A perovskite thin film quality management and control system integrating optical detection and data analysis

Abstract

The application belongs to the technical field of semiconductor material detection, and relates to a perovskite film quality management and control system integrating optical detection and data analysis, which comprises a data acquisition and preprocessing module, a dynamic reference construction module, a defect index calculation module and a defect analysis and regulation module. The perovskite film surface photoluminescence intensity data and reflectivity data are collected, spatial pixel registration and preprocessing are performed, a two-dimensional feature space is constructed, and a dynamic reference light intensity curve is fitted by using a dynamic sliding window algorithm. Based on the dynamic reference light intensity curve and the measured photoluminescence intensity, the defect severity index of each pixel point is calculated by combining a nonlinear index mapping model. The defect severity index is rendered into a defect distribution heat map, and production equipment regulation instructions are generated according to the defect area proportion and distribution form. The application dynamically decouples the interference of film thickness on the photoluminescence signal by using data statistical rules, and improves the accuracy of perovskite film defect detection and the production yield.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor material detection, and particularly relates to a perovskite film quality control system integrating optical detection and data analysis. BACKGROUND

[0002] As a new photovoltaic technology, perovskite solar cells have realized high-efficiency verification in laboratory small-area devices due to excellent photoelectric conversion performance, and have great potential to move towards GW-level mass production scale. However, the preparation process of large-area peroviskite films still faces core bottlenecks, which are key factors restricting its large-scale application. Currently, the mainstream industry adopts a slot coating method to prepare large-area peroviskite films. This process has the advantages of fast film formation speed and high production efficiency, and can meet the production capacity demand of mass production scenarios. However, in the actual production process, the film surface is prone to micron-level thickness fluctuations due to the influence of physical factors such as fluid viscosity, air knife drying rate, and substrate surface flatness. This thickness fluctuation is manifested as obvious fluid stripes in the macroscopic view, which not only destroys the appearance consistency of the film, but also directly negatively affects the photoelectric performance of the film.

[0003] In the quality detection link of peroviskite films, photoluminescence detection technology is widely used for crystal quality evaluation due to its core advantage of non-destructive detection. The detection principle is based on capturing the fluorescence intensity signal to judge the internal state of the crystal after laser excitation of the film. Non-radiative recombination centers in the crystal, as the main form of defects, will cause fluorescence quenching and form characteristic dark spot signals. However, the existing photoluminescence detection technology has a serious signal coupling problem: the fluorescence intensity signal is not only affected by the defect density of the crystal, but also has a significant positive correlation with the physical thickness of the film. In the area with thinner film, the total amount of material excited by the laser decreases, and the fluorescence signal naturally weakens. In the area with thicker film, the total amount of excited material increases, and the fluorescence signal correspondingly strengthens. This thickness-related signal change and the fluorescence quenching signal caused by defects are superimposed on each other, making it difficult for the detection system to distinguish the signal strength and restricting the accuracy of the detection result.

[0004] Existing detection algorithms primarily rely on two core logics: fixed thresholding and full-image mean normalization. When using a fixed thresholding algorithm, the naturally weak fluorescence signal in thinner regions may be misinterpreted as dark spots caused by defects, resulting in numerous false positives. Conversely, in thicker regions, fluorescence quenching signals caused by defects may be masked by strong background fluorescence, leading to serious false negatives. While full-image mean normalization attempts to calibrate the signal through overall mean, it cannot resolve signal biases caused by local thickness fluctuations, and similarly struggles to accurately reflect the true quality of the crystal. This distortion in detection data directly prevents production lines from obtaining accurate crystal quality information, hindering targeted adjustments to production process parameters. Consequently, it severely restricts the production yield of perovskite thin films, increases the cost of large-scale production, and impedes the industrialization of perovskite solar cell technology. Summary of the Invention

[0005] In view of this, the present invention proposes a perovskite thin film quality control system that integrates optical detection and data analysis, in order to solve the technical problem that the existing technology is prone to misjudgment and missed detection due to film thickness fluctuations, and cannot accurately reflect the true crystal quality of perovskite thin films.

[0006] To solve the above-mentioned technical problems, the technical solution of the perovskite thin film quality control system integrating optical detection and data analysis proposed in this invention is as follows:

[0007] The integrated optical inspection and data analysis perovskite thin film quality control system includes the following modules:

[0008] The data acquisition and preprocessing module is used to acquire photoluminescence intensity data and reflectance data on the surface of the perovskite thin film, and to perform spatial pixel registration and preprocessing on the photoluminescence intensity data and the reflectance data to obtain a light intensity matrix and a reflectance matrix.

