A kind of pit detection method based on carbon-based composite coating of fuel cell metal bipolar plate

Abstract

The application discloses a kind of based on fuel cell metal bipolar plate carbon-based composite coating pit detection method, specifically relates to surface defect optical detection technical field, for solving the misjudgment problem caused by thermal deformation interference of existing optical detection method, is by collecting the reflection light signal sequence of illumination light source opening and closing attenuation stage, extracts double-band reflectivity ratio sequence and calculates its change rate curve, identifies signal attenuation stage and obtains thermal inertia relaxation time constant by fitting, finally through comprehensive matching degree calculation distinguishes thermal deformation interference and real pit defect to realize the improvement of detection precision.

Description

Technical Field

[0001] This invention relates to the field of optical detection technology for surface defects, and more specifically, to a method for detecting pits based on a carbon-based composite coating on a fuel cell metal bipolar plate. Background Technology

[0002] The integrity of the carbon-based composite coating on the surface of fuel cell metal bipolar plates is crucial for ensuring their long-term reliable operation. In existing technologies, surface visual inspection methods based on optical imaging are widely used for detecting macroscopic defects in such coatings due to their high resolution, high efficiency, and non-contact nature. These methods typically utilize a high-brightness light source to illuminate the surface of the coating under test and identify surface morphology anomalies by collecting the reflected light signals.

[0003] However, in actual high-throughput industrial inspection processes, to achieve high signal-to-noise ratio and high resolution, inspection systems often require illumination sources with high power density. Light energy accumulates in localized micro-regions on the coating surface and is converted into heat. Due to the difference in thermophysical parameters between the carbon-based composite coating and the underlying metal substrate material, this thermal effect induces non-permanent thermal deformation in the micro-regions. The signal characteristics formed by this deformation on the optical sensor are extremely similar to those of real pit defects, leading to misjudgments by the inspection system and severely affecting the accuracy and reliability of the inspection results. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a pit detection method based on carbon-based composite coating of metal bipolar plate of fuel cell to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for detecting pits based on a carbon-based composite coating on a fuel cell metal bipolar plate includes the following steps:

[0007] S1. Illuminate the surface of the coating under test with a light source of preset power density, and collect the reflected light signal sequence during the on-time and attenuation stages after the light source is turned off.

[0008] S2. Extract the light intensity signals of the first band and the second band from the reflected light signal sequence, and calculate the ratio of the light intensity signal of the first band to the light intensity signal of the second band to obtain the sequence of the change of the dual-band reflectivity ratio over time.

[0009] S3. Calculate the rate of change of the dual-band reflectivity ratio sequence over time to obtain the curve of the rate of change of the dual-band reflectivity ratio.

[0010] S4. Identify the signal attenuation stage corresponding to the lighting source after it is turned off in the dual-band reflectivity ratio change rate curve, and fit the corresponding thermal inertia relaxation time constant of the attenuation process.

[0011] S5. The curve of the change rate of the dual-band reflectivity ratio and the thermal inertia relaxation time constant are combined with the pre-stored thermal deformation characteristics to calculate the matching degree and obtain the comprehensive matching degree index.

[0012] S6. When the comprehensive matching index exceeds the preset threshold, it is determined that there is thermal deformation interference in the current detection area and the pit defect judgment is excluded.

[0013] Furthermore, the surface of the coating under test is illuminated by a light source with a preset power density, and the reflected light signal sequence is collected during the on-time and attenuation phases after the light source is turned off, including:

[0014] The illumination source is controlled to irradiate the surface of the coating under test in a pulse modulation mode with a preset duty cycle;

[0015] Within a single pulse cycle of the lighting source, the signal from the rising edge of the pulse to the top of the pulse is collected as the reflected light signal during the lighting source's on-time.

[0016] The signal after the falling edge of the pulse is collected as the reflected light signal during the decay phase after the illumination source is turned off;

[0017] The reflected light signals collected within multiple pulse cycles are combined in chronological order to form a reflected light signal sequence.

[0018] Furthermore, the light intensity signals of the first and second bands are extracted from the reflected light signal sequence, and the ratio of the light intensity signal of the first band to the light intensity signal of the second band is calculated to obtain a sequence of the dual-band reflectivity ratio changing with time, including:

[0019] The reflected light signal sequence is separated into light signals of different wavelengths using a beam splitter.

[0020] Select the first band optical signal with a preset center wavelength and detect its light intensity to obtain the first band optical intensity signal;

[0021] Select the second band optical signal with a preset center wavelength and detect its intensity to obtain the second band optical intensity signal;

[0022] The first band light intensity signal and the second band light intensity signal are divided by corresponding points according to the time sequence to obtain the sequence of the ratio of reflectivity of the two bands changing with time.

[0023] Furthermore, performing corresponding point division according to the time series means: dividing the light intensity signal value of the first band and the light intensity signal value of the second band at the same timestamp to obtain the dual-band reflectivity ratio corresponding to the corresponding time point, and arranging the ratios of all time points in chronological order to form a sequence.

[0024] Furthermore, the rate of change of the dual-band reflectivity ratio sequence over time is calculated to obtain the dual-band reflectivity ratio change rate curve, including:

[0025] A smoothing filter is applied to the time sequence of the dual-band reflectivity ratio to obtain a preprocessed ratio sequence.

