Optical darkroom optical performance detection method and system based on intelligent sensor

Abstract

The application provides an optical darkroom optical performance detection method and system based on an intelligent sensor, relates to the technical field of optical darkroom detection, and comprises the following steps: establishing sample initial classification data set; generating defect space coordinate data set and constructing defect positioning atlas; constructing optical performance core database; marking abnormal wavelength interval and fluorescent signal, establishing abnormal correlation data set; matching abnormal characteristics and component information, and determining the correlation between pollutants and optical abnormalities; constructing a comprehensive feature matrix, and completing performance grade prediction and defect root cause analysis through a deep learning model; and calculating the comprehensive performance score by preset weight and dividing the grade. The application constructs a three-dimensional correlation system, quantifies the correlation by combining various algorithms, accurately locates the defect root cause, improves the analysis scientificity, replaces the manual link with full-process automatic operation, improves the detection and classification efficiency with the aid of the optimization model, reduces the artificial dependence and subjective error, and is suitable for batch detection scenes.

Description

Technical Field

[0001] This invention relates to the field of optical anechoic chamber testing technology, and in particular to a method and system for testing the optical performance of an optical anechoic chamber based on intelligent sensors. Background Technology

[0002] An optical anechoic chamber is a specialized environmental space with light shielding capabilities. Its core design goal is to create a standardized optical experimental environment that allows for precise control of light intensity and elimination of stray light interference through structural sealing, the application of light-shielding materials, and ambient light management. These spaces are widely used in optical instrument research and development, optoelectronic product calibration, imaging system testing, and optical material property analysis, serving as fundamental hardware to ensure the accuracy of optical experiments and testing. The construction of an optical anechoic chamber requires the integration of light-shielding and sealing technologies, ambient light attenuation technologies, and anti-interference designs. Depending on the application requirements, it can be categorized into fully dark chambers and semi-dark chambers. It can also be equipped with integrated light source systems, optical adjustment frames, imaging acquisition equipment, and other auxiliary devices to adapt to the specific needs of different optical testing tasks.

[0003] Optical performance testing is a technical process that relies on optical principles and specialized testing equipment to quantitatively and qualitatively analyze the core optical indicators of optical systems, optoelectronic components, and optical materials. The test results directly provide data support for the R&D optimization, production quality control, and performance verification of optical products. With the widespread penetration of optoelectronic technology in aerospace, precision manufacturing, security monitoring, medical equipment, and consumer electronics, the precision requirements for core performance indicators of optical products, such as imaging quality, light transmittance, resolution, contrast, stray light suppression capability, and spectral characteristics, are continuously increasing. This has driven the iteration of optical performance testing technology and the standardization of optical darkrooms. Optical performance testing and optical darkrooms are closely technically linked. External stray light and ambient light fluctuations can significantly affect the accuracy of optical testing. Optical darkrooms, through strict control of ambient light, provide a stable and repeatable experimental basis for optical performance testing. In the process of technological development, the design of optical darkrooms has gradually moved towards modularity and intelligence, enabling precise control and real-time monitoring of environmental parameters such as light intensity, spectrum, and temperature, adapting to the compatibility needs of various types of optical testing tasks.

[0004] Typical optical performance testing in darkrooms focuses primarily on detecting surface defects in samples, failing to effectively identify hidden defects such as internal pores and interlayer separation, resulting in a lack of comprehensive testing. Furthermore, the absence of a correlation between defect depth and the ability to detect optical anomalies makes it difficult to quantify their causes and correlations, hindering the precise identification of the root cause of defects and leading to inefficient and ineffective remediation efforts. Summary of the Invention

[0005] This invention provides a method and system for optical performance testing in an optical darkroom based on intelligent sensors, which solves the problem that existing technologies do not establish a correlation between defect depth and cannot accurately locate the root cause of defects.

[0006] On one hand, the present invention provides a method for detecting the optical performance of an optical darkroom based on intelligent sensors, comprising:

[0007] S1: Macroscopic visual screening of samples was completed in a dark room with low light assisted environment, and appearance defects and uniformity data were recorded to establish a sample initial classification dataset.

[0008] S2: Based on the three-dimensional morphology and internal structure data of the samples in the initial sample classification dataset, generate a defect spatial coordinate dataset and construct a defect location map;

[0009] S3: Based on the defect location map, construct a core database of optical performance; mark abnormal wavelength ranges and fluorescence signals, and establish an abnormal correlation dataset;

[0010] S4: Based on the anomaly association dataset, match the anomaly features with the component information to clarify the association between pollutants and optical anomalies, and obtain a comprehensive dataset;

[0011] S5: Based on a comprehensive dataset, construct a comprehensive feature matrix, use a deep learning model to predict performance level and analyze the root causes of defects, and output performance analysis conclusions;

[0012] S6: Based on the performance analysis results, calculate the overall performance score according to the preset weights and classify the grades, and generate a standardized test report.