[0009] The dynamic benchmark construction module is used to construct a two-dimensional feature space with reflectance values ​​as the horizontal axis and photoluminescence intensity values ​​as the vertical axis. It maps the pixels in the light intensity matrix and the reflectance matrix to the two-dimensional feature space and uses a dynamic sliding window algorithm to fit a dynamic benchmark light intensity curve that varies with reflectance.

[0010] The defect index calculation module is used to calculate the current theoretical reference light intensity of each pixel based on the dynamic reference light intensity curve, and combine the measured photoluminescence intensity of the pixel to calculate the defect severity index of each pixel using a nonlinear exponential mapping model.

[0011] The defect analysis and regulation module is configured to render the calculated defect severity index into a defect distribution heat map, and generate corresponding production equipment regulation instructions to adjust the production process parameters according to the defect area proportion and defect distribution form in the defect distribution heat map.

[0012] Further, the formula for constructing the dynamic reference light intensity curve is:

[0013]

[0014] In the formula, is the dynamic reference light intensity when the reflectivity is is a smoothing function, is a high quantile statistical function, is photoluminescence intensity data, is reflectivity data, is a statistical window radius, is a local sample set, is a compensation coefficient, is a robust local standard deviation. By combining the high quantile statistics and the robust local standard deviation, the dynamic reference light intensity curve constructed represents the optimal quality at the thickness and has a certain statistical tolerance. This design cleverly eliminates the high-brightness outliers caused by dust or noise, and at the same time uses the compensation coefficient to ensure that the reference line covers most of the normal points, ensuring the robustness and reliability of the detection standard and avoiding misjudgment caused by statistical fluctuations.

[0015] Further, the high quantile statistical function

[0016] takes the value of the 95th percentile in the local sample set; and the robust local standard deviation only calculates the standard deviation of the samples in the local sample set whose values are between the median and the maximum value. Further, the statistical window radius

[0017] is set to 1% to 2% of the reflectivity range; and the compensation coefficient has a value range of 2 to 3. Further, constructing the dynamic reference light intensity curve further includes: if the number of samples in a certain reflectivity interval is less than a preset threshold, then the

[0018] is generated by the adjacent effective window values using linear interpolation. Further, the formula for calculating the defect severity index is:

[0019]

[0020]

[0021] ​In the formula, is a coordinate is a defect severity index at the coordinate is a sensitivity gain coefficient is a theoretical reference light intensity of a current pixel point is a measured photoluminescence intensity of the current pixel point is a noise floor constant is a one-way truncation function.

[0022] The formula design uses the measured value of the photoluminescence intensity as the denominator, similar to the concept of signal-to-noise ratio, so that when the photoluminescence intensity is quenched to be very low, a small absolute difference can also be converted into a huge defect index. This nonlinear mapping greatly improves the detection sensitivity of serious defects, and the introduction of the noise floor constant prevents calculation overflow under low light intensity, ensuring the numerical stability of the algorithm in various extreme cases.

[0023] Further, the noise floor constant is a dark noise mean value collected by the sensor under a darkroom environment with the excitation light source turned off; the sensitivity gain coefficient is valued at 5-10.

[0024] Further, the generating of the corresponding production equipment control instruction comprises:

[0025] Real-time statistics of the defect area proportion whose defect severity index exceeds a preset warning value; if the defects are point-like and dispersedly distributed, it is determined that the crystallization annealing temperature is abnormal, and an instruction of adjusting the annealing furnace temperature curve is output.

[0026] Further, the generating of the corresponding production equipment control instruction further comprises: if the defects are linearly and continuously distributed, it is determined that the slot coating head is blocked or scratched, and a stop alarm signal is output.

[0027] Further, the smoothing function uses a Savitzky-Golay filter to smoothly connect the calculated discrete points to form a continuously derivable dynamic reference light intensity curve.