[0026] The instantaneous rate of change sequence is obtained by calculating the difference between data at adjacent time points in the preprocessed ratio sequence;

[0027] The instantaneous rate of change sequence is connected in chronological order to form a dual-band reflectivity ratio rate of change curve.

[0028] Furthermore, the signal attenuation stage corresponding to the illumination source being turned off is identified in the dual-band reflectivity ratio change rate curve, and the corresponding thermal inertial relaxation time constant of the attenuation process is obtained by fitting, including:

[0029] Locate the time point corresponding to the moment when the lighting source is turned off in the dual-band reflectivity ratio change rate curve;

[0030] The curve segment in which the signal amplitude changes monotonically from the corresponding time point onward is extracted as the signal attenuation stage;

[0031] The curve data during the signal attenuation phase are fitted with an exponential decay function;

[0032] The characteristic time parameter is extracted from the fitted exponential decay function and used as the thermal inertia relaxation time constant.

[0033] Furthermore, the exponential decay function fitting refers to fitting the curve data of the signal decay stage to an exponential function of the form y=A·exp(-t / τ)+B using the least squares method, where t is the time variable, y is the signal amplitude, A and B are constant terms, and τ is the thermal inertia relaxation time constant to be determined.

[0034] Furthermore, the dual-band reflectivity ratio change rate curve and the thermal inertia relaxation time constant are combined with pre-stored thermal deformation characteristics to calculate the matching degree, resulting in a comprehensive matching degree index, including:

[0035] Calculate the waveform similarity between the dual-band reflectivity ratio change rate curve and the pre-stored thermal deformation characteristic curve;

[0036] Calculate the degree of agreement between the thermal inertia relaxation time constant and the pre-stored range of thermal deformation time constant;

[0037] The overall matching index is obtained by weighted fusion calculation of waveform similarity and conformity.

[0038] Furthermore, weighted fusion calculation refers to: after normalizing the waveform similarity and conformity, multiplying the waveform similarity by the first weight coefficient, multiplying the conformity by the second weight coefficient, and then adding the two weighted results to obtain the comprehensive matching index.

[0039] Furthermore, when the overall matching index exceeds a preset threshold, it is determined that thermal deformation interference exists in the current detection area and the pit defect judgment is excluded, including:

[0040] Compare the overall matching index with a preset threshold;

[0041] When the overall matching index exceeds the preset threshold, the current detection area is marked as a thermal deformation interference area;

[0042] The test results marked as areas of thermal deformation interference were excluded from the determination of pit defects.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. By analyzing the dynamic changes in the dual-band reflectivity ratio of the coating surface during illumination and light attenuation, the transient optical effects caused by thermal deformation and the permanent morphological features caused by real pit defects were effectively distinguished. The dual-band reflectivity ratio method can eliminate common-mode interference caused by factors such as light source fluctuations and surface tilt, while enhancing temperature-sensitive optical response characteristics, thus significantly improving the sensitivity to thermal deformation signals. By extracting the intrinsic physical parameter of thermal inertia relaxation time constant, the differences in thermophysical properties between the coating and the substrate material can be accurately characterized, providing a reliable quantitative basis for distinguishing between thermal deformation and real defects.

[0045] 2. By combining dynamic optical response characteristics with thermophysical parameters, accurate identification of thermal deformation interference is achieved through multi-feature fusion matching degree calculation. This not only avoids the limitations of single optical feature criteria but also enables quantitative assessment of thermal deformation interference by establishing a comprehensive matching degree index. This physics-based discrimination method effectively overcomes the dependence of traditional visual inspection methods on empirical thresholds, significantly improving the accuracy and reliability of pit defect detection and providing an effective technical means for quality control of fuel cell metal bipolar plates. Attached Figure Description

[0046] Figure 1 This is a flowchart of a pit detection method based on a carbon-based composite coating for a fuel cell metal bipolar plate, according to the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0048] Example: Figure 1 This invention discloses a pit detection method based on a carbon-based composite coating for a fuel cell metal bipolar plate, comprising the following steps:

[0049] S1. Illuminate the surface of the coating under test with a light source of preset power density, and collect the reflected light signal sequence during the on-time and attenuation stages after the light source is turned off.

[0050] S2. Extract the light intensity signals of the first band and the second band from the reflected light signal sequence, and calculate the ratio of the light intensity signal of the first band to the light intensity signal of the second band to obtain the sequence of the change of the dual-band reflectivity ratio over time.

[0051] S3. Calculate the rate of change of the dual-band reflectivity ratio sequence over time to obtain the curve of the rate of change of the dual-band reflectivity ratio.

[0052] S4. Identify the signal attenuation stage corresponding to the lighting source after it is turned off in the dual-band reflectivity ratio change rate curve, and fit the corresponding thermal inertia relaxation time constant of the attenuation process.

[0053] S5. The curve of the change rate of the dual-band reflectivity ratio and the thermal inertia relaxation time constant are combined with the pre-stored thermal deformation characteristics to calculate the matching degree and obtain the comprehensive matching degree index.

[0054] S6. When the comprehensive matching index exceeds the preset threshold, it is determined that there is thermal deformation interference in the current detection area and the pit defect judgment is excluded.

