. For each electric field amplitude, the first number of cycles is initially collected and the standard deviation of the fundamental phase delay of each cycle is calculated. If the standard deviation exceeds the noise threshold, the number of cycles is increased sequentially to the second and third cycles until the standard deviation is lower than the noise threshold or the maximum number of cycles is reached.

7. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, Before extracting the effective piezoelectric coefficient under the preset electric field from the amplitude curve, the method further includes extracting the inflection point electric field and slope ratio from the phase delay curve corresponding to the amplitude curve: dividing the phase delay curve into a low electric field segment and a high electric field segment according to the electric field amplitude from small to large, with the dividing point as a candidate inflection point; performing a first linear fitting on the phase delay data points of the low electric field segment to obtain a first slope, and performing a second linear fitting on the phase delay data points of the high electric field segment to obtain a second slope; determining the electric field amplitude corresponding to the dividing point as the inflection point electric field, and calculating the ratio of the second slope to the first slope as the slope ratio.

8. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, The construction of the two-dimensional feature distribution corresponding to the health reference region specifically includes: for all measurement points within the health reference region, calculating the first arithmetic mean and first standard deviation of the effective piezoelectric coefficient under the preset electric field, calculating the second arithmetic mean and second standard deviation of the fundamental phase delay under the preset electric field, and calculating the correlation coefficient between the effective piezoelectric coefficient and the fundamental phase delay; constructing a covariance matrix based on the first standard deviation, the second standard deviation, and the correlation coefficient; the degree of deviation is measured using Mahalanobis distance, specifically: dividing the difference between the effective piezoelectric coefficient of the candidate fatigue point and the first arithmetic mean by the first standard deviation to obtain a first standardized value, dividing the difference between the fundamental phase delay of the candidate fatigue point and the second arithmetic mean by the second standard deviation to obtain a second standardized value; and calculating the Mahalanobis distance based on the first standardized value, the second standardized value, and the inverse matrix of the covariance matrix.

9. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, It also includes a fatigue grading step: obtaining the difference between the inflection point electric field of the actual fatigue defect region and the inflection point electric field of the healthy reference region, and obtaining the ratio of the slope ratio of the actual fatigue defect region to the slope ratio of the healthy reference region; When the ratio of the difference to the coercive field of the material is less than or equal to a first threshold and the ratio is less than or equal to a second threshold, it is determined to be mild fatigue. When the ratio of the difference to the coercive field of the material is greater than the first threshold and less than or equal to the third threshold, or when the ratio is greater than the second threshold and less than or equal to the fourth threshold, it is determined to be moderate fatigue; when the ratio of the difference to the coercive field of the material is greater than the third threshold or when the ratio is greater than the fourth threshold, it is determined to be severe fatigue.

10. The method for defect identification of electronic ceramic products based on machine vision according to claim 1, characterized in that, It also includes a lifetime prediction step: obtaining the pre-stored inflection point electric field increment of the same batch of materials and the corresponding failure cycle number; calculating the difference between the inflection point electric field of the actual fatigue defect region and the inflection point electric field of the healthy reference region as the current inflection point electric field increment; Divide the current inflection point electric field increment by the failure inflection point electric field increment to obtain the relative fatigue ratio; subtract the relative fatigue ratio from the current inflection point and multiply by the failure cycle number to obtain an estimate of the remaining cycle number.

Images