[0028] The beneficial effects of this invention are as follows: By constructing a multi-dimensional feature space and an adaptive benchmark model, this invention dynamically decouples film thickness interference using statistical data without requiring prior knowledge of the film thickness distribution model. Compared with existing technologies, this invention abandons the traditional fixed threshold judgment logic and constructs a benchmark line that dynamically changes with film thickness using reflectivity data, solving the problem of large-area misjudgment caused by fluid stripes in the slot coating process. Regardless of the film thickness, as long as its luminous efficiency reaches the expected level for that thickness, it is judged as qualified. In addition, the nonlinear defect severity index constructed by this invention conforms to the semiconductor physical quenching mechanism and has extremely high sensitivity in the low-light region, effectively capturing minute defects. The design of noise floor constant and local robust standard deviation introduced in this invention endows the system with extremely high robustness, enabling it to quickly adapt to production scenarios under different formulations and processes, significantly improving the quality control level of perovskite thin film production lines. Attached Figure Description

[0029] Figure 1 This is a system block diagram of a perovskite thin film quality control system integrating optical detection and data analysis according to the present invention.

[0030] Figure 2 This is a schematic diagram of fitting a two-dimensional feature space to a dynamic benchmark.

[0031] Figure 3 This is a comparison diagram of the final imaging effect between the present invention and the prior art. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0033] Specific embodiments of the perovskite thin film quality control system integrating optical detection and data analysis proposed in this invention are as follows:

[0034] like Figure 1 As shown, the perovskite thin film quality control system integrating optical inspection and data analysis includes a data acquisition and preprocessing module, a dynamic benchmark construction module, a defect index calculation module, and a defect analysis and control module.

[0035] The data acquisition and preprocessing module is used to acquire photoluminescence intensity data and reflectance data on the surface of the perovskite thin film, and to perform spatial pixel registration and preprocessing on the photoluminescence intensity data and the reflectance data to obtain the light intensity matrix and the reflectance matrix.

[0036] Specifically, an integrated optical inspection module is deployed above the conveyor belt of the perovskite thin film production line. This integrated optical inspection module includes a line-scan laser excitation unit, a PL fluorescence acquisition unit, and a reflectance spectroscopy acquisition unit. As the thin film moves at a constant speed with the conveyor belt, the system simultaneously acquires optical data in two dimensions using a line-scan method.

[0037] A high-sensitivity photodetector is used to receive the stimulated emission fluorescence signal from the thin film. After analog-to-digital conversion, a light intensity matrix is ​​formed, denoted as . The light intensity matrix characterizes the luminescence capability of the thin film and is affected by both film thickness and defects. The reflectance signal of the thin film surface to a specific wavelength band is collected using a reflectance spectroscopy probe to form a reflectance matrix, denoted as [equation missing]. In one example, the specific wavelength is the 550nm green light band, which is sensitive to thickness interference effects. Since films of different thicknesses produce different degrees of interference reflection of light, reflectivity data is directly used as a proxy variable to characterize the relative change in film thickness.

[0038] The encoder trigger signal is used to perform spatial pixel registration on the two sensors, a high-sensitivity photodetector and a reflectance spectral probe, to ensure... and The same coordinate in It corresponds to the same physical micro-element on the thin film; at the same time, it removes dead pixels and saturation noise from the data.

[0039] The data acquisition and preprocessing module simultaneously acquires PL data (characterizing luminescence capability) and reflectivity data (characterizing film thickness variation), and performs precise spatial registration, providing a multi-dimensional data foundation for subsequent decoupling of film thickness interference. This dual-channel data acquisition method ensures that the optical properties on the same physical element can be accurately mapped, eliminating data errors caused by differences in sensor position or asynchronous acquisition time, and providing reliable raw data support for accurately evaluating film quality.

[0040] The dynamic reference construction module is used to construct a two-dimensional feature space with reflectance values ​​as the horizontal axis and photoluminescence intensity values ​​as the vertical axis. It maps the pixels in the light intensity matrix and the reflectance matrix to the two-dimensional feature space and uses a dynamic sliding window algorithm to fit a dynamic reference light intensity curve that varies with reflectance.

[0041] Specifically, the system iterates through all pixels in the light intensity matrix and the reflectivity matrix, mapping them to a two-dimensional feature space coordinate system. The horizontal axis represents the reflectivity value. The vertical axis represents the photoluminescence intensity value. This forms a data cloud whose shape varies with the thickness of the film.

[0042] The system utilizes a dynamic sliding window algorithm to move along the horizontal axis and fit the upper envelope of the data cloud. This dynamic reference light intensity curve represents the theoretically optimal luminescence intensity under different film thickness conditions, which is also the defect-free reference light intensity determined by the system. It forms the basis for calculating the defect severity index and provides a pass / fail standard that adapts to film thickness to eliminate interference from film thickness fluctuations in the detection process.