[0055] In the implementation of the pit detection method, step S1 achieves the acquisition of the reflected light signal sequence in the following specific manner: First, a high-brightness LED lighting source is controlled by a function generator to operate in pulse modulation mode. The pulse duty cycle is preset within the range of 10% to 50% based on the thermal response characteristics of the coating material, for example, 30%. This range ensures a detectable balance between heat accumulation and heat dissipation. The pulse frequency is set between 10Hz and 100Hz, for example, 20Hz. This frequency range ensures sufficient time to record the complete signal attenuation process. The power density of the lighting source is controlled between 0.5W / cm² and 5W / cm² by an adjustable DC power supply, for example, 2W / cm². This power density range must ensure that sufficient thermal response signals are excited without causing permanent thermal damage to the coating.

[0056] Within a single pulse cycle, a high-speed photodetector is used to acquire the reflected light signal, with a sampling rate of no less than 100 kS / s to ensure complete capture of rapid changes in the light signal. Specifically, the signal acquired within the time interval from 0.1 ms after the pulse rising edge to 0.1 ms before the pulse falling edge is defined as the reflected light signal during the illumination source's on-time, and this phase mainly reflects the steady-state reflection characteristics of the coating surface. The signal acquired within the time interval from 0 ms to 50 ms after the pulse falling edge is defined as the reflected light signal during the attenuation phase after the illumination source is turned off. This phase records the attenuation process of reflectivity on the coating surface due to thermal inertia, and the 50 ms acquisition duration can cover most of the thermal relaxation process of the carbon-based coating.

[0057] To improve the signal-to-noise ratio (SNR), the reflected light signals acquired over multiple consecutive pulse cycles are aligned and averaged sequentially. The number of pulse cycles is typically chosen as a power of 2, such as 64 cycles. The alignment method uses the rising edge of each pulse as the time reference point, shifting and aligning the signals acquired over all cycles along the time axis. Then, the signal amplitude at each time point is arithmetically averaged, ultimately forming a complete, high SNR sequence of reflected light signals. This sequence includes the steady-state signal during the illumination source's on-state and the attenuated signal after it is turned off. The total duration is typically the sum of the durations of multiple pulse cycles; for example, when using a 20Hz pulse, the total duration for 64 cycles is 3.2 seconds.

[0058] The power density of the illumination source was determined through prior experiments. Standard samples were irradiated at different power densities, and the temperature rise of the coating surface was monitored using a thermal imager. A power density value was selected that kept the temperature rise within the range of 1K to 5K. The principle for setting the pulse duty cycle was to ensure that the coating surface reached thermal equilibrium during the illumination period, while allowing a sufficiently long decay acquisition time, typically ensuring that the on-time does not exceed 50% of the cycle length. The sampling rate of the photodetector was selected based on the Nyquist sampling theorem. Considering that the main frequency components of the thermal relaxation process are usually below 1kHz, a sampling rate of 100kS / s was chosen to meet the sampling requirements.

[0059] During signal alignment, the rising edge of each pulse cycle is first detected, and a sliding window cross-correlation algorithm is used to determine the optimal alignment position. The alignment accuracy is required to be within one-tenth of the sampling time interval. During superposition and averaging, the arithmetic mean of the signal value at each time point is calculated, and the standard deviation is also calculated to evaluate signal stability. When the standard deviation of the signal at a certain time point exceeds three times the mean, that time point is marked as an outlier and interpolation is performed.

[0060] The reflected light signal sequence is stored using a two-dimensional array data structure. The first dimension is the time axis, and the second dimension is the signal amplitude. Metadata such as sampling rate and pulse parameters are also recorded. The time axis coordinates are set to zero at the rising edge of the first pulse, with the unit being milliseconds. The signal amplitude, after AD conversion, is stored as a 16-bit signed integer and subsequently converted to floating-point numbers for calculations.

[0061] The entire acquisition process is conducted in a darkroom environment to avoid interference from ambient light. The angle between the illumination source and the detector is set to 45 degrees, conforming to standard optical detection configurations. The distance between the detector probe and the coating surface is controlled within the range of 10cm to 50cm, for example, set to 30cm. This distance range ensures that the illumination spot covers a sufficient detection area while maintaining sufficient signal strength. Acquisition of each detection area is repeated three times, and the average value of the reflected light signal sequence is taken as the final output to improve measurement reliability.

[0062] In step S2, the reflected light signal sequence is first processed by a beam splitter, which uses either prism dispersion or grating diffraction to separate the light signals. Specifically, a blazed grating is used as the dispersive element, and its line density is selected based on the required wavelength resolution, for example, a grating with 600 lines per millimeter. This disperses the incident reflected light signal into a continuous spectrum. The spectral range covers 400 nm to 1800 nm to accommodate the characteristic responses of the carbon-based coating in different wavelength bands. The dispersed beam is then focused onto a linear array detector by a focusing lens. The number of detector pixels is configured according to the spectral resolution requirements, for example, a 256-pixel indium gallium arsenide array detector.

[0063] The preset center wavelength of the first band is based on the absorption characteristics of the carbon-based coating in the near-infrared band. It is typically selected from the band where the coating material exhibits a significant thermally induced absorption change, for example, 1310 nm as the first center wavelength, with a bandwidth set to ±20 nm. The preset center wavelength of the second band is selected from a reference band with a relatively stable thermal response, for example, 1550 nm as the second center wavelength, with a bandwidth also set to ±20 nm. The selection principle for these two bands is to make the first band sensitive to temperature changes, while the second band is relatively insensitive to temperature changes, thereby eliminating common-mode interference through a ratio method. The specific values ​​of the band center wavelengths are determined by pre-measuring the spectral response characteristics of the coating material. During measurement, a Fourier transform infrared spectrometer is used to collect the spectral curves of the coating at different temperatures. The band where the absorption coefficient changes most significantly with temperature is selected as the first band, and the band with a stable absorption coefficient is selected as the second band.

[0064] The detection of light intensity signals is achieved through photoelectric conversion and signal conditioning circuitry. The output signal of each pixel corresponding to each band is converted into a voltage signal by a transimpedance amplifier, with the amplification factor adjusted according to the light intensity range. Subsequently, bandpass filtering is performed, with the filter's -3 dB bandwidth matched to a preset band width, for example, a bandwidth of 40 nanometers, and the center frequency corresponding to its respective center wavelength. The filtered signal is sampled by a 24-bit analog-to-digital converter, with the sampling rate consistent with the sampling rate of the reflected light signal sequence, for example, set to 100 kiloHz.

[0065] When performing a point-to-point division operation on the first and second band light intensity signals according to their time sequences, a precise time synchronization mechanism is first established. Using the sampling timestamps of the reflected light signal sequence as a reference, time alignment processing is performed on the two band light intensity signals to ensure strict synchronization of the two band signals at each sampling time point. Time alignment is achieved using an interpolation algorithm. When there is a slight deviation in the sampling times of the two channels, cubic spline interpolation is used to unify the signals onto the same time coordinate axis, with an interpolation accuracy requirement of within one-tenth of the sampling time interval.

[0066] The specific implementation process of the corresponding point division operation is as follows: At each time point, the value of the first band light intensity signal is divided by the value of the second band light intensity signal to obtain the dual-band reflectivity ratio at that time point. To prevent division by zero errors, when the value of the second band light intensity signal is lower than the noise threshold, the ratio at that time point is marked as invalid. The noise threshold is determined by acquiring dark field signals; for example, three times the standard deviation of the dark field signal is taken as the threshold. The dark field signal is the background noise signal acquired when the illumination source is off. The calculated dual-band reflectivity ratios are stored in chronological order as an array structure, with the time interval consistent with the sampling interval of the original reflected light signal sequence. For example, when the sampling rate is 100 kiloseconds per second, the time interval is 10 microseconds.

[0067] To improve the stability of the ratio signal, the original light intensity signal is preprocessed. First, a moving average filter is applied to the light intensity signal of each band, with a window length of 11 sampling points, equivalent to a time window of 0.11 milliseconds. Then, baseline correction is performed to subtract the background noise of each signal before the illumination source is turned on. Outliers are processed using median filtering with a window length of 5 sampling points. The final dual-band reflectivity ratio sequence is then logarithmically transformed to enhance the signal's resolution in the low-amplitude region; the logarithmic transformation is implemented using the natural logarithm function.

[0068] Throughout the processing, the parameters were set based on the following: The choice of bandwidth balances spectral resolution and signal strength, typically ranging from 20 to 50 nanometers. Too narrow a bandwidth reduces signal strength, while too wide a bandwidth reduces spectral resolution. Time alignment accuracy is required to be within one-tenth of the sampling time interval to ensure the accuracy of the ratio calculation. The interpolation algorithm selection considers both computational complexity and accuracy requirements; cubic spline interpolation ensures no distortion of the signal waveform. All signal processing algorithms are implemented in an embedded system, using floating-point arithmetic to ensure computational accuracy and outputting a real-time sequence of the dual-band reflectivity ratio over time. Intermediate data generated during signal processing is timestamped to ensure temporal correlation with the original reflected light signal sequence. The final output dual-band reflectivity ratio sequence is stored as a two-dimensional array, with the first dimension representing the time coordinate and the second dimension representing the ratio value, while also recording metadata such as band parameters and processing parameters.

[0069] In step S3, the sequence showing the time-varying dual-band reflectivity ratio is first smoothed using a smoothing filter. The smoothing filter employs a Savitzky-Golay filter, which smooths the signal by performing polynomial least-squares fitting within a sliding window, effectively suppressing noise while preserving signal characteristics to the greatest extent possible. The filter window length is determined based on the signal sampling rate and characteristic time scale. For example, a window containing 21 sampling points corresponds to a time span of 0.21 milliseconds. The principle for selecting the window length is to make it longer than the noise fluctuation period but shorter than the signal characteristic change time scale. The polynomial order is chosen as 3rd order. This order is based on the analysis of the thermal response signal characteristics of the carbon-based coating; a 3rd order polynomial is sufficient to describe the local curvature characteristics of the signal while avoiding overfitting. Before filtering, the sequence is first extended at the boundaries using a mirror-symmetric method. This involves mirroring subsequent data forward at the beginning of the sequence and mirroring the beginning data backward at the end. The extension length is half the window length to avoid data distortion caused by boundary effects. After smoothing and filtering, a preprocessed ratio sequence is obtained, which retains the trend characteristics of the original signal while suppressing high-frequency noise.