[0043] In this embodiment, the dynamic reference light intensity curve The construction formula is as follows:

[0044]

[0045] In the formula, For reflectivity Dynamic reference light intensity at that time. The statistical window radius is set to 1% to 2% of the reflectivity range, and is used to extract a local slice in the two-dimensional feature space. For a local sample set, it means that all reflectances fall within the range of The set of photoluminescence intensities of pixels within the range. The high quantile statistical function sorts all photoluminescence intensity values ​​in the local sample set from smallest to largest, and then selects the value at the 95th percentile. This value represents the higher luminescence intensity that most normal points can achieve under this specific film thickness, which is close to the theoretical optimal value. The maximum value is not selected in order to eliminate high-brightness anomalies caused by dust or thermal noise. To achieve robust local standard deviation, and to prevent defective pixels from dragging down statistical fluctuations, this robust local standard deviation only calculates the standard deviation of samples whose values ​​in the photoluminescence intensity set are between the median and the maximum value. This is the compensation coefficient, typically ranging from 2 to 3; however, due to unavoidable random noise and statistical fluctuations in the measurement process, if only... If it's an absolute standard, then normal pixels that are slightly above or below that value due to normal statistical fluctuations might be misjudged. Therefore, adding... A robust local standard deviation of 10 times is used as a safety margin. To smooth the function, a Savitzky-Golay filter is used to smoothly connect the calculated discrete points. Furthermore, if the reflectivity interval... If the number of samples is less than a preset threshold, such as 50 points, then... Linear interpolation is used to generate adjacent valid window values.

[0046] To better understand this step, the following calculation example is provided:

[0047] Assuming the current reflectivity range is 0 to 100%, set the statistical window radius. It is necessary to calculate the reflectivity. Dynamic reference light intensity at time The system selects all reflectivities within a certain range. arrive The pixels between, assuming the set of photoluminescence intensity values ​​of these pixels is The set contains 1000 data points.

[0048] First, calculate the higher quantile values. Assuming the 950th value after sorting is 100, then Then calculate the robust local standard deviation. Assuming the median is 90 and the maximum value after removing extreme outliers is 120, calculate the standard deviation of the sample between the median and the maximum value. Assume the result is... Set the compensation coefficient. Then the initial value for the baseline calculation under this reflectivity is: Finally, after Smoothing process yields the final result. This means that, theoretically, for an area with a reflectivity of 30%, its photoluminescence intensity should reach around 110 to be considered perfect and without defects.

[0049] The dynamic benchmark construction module dynamically fits a baseline, essentially setting a pass / fail standard for each pixel based on its own thickness attribute. This method can effectively identify the photoluminescence intensity that theoretically yields the best crystal quality at different thicknesses, thereby eliminating interference caused by film thickness fluctuations in principle. It avoids misjudgments caused by film thinning in traditional fixed threshold methods and achieves dynamic tracking of film thickness changes by the detection standard.

[0050] The defect index calculation module is used to calculate the current theoretical reference light intensity of each pixel based on the dynamic reference light intensity curve, and combine it with the measured photoluminescence intensity of the pixel to calculate the defect severity index of each pixel using a nonlinear exponential mapping model.

[0051] Specifically, after obtaining a baseline that adaptively floats with film thickness, the system calculates the light intensity defect rate for each pixel. To conform to the exponential quenching of carrier lifetime by nonradiative recombination centers in semiconductor physics, a nonlinear defect severity index is constructed. Defect Severity Index The calculation formula is as follows:

[0052]

[0053] In the formula, For the current pixel The defect severity index of the current pixel point, ranging between 0 and 1. The theoretical reference light intensity of the current pixel point, the reflectivity of the pixel point is substituted into the calculation formula of the dynamic reference light intensity curve to obtain the theoretical light intensity at this thickness; The measured photoluminescence intensity of the current pixel point; The one-way truncation function is used to process the case that the measured value is occasionally higher than the reference value due to measurement noise; The noise floor constant is the average dark noise collected in a dark room environment with the excitation light source turned off; The sensitivity gain coefficient, usually ranging between 5 and 10.

[0054] In order to more clearly illustrate the calculation process and effect of the formula, a calculation example is provided as follows:

[0055] Assume that the system setting parameters are: sensitivity gain , noise floor constant .

[0056] Scenario one: a normal point located in the thick film area is detected, the reference value corresponding to the reflectivity is measured , the measured photoluminescence intensity value is , these data are substituted into the above calculation formula to obtain:

[0057] =0.18, which is close to 0, and is determined as qualified.