[0070] The instantaneous rate of change sequence is obtained by calculating the difference between adjacent time points in the preprocessed ratio sequence. The difference calculation uses the central difference method. For the rate of change at the i-th time point in the sequence, the value at the (i+1)-th point is subtracted from the value at the (i-1)-th point, and then divided by twice the time interval. The time interval is taken as the sampling interval of the reflected light signal sequence; for example, when the sampling rate is 100 kiloseconds per second, the time interval is 10 microseconds. This calculation method has higher accuracy than forward or backward differencing, and its truncation error is proportional to the square of the time interval. For the start and end points of the sequence, since there are no preceding or succeeding data points, forward and backward differencing are used as supplementary methods. That is, the start point is calculated by subtracting the first point from the second point and dividing by the time interval, and the end point is calculated by subtracting the second-to-last point from the last point and dividing by the time interval, ensuring the continuity of the entire time series. The calculated instantaneous rate of change sequence is stored in array form, with each rate of change value corresponding to a time point aligned with the time point of the original sequence. The unit of the rate of change is changes per second.

[0071] The instantaneous rate of change sequences are concatenated chronologically to form a dual-band reflectivity ratio rate of change curve. During the concatenation process, the continuity and integrity of the time axis are maintained, ensuring that each time point has a corresponding rate of change value. For the two endpoints lost due to differential calculation, the corresponding rate of change values ​​are supplemented through linear extrapolation; that is, the starting endpoint uses the linear extrapolation of the first two rate of change values, and the ending endpoint uses the linear extrapolation of the last two rate of change values, ensuring that the curve's time range is consistent with the original ratio sequence. The final dual-band reflectivity ratio rate of change curve is stored in a two-dimensional time-rate of change data format, with the time axis unit maintaining microsecond-level precision and the rate of change values ​​retaining sufficient decimal places to ensure calculation accuracy. The curve data also includes quality flags to mark time segments where calculation anomalies may exist. These anomalies are usually caused by instantaneous signal drops or noise interference. When the rate of change value exceeds three standard deviations of the normal range, the corresponding time point is marked as a suspicious point.

[0072] Throughout the processing, the selection of key parameters was based on the following criteria: The window length of the Savitzky-Golay filter was chosen based on the estimation of the signal's characteristic time constant. The characteristic time scale was obtained by calculating the signal's autocorrelation function, typically taking 3 to 5 times the number of sampling points corresponding to the characteristic time constant. The polynomial order was determined according to the signal complexity. A local polynomial fitting test was performed on the signal to select the order with the smallest residual and without introducing spurious extrema. The time interval for difference calculations strictly followed the sampling theorem requirements to ensure no aliasing errors were introduced. The determination of the time interval needed to consider the relationship between the highest frequency component of the signal and the sampling rate. All calculations used floating-point operations, achieving the precision required for single-precision floating-point numbers. Accumulated errors were carefully avoided during the calculation process. Data validity checks were also implemented during processing. When an abnormally large rate of change was detected, an automatic recalculation mechanism was triggered, using adjacent valid data for cubic spline interpolation replacement. The final output dual-band reflectivity ratio change rate curve maintained strict consistency with the time base of subsequent processing steps, providing accurate input for the extraction of the thermal inertia relaxation time constant. Timestamp alignment was used when storing the curve data to ensure accurate time correspondence with the original light intensity signal.

[0073] In step S4, the time point corresponding to the moment the lighting source turns off is first located in the dual-band reflectivity ratio change rate curve. This time point is determined by the falling edge triggering a timestamp on the pulse modulation signal. Specifically, the driving signal of the lighting source is time-correlated with the acquired light signal, using the instant the driving signal transitions from high to low as the time reference point. Considering the delay in signal transmission and acquisition links, a calibration experiment is conducted to determine the time offset between the actual light source turning off and the falling edge of the electrical signal. This calibration experiment uses a high-speed photodetector to directly measure the actual light output of the light source and compares it with the driving signal. The resulting delay compensation is typically on the order of microseconds, for example, a specific value between 5 and 20 microseconds. The final determined light source turning-off time point is recorded as an absolute time value and precisely aligned with the time axis of the dual-band reflectivity ratio change rate curve, with an alignment accuracy required to be within one-tenth of the sampling time interval.

[0074] Starting from a defined time point, the signal amplitude is extracted as a monotonically decreasing curve segment as the signal attenuation phase. The extraction process employs a sliding window detection algorithm. The window length is set according to the signal's sampling rate; for example, a window containing 11 sampling points is selected, with an odd number of sampling points within the window to ensure symmetry. The first derivative of the signal is calculated within each window. When the first derivative is negative for three consecutive windows and the signal amplitude decreases beyond a noise threshold, the segment is determined to be a monotonically decreasing phase. The noise threshold is determined by analyzing signal fluctuations during the illumination source's operation, using twice the steady-state signal standard deviation as the threshold. For example, when the steady-state signal standard deviation is 0.005, the noise threshold is set to 0.01. The extracted attenuation phase curve must meet a minimum length requirement, typically containing at least 50 sampling points, to ensure the reliability of subsequent fitting. Non-monotonic fluctuations or signal plateaus are eliminated by setting a rate-of-change threshold, ensuring that the extracted attenuation phase curve exhibits a clear exponential decay characteristic.

[0075] An exponential decay function was fitted to the signal decay curve data during the decay phase. The least squares method was used to fit the data to an exponential function of the form y = A·exp(-t / τ) + B, where t is the time variable, y is the signal amplitude, A and B are constants, and τ is the thermal inertia relaxation time constant to be determined. Before fitting, the data was preprocessed, including subtracting the baseline offset and smoothing / denoising. The initial value of the baseline offset B was the average of the last 10 sampling points in the decay phase. Parameter initialization was achieved through linearization. The natural logarithm of the decay curve was taken, and then linear fitting was performed to obtain the initial estimate of τ. Iterative optimization used the Levenberg-Marquardt algorithm, with the convergence condition set as the change in the sum of squared residuals between two adjacent iterations being less than 0.000001, and the maximum number of iterations set to 200. During the fitting process, outliers were automatically removed. When a data point deviated from the fitted curve by more than three times the standard deviation, the point was marked as an outlier and not included in the next iteration.

[0076] Feature time parameters are extracted from the fitted exponential decay function as the thermal inertia relaxation time constant. Specifically, the parameter τ is extracted, representing the time required for the signal to decay to 1 / e of its initial value, where e is the base of the natural logarithm, approximately 2.71828. Simultaneously, the goodness-of-fit index R-squared is calculated. A goodness-of-fit index greater than 0.95 is considered reliable. The R-squared value is calculated as 1 minus the ratio of the residual sum of squares to the total sum of squares. For cases of poor fit quality, the start and end points of the decay phase are automatically adjusted, and the fitting is re-applied with an adjustment range of ±10% of the original length, up to a maximum of three refit attempts. The final thermal inertia relaxation time constant is stored as a floating-point number, with units consistent with the original time axis, such as microseconds or milliseconds. All intermediate parameters and fit quality indices during the fitting process are recorded in the result data structure, including fitting error, confidence intervals, and other auxiliary information, providing data support for subsequent matching degree calculations. The entire process realizes a complete workflow from signal decay feature extraction to time constant calculation, ensuring the accuracy and reliability of thermal inertia parameter extraction.

[0077] In step S5, the waveform similarity between the dual-band reflectivity ratio change rate curve and the pre-stored thermal deformation characteristic curve is first calculated. The waveform similarity calculation employs a dynamic time warping algorithm, which effectively handles the nonlinear deformation of the two curves along the time axis. Specifically, the amplitudes of the two curves are first normalized to a uniform range between 0 and 1. The normalization formula is to subtract the minimum value from each data point and divide by the difference between the maximum and minimum values. Then, an Euclidean distance matrix is ​​constructed between corresponding points on the two curves. The distance is calculated as the square root of the sum of the squares of the amplitude differences at corresponding time points. The optimal warping path is found using dynamic programming. The path search constraints include monotonicity constraints, continuity constraints, and boundary constraints. The final minimum cumulative distance serves as the basis for the similarity metric. The similarity value is converted to a similarity index between 0 and 1 by subtracting the normalized distance value using Formula 1, where the normalized distance value is the cumulative distance divided by the sum of the lengths of the two curves. The pre-stored thermal deformation characteristic curves are typical curves obtained by statistical analysis of a large number of known thermal deformation samples. Specifically, they are obtained by time alignment and amplitude averaging of more than 100 verified thermal deformation sample curves and stored in the form of time-amplitude data pairs with a time resolution of 1 microsecond and an amplitude accuracy of 0.001.

[0078] The conformity between the thermal inertia relaxation time constant and the pre-stored range of thermal deformation time constants is calculated. The pre-stored range of thermal deformation time constants is determined through statistical analysis. Specifically, relaxation time constant data from over 200 validated thermal deformation samples are collected, and the mean and standard deviation of these data are calculated. The mean plus or minus three times the standard deviation is used as the upper and lower bounds of the range. For example, when the mean is 50 microseconds and the standard deviation is 5 microseconds, the time constant range is set to 35 microseconds to 65 microseconds. The conformity calculation uses a piecewise linear function. When the measured time constant falls within the pre-stored range, the conformity is 1. When it deviates from the range, the conformity decreases linearly, dropping to 0 at a certain distance from the upper and lower bounds of the range. The deviation distance is set based on the statistical standard deviation of the time constant, using twice the standard deviation as the length of the linearly decreasing interval. For example, when the standard deviation is 5 microseconds, the linearly decreasing interval length is 10 microseconds. The conformity calculation results are also normalized to between 0 and 1 using the minimum-maximum scaling method.

[0079] The overall matching score is calculated by weighted fusion of waveform similarity and conformity. Before weighted fusion, both scores are standardized to ensure consistent numerical ranges and similar distribution characteristics. The determination of the first and second weighting coefficients is based on the discriminative ability of each score in thermal deformation recognition. The discriminative power of each score is determined through receiver operating characteristic (ROC) curve analysis, and then the weighting coefficients are calculated. Specifically, 100 positive samples (thermal deformation) and 100 negative samples (non-thermal deformation) are collected, and the distribution of each score on these samples is calculated. The relative importance of each score is determined by calculating its Gini coefficient or information gain. For example, when the Gini coefficient for waveform similarity is 0.8 and the Gini coefficient for conformity is 0.6, the first weighting coefficient is 0.57, the second weighting coefficient is 0.43, and the sum of the two coefficients is 1. The weighted calculation uses a linear weighted summation formula, that is, the overall matching score equals waveform similarity multiplied by the first weighting coefficient plus conformity multiplied by the second weighting coefficient. The final comprehensive matching score ranges from 0 to 1, with values ​​closer to 1 indicating a higher degree of matching with the thermal deformation features. All intermediate results and parameters generated during the calculation process are recorded, including original similarity values, conformity values, and the basis for weighting coefficients, for subsequent verification and adjustment. The entire matching score calculation process realizes a complete workflow from feature extraction to comprehensive evaluation, providing a quantitative basis for the accurate identification of thermal deformation interference. After the comprehensive matching score is calculated, a confidence level assessment is performed. The reliability of the results is evaluated by calculating the variance and confidence interval of the score. When the confidence level falls below a preset threshold, a recalculation mechanism is automatically triggered.