[0058] Scenario two: a serious defect point located in the thin film area is detected, since the film is thin, the reference value is low, assuming that at this time . Due to the defect, the fluorescence is quenched, the measured photoluminescence intensity value is , these data are substituted into the above calculation formula to obtain:

[0059] ; the calculation result is close to 1, and is determined as a serious defect.

[0060] It should be noted that if it is a simple linear difference, the difference (40) of scenario two is much smaller than the reference value (200) of scenario one, but in the present formula, since the measured photoluminescence intensity value is used, the difference is extremely sensitive in the low light intensity area, so that the defect in the thin film area can be accurately identified.

[0061] ​By introducing the defect severity index, the exponential quenching rule of non-radiative recombination centers in semiconductor materials on carrier lifetime is highly consistent. Especially in the weak light area, the sensitivity of the index increases exponentially, so it can sensitively capture the tiny defects with extremely low contrast in the traditional gray map, and realize the accurate evaluation of the defect severity, greatly reducing the missed detection rate.

[0062] The defect analysis and regulation module is used to render the calculated defect severity index into a defect distribution heat map, and generate corresponding production equipment regulation instructions according to the defect area ratio and defect distribution form in the defect distribution heat map to adjust the production process parameters.

[0063] Specifically, since the originally collected data is processed into the light intensity matrix and the reflectivity matrix, and the calculation of the defect severity index is carried out for each pixel point in the two matrices one by one, the calculated defect severity index is a two-dimensional matrix consistent with the length and width dimensions of the input image. The defect analysis and regulation module renders the two-dimensional matrix into a pseudo-color heat map. The low light intensity area caused by the thinning of the film thickness matches the measured photoluminescence intensity with the dynamic reference light intensity at this thickness, and the calculation result is 0, which is displayed as a white background. The real crystal defect area, whether in the thick film area or the thin film area, will show high values and display red / yellow patches.

[0064] The system real-time statistics the defect area ratio of the defect severity index exceeding the preset warning value, such as 0.5. If the defects are point-like and dispersed, it is determined that the crystallization annealing temperature is abnormal, and the output signal adjusts the annealing furnace temperature curve; if the defects are linear and continuous, it is determined that the slit coating head is blocked or scratched, and the output stops the alarm signal.

[0065] By converting the detection results into intuitive images and specific control instructions, the closed-loop management from detection to control is realized. This intelligent management and control method can trigger corresponding processing strategies according to different defect characteristics, correct production deviations in time, prevent the generation of batch waste, and thus ensure the efficiency and stability of the production process.

[0066] The technical solutions and technical effects of the present application will be further described below. Figure 2 and Figure 3 The technical solutions and technical effects of the present application will be further described below.

[0067] Figure 2The two-dimensional feature space constructed by the present application is fitted with a dynamic reference diagram. In the diagram, the horizontal axis is reflectivity (representing film thickness), and the larger the value, the thicker the film layer; the vertical axis is photoluminescence (PL) signal intensity, representing the material's light-emitting ability; each gray point represents a measured pixel point on the film. As can be seen from the diagram, the data points show a diagonal distribution trend, which conforms to the physical law that the thicker the film layer, the more the total amount of material excited, and the stronger the PL signal. The horizontal dashed line is the fixed threshold line of the prior art, i.e., a fixed photoluminescence intensity value is set as the passing line. In the low film thickness region, a large number of normal data points, although belonging to the normal distribution, are incorrectly judged as unqualified because their absolute positions are lower than the fixed threshold line, which directly shows why the traditional method produces a large number of false positives in the thin film region. Figure 2 The diagonal curve in the diagram is the dynamic reference light intensity curve constructed by the present application, which is generated by using a dynamic sliding window algorithm to closely follow the upper envelope of the data cloud, and is self-adaptive and floating according to the change of reflectivity. In the low film thickness region, the dynamic reference line automatically lowers the standard; in the high film thickness region, the dynamic reference line automatically increases. Only those points that are significantly lower than the dynamic reference line, i.e., points far below the cloud, are calculated as high values. This realizes the decoupling of film thickness and defects, and no matter how the film thickness changes, as long as the light-emitting efficiency reaches the required level at that thickness, it is judged as qualified.