[0080] In step S6, the overall matching index is first compared with a preset threshold. The preset threshold is determined based on statistical analysis of a large amount of experimental data, specifically by analyzing the distribution of the overall matching index of known heat-deformed and non-heat-deformed samples. More than 200 validated sample data points are collected, including 100 heat-deformed positive samples and 100 normal samples, and the overall matching index values ​​for these samples are calculated. By plotting the receiver operating characteristic (ROC) curve, the true positive rate and false positive rate at different thresholds are calculated. The threshold that maximizes the Youden index is selected as the preset threshold, where both the false positive rate and true positive rate are within acceptable ranges; for example, the false positive rate is controlled below 5%, and the true positive rate is maintained above 95%. The comparison operation uses a numerical comparison algorithm. When the overall matching index is greater than or equal to the preset threshold, it is considered to have exceeded the threshold. For example, when the preset threshold is 0.85, an overall matching index of 0.85 or higher is considered to have exceeded the threshold. The comparison process also includes a hysteresis interval to avoid frequent switching near the threshold. For example, the original state is kept unchanged within a range of ±0.02 of the threshold, and the judgment state is changed only when the indicator changes beyond the hysteresis interval.

[0081] When the overall matching degree index exceeds a preset threshold, the current detection area is marked as a thermal deformation interference area. The marking operation is implemented by setting specific status flag bits in the detection area data structure. These flag bits use 8-bit binary encoding, where bits 3 to 5 represent the thermal deformation interference status, with different bit combinations representing different levels of interference. Simultaneously, detailed information such as the marking timestamp, location coordinates, and matching degree index value are recorded. This information is stored in the detection result database in the form of a structure containing a timestamp field (64-bit integer), coordinate fields (two 32-bit floating-point numbers), and a matching degree field (32-bit floating-point number). Visual prompts are also generated during the marking process, highlighting the marked area in a specific color in the detection system's user interface. For example, a red border is used to mark the thermal deformation interference area, with a border width of 3 pixels and a color value of RGB(255,0,0). All marking operations are logged, including marking time, operator, and judgment criteria, for subsequent traceability and verification. The log records are stored in JSON format and include fields such as operation time, operation type, area coordinates, and matching degree value.

[0082] The system excludes areas marked as heat deformation interference from the assessment of pit defects. Exclusion is achieved by modifying the status flag of the inspection result, changing the status of an area that might have been judged as a pit defect to an excluded status. Two bits are specifically allocated in the status code to indicate the exclusion status: 00 indicates not excluded, 01 indicates heat deformation exclusion, and 10 indicates exclusion for other reasons. The exclusion decision employs a three-layer verification mechanism: first, it checks whether the area has already been marked as a heat deformation interference area; second, it verifies whether the confidence level of the marking meets the requirements, with a confidence threshold set to 0.9; and finally, it confirms the rationality of the exclusion operation through a manual review interface. A detailed exclusion report is generated during the exclusion process, recording the reason for exclusion, the exclusion time, and supporting evidence. The report format uses a standardized XML structure and includes information such as the exclusion time, area identifier, matching degree value, threshold parameter, and operator. For excluded areas, the system automatically removes them from the defect statistics results, but does not completely delete the inspection data. Instead, it archives them in a special category for subsequent analysis and reference. The archive database uses time-partitioned storage, with each partition storing one month's worth of exclusion data. The entire exclusion process ensures that it does not affect the defect determination of other normal areas, while maintaining full data traceability. All operations follow pre-set business rules and workflows to ensure the consistency and reliability of the determination results. After the exclusion operation is completed, the system automatically generates an exclusion confirmation notification, which is sent to relevant quality management personnel via a message queue. The notification includes information about the excluded area, the reason for exclusion, and a confirmation link. Management personnel can quickly access the detailed exclusion report for review and confirmation via the link.

[0083] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0084] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0085] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0086] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0087] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0089] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting pits based on carbon-based composite coating of a fuel cell metal bipolar plate, characterized by, The method comprises the following steps: S1, irradiating the surface of the coating to be measured by using a light source with a preset power density, and collecting a sequence of reflected light signals during the opening of the light source and the decay stage after the light source is turned off; S2, extracting light intensity signals of a first wave band and a second wave band from the sequence of reflected light signals, and calculating the ratio of the light intensity signals of the first wave band and the second wave band to obtain a sequence of the ratio of the dual-wave band reflectivity changing with time; S3, calculating the rate of change of the sequence of the ratio of the dual-wave band reflectivity with time to obtain a curve of the rate of change of the ratio of the dual-wave band reflectivity; S4, identifying the signal decay stage corresponding to the stage after the light source is turned off in the curve of the rate of change of the ratio of the dual-wave band reflectivity, and fitting to obtain a thermal inertia relaxation time constant corresponding to the decay process; S5, calculating the matching degree of the curve of the rate of change of the ratio of the dual-wave band reflectivity and the thermal inertia relaxation time constant with the pre-stored thermal deformation characteristics to obtain a comprehensive matching degree index; S6, when the comprehensive matching degree index exceeds a preset threshold, determining that there is thermal deformation interference in the current detection area and excluding the pit defect determination.

2. A method for detecting pits in a carbon-based composite coating on a fuel cell metal bipolar plate according to claim 1, characterized in that, The method comprises the following steps: controlling the light source to irradiate the surface of the coating to be measured in a pulse modulation mode with a preset duty cycle; in a single pulse period of the light source, collecting signals from the rising edge to the flat-top stage of the pulse as the reflected light signals during the opening of the light source; collecting signals after the falling edge of the pulse as the reflected light signals during the decay stage after the light source is turned off; combining the reflected light signals collected in multiple pulse periods in time sequence to form a sequence of reflected light signals.

3. A method for detecting pits in carbon-based composite coating of a fuel cell metal bipolar plate according to claim 1, characterized in that, The method comprises the following steps: separating the sequence of reflected light signals into light signals of different wavelengths through a light splitting device; selecting a first wave band of a preset central wavelength and detecting the light intensity to obtain a first wave band light intensity signal; selecting a second wave band of a preset central wavelength and detecting the light intensity to obtain a second wave band light intensity signal; performing corresponding point division operation on the first wave band light intensity signal and the second wave band light intensity signal in time sequence to obtain a sequence of the ratio of the dual-wave band reflectivity changing with time.

4. A method of detecting pits in a carbon-based composite coating of a fuel cell metal bipolar plate according to claim 3, characterized in that, The corresponding point division operation in time sequence refers to performing division operation on the first wave band light intensity signal value and the second wave band light intensity signal value at the same time stamp to obtain the ratio of the dual-wave band reflectivity corresponding to the time point, and arranging the ratios of all time points in time sequence to form a sequence.

5. The method of claim 1, wherein the method is a method of detecting pits in a carbon-based composite coating of a fuel cell metal bipolar plate. The method comprises the following steps: performing smoothing filter processing on the sequence of the ratio of the dual-wave band reflectivity changing with time to obtain a preprocessed ratio sequence; calculating the difference values of adjacent time point data in the preprocessed ratio sequence to obtain an instantaneous change rate sequence; connecting the instantaneous change rate sequence in time sequence to form a curve of the rate of change of the ratio of the dual-wave band reflectivity.

6. A method for detecting pits in carbon-based composite coating on a fuel cell metal bipolar plate according to claim 1, characterized in that, Identify the signal decay stage corresponding to the light source being turned off in the dual-band reflectance ratio change rate curve, and fit the thermal inertia relaxation time constant corresponding to the decay process, including: Locating the time point corresponding to the time when the light source is turned off in the dual-band reflectance ratio change rate curve; Extracting the curve segment with monotonically decreasing signal amplitude from the corresponding time point as the signal decay stage; Exponential decay function fitting of the curve data in the signal decay stage; Extract the feature time parameter from the fitted exponential decay function as the thermal inertia relaxation time constant.

7. A method of detecting pits in a carbon-based composite coating of a fuel cell metal bipolar plate according to claim 6, characterized in that, Exponential decay function fitting refers to: using the least squares method to fit the curve data in the signal decay stage into an exponential function of the form y=A·exp(-t / τ)+B, where t is the time variable, y is the signal amplitude, A and B are constant terms, and τ is the thermal inertia relaxation time constant to be solved.

8. A method for detecting pits in carbon-based composite coating of a fuel cell metal bipolar plate according to claim 1, characterized in that, Match the dual-band reflectance ratio change rate curve and the thermal inertia relaxation time constant with the pre-stored thermal deformation features to calculate the comprehensive matching degree index, including: Calculate the waveform similarity between the dual-band reflectance ratio change rate curve and the pre-stored thermal deformation feature curve; Calculate the compliance degree between the thermal inertia relaxation time constant and the pre-stored thermal deformation time constant range; Weighted fusion calculation of waveform similarity and compliance degree to obtain comprehensive matching degree index.

9. A method of detecting pits in a carbon-based composite coating of a fuel cell metal bipolar plate according to claim 8, characterized in that, Weighted fusion calculation refers to: after normalizing the waveform similarity and the compliance degree, multiplying the waveform similarity by the first weight coefficient, multiplying the compliance degree by the second weight coefficient, and then adding the two weighted results to obtain the comprehensive matching degree index.

10. The method of claim 1, wherein the method is a pit detection method for a carbon-based composite coating of a fuel cell metal bipolar plate. When the comprehensive matching degree index exceeds the preset threshold, it is determined that there is thermal deformation interference in the current detection area and the pit defect judgment is excluded, including: Compare the comprehensive matching degree index with the preset threshold; When the comprehensive matching degree index exceeds the preset threshold, mark the current detection area as a thermal deformation interference area; Exclude pit defect judgment for detection results marked as thermal deformation interference area.

Images