[0068] Figure 3 The comparison diagram between the present application and the prior art in the final imaging effect is shown in the diagram. The prior art detection result diagram is a black and white binary image, and the background is extremely dirty; the large black area in the diagram is actually a film thickness thin area caused by fluid stripes. Since these areas are thin, the photoluminescence (PL) signal intensity is weak, which is lower than the fixed threshold (reference Figure 2 the fixed threshold line of the prior art), so it is directly marked as a black defect by the traditional algorithm; while the real defects, the areas marked with a frame in the diagram, are submerged in these large false defects, causing the system to fail to identify the real process problems, and even fail to produce normally. On the contrary, the detection result diagram of the present application is a color pseudo-color heat map, and the background is pure. The color bar value on the right represents the defect severity index. The vertical wavy lines in the background can still be seen, which are physical thickness fluctuations, but they are all rendered as qualified, indicating that the system confirms through formula calculation that these areas are normal relative to their thickness although the light is weak, eliminating the misjudgment, and the several square-shaped spots clearly shown in the diagram are real defects. It can be seen that the present application eliminates the misjudgment interference caused by film thickness fluctuations, and realizes high signal-to-noise ratio crystal quality imaging.

[0069] While the specification has illustrated and described various embodiments of the application, it will be readily apparent to those of ordinary skill in the art that many modifications, changes and substitutions can be made without departing from the spirit and scope of the application.

Claims

1. A perovskite thin film quality control system integrating optical inspection and data analysis, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to acquire photoluminescence intensity data and reflectance data on the surface of the perovskite thin film, and to perform spatial pixel registration and preprocessing on the photoluminescence intensity data and the reflectance data to obtain a light intensity matrix and a reflectance matrix. A dynamic benchmark construction module is used to construct a two-dimensional feature space with reflectance values ​​as the horizontal axis and photoluminescence intensity values ​​as the vertical axis. It maps the pixels in the intensity matrix and reflectance matrix to this two-dimensional feature space and uses a dynamic sliding window algorithm to fit a dynamic benchmark light intensity curve that varies with reflectance, satisfying the following: For reflectivity Dynamic reference light intensity at time, For smoothing functions, This is a high quantile statistical function. For photoluminescence intensity data, For reflectivity data, To calculate the statistical window radius, For a local sample set, For compensation coefficient, For robust local standard deviation; The defect index calculation module is used to calculate the current theoretical reference light intensity of each pixel based on the dynamic reference light intensity curve, and combine it with the measured photoluminescence intensity of the pixel to calculate the defect severity index of each pixel using a nonlinear exponential mapping model, satisfying the following: coordinates The severity index of defects at the location. This is the sensitivity gain coefficient. The theoretical reference light intensity for the current pixel. The measured photoluminescence intensity of the current pixel. The noise basis constant, It is a one-way truncation function; The defect analysis and control module is used to render the calculated defect severity index into a defect distribution heatmap, and generate corresponding production equipment control commands to adjust production process parameters based on the defect area ratio and defect distribution pattern in the defect distribution heatmap.

2. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 1, characterized in that, The higher quantile statistical function Take the 95th percentile value from the local sample set; the robust local standard deviation. The standard deviation is calculated only for samples in a local sample set whose values ​​are between the median and the maximum.

3. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 1, characterized in that, The statistical window radius The compensation coefficient is set to 1%~2% of the reflectivity range; The value range is 2 to 3.

4. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 1, characterized in that, Constructing a dynamic reference light intensity curve also includes: if the number of samples within a certain reflectivity interval is less than a preset threshold, then the corresponding interval... Linear interpolation is used to generate adjacent valid window values.

5. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 1, characterized in that, The noise floor constant The mean value of dark noise acquired by a sensor in a dark room environment with the excitation light source turned off; the sensitivity gain coefficient The value ranges from 5 to 10.

6. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 1, characterized in that, The generation of corresponding production equipment control commands includes: The system calculates the percentage of defect area whose severity index exceeds the preset warning value in real time. If the defects are distributed in a dotted pattern, it is determined that the crystallization annealing temperature is abnormal, and an instruction to adjust the annealing furnace temperature curve is output.

7. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 6, characterized in that, The generation of corresponding production equipment control instructions also includes: If the defects are distributed in a continuous linear pattern, it is determined that the slit coating head is blocked or scratched, and a shutdown alarm signal is output.

8. The perovskite thin film quality control system integrating optical inspection and data analysis according to claim 1, characterized in that, The smoothing function uses a Savitzky-Golay filter to smoothly connect the calculated discrete points to form a continuously differentiable dynamic reference light intensity curve.

Images