[0013] According to the optical performance testing method for an optical darkroom based on intelligent sensors provided by the present invention, step S2, the step of constructing a defect localization map, includes:

[0014] S21: Analyze the internal defects in the three-dimensional morphology data and internal structure data, including internal pores, impurity distribution, and interlayer bonding state;

[0015] S22: Establish a three-dimensional rectangular coordinate system with the geometric center of the sample as the origin, map the appearance defects and internal defects to the three-dimensional rectangular coordinate system, and generate a defect space coordinate dataset containing defect type, size and position coordinates;

[0016] S23: Based on the defect spatial coordinate dataset, draw two-dimensional planar distribution maps and three-dimensional solid distribution maps, label the confidence level and associated regions of each defect, and construct a defect location map.

[0017] According to the optical performance detection method for an optical darkroom based on intelligent sensors provided by the present invention, step S3, the step of establishing an anomaly correlation dataset includes:

[0018] S31: Based on the defect location map, select the defect area and the corresponding defect-free control area as the detection target point, obtain the optical spectral data of the target wavelength range, and construct a core database of optical performance containing parameters of transmittance, reflectance and absorptivity.

[0019] S32: Based on the core database of optical performance, set the normal threshold range of optical parameters, screen out abnormal wavelength ranges that exceed the threshold range, and capture the fluorescence signal characteristics of the sample under a preset specific excitation light;

[0020] S33: Associate and match the abnormal wavelength range, fluorescence signal characteristics, and the location and type information of the corresponding defects, record the variation range of abnormal parameters, the correspondence between fluorescence signals and defects, remove interference data with no obvious correlation, and establish an abnormal association dataset.

[0021] According to the optical performance detection method for an optical darkroom based on intelligent sensors provided by the present invention, step S4, the step of matching abnormal features with component information, includes:

[0022] S41: Analyze the intrinsic properties of the sample surface and interior for the abnormal regions marked in the abnormal correlation dataset;

[0023] S42: Establish a mapping model between abnormal features and component information, match the changes in optical parameters, fluorescence signal features and intrinsic attribute information corresponding to the abnormal wavelength range in the abnormal correlation dataset, and obtain the correlation information between attribute features and the degree of optical anomaly.

[0024] S43: Based on correlation information, clarify the influence mechanism of intrinsic attribute information on optical performance, eliminate irrelevant attribute interference, and integrate the matching results, correlation analysis and influence mechanism to obtain a comprehensive dataset.

[0025] According to the optical performance detection method for an optical darkroom based on a smart sensor provided by the present invention, step S42, the step of establishing a mapping model includes:

[0026] S421: Normalize the abnormal features and intrinsic attribute information, and output standard abnormal features and standard intrinsic attribute information;

[0027] S422: Based on standard anomaly features and standard intrinsic attribute information, construct a mapping model framework to obtain an initial mapping model;

[0028] S423: Train the initial mapping model based on historical detection data to obtain the mapping model;

[0029] S424: Input the data to be matched into the mapping model and output the correlation coefficient between the attribute features and the degree of optical anomaly.

[0030] According to the optical performance testing method for an optical darkroom based on intelligent sensors provided by the present invention, step S5, which involves using a deep learning model to predict performance level and analyze the root causes of defects, includes:

[0031] S51: Based on the comprehensive dataset, extract core indicators including defect features, optical parameter features, and component features, and construct a comprehensive feature matrix;

[0032] S52: Based on the comprehensive feature matrix, a hybrid prediction model is used to predict the optical performance level of the sample and output the prediction result of the optical performance level of the sample.

[0033] S53: Combining the sample optical performance level prediction results with the anomaly correlation dataset, the root cause of the defect is analyzed through causal reasoning algorithm, and the performance analysis conclusion is output.

[0034] According to the optical performance detection method for an optical darkroom based on a smart sensor provided by the present invention, step S52, which involves predicting the optical performance level of the sample using a hybrid prediction model, includes:

[0035] S521: Divide the comprehensive feature matrix into training set, validation set and test set according to a preset ratio;

[0036] S522: Spatial features are extracted from the comprehensive feature matrix using the CNN module, and temporal features are extracted using the LSTM module;

[0037] S523: Set the model hyperparameters, train the hybrid prediction model based on the training set, and dynamically adjust the hyperparameters through the validation set to obtain the trained hybrid prediction model;

[0038] S524: Input the test set into the trained hybrid prediction model. When the prediction accuracy reaches the preset accuracy level, the model is deemed qualified, and the prediction result of the optical performance level of the sample is output.

[0039] According to the optical performance testing method for an optical darkroom based on a smart sensor provided by the present invention, step S53, the step of outputting performance analysis conclusions, includes:

[0040] S531: Based on the causal reasoning algorithm, and combining the defect location, type, optical parameter anomaly characteristics and component information of the anomaly association dataset, a causal relationship network between defects and optical performance anomalies is constructed.

[0041] S532: Quantify the weights of each influencing factor in the causal network, prioritize the identification of core root causes whose weight exceeds the preset weight, and output a root cause impact report.

[0042] S533: Based on the root cause impact report and combined with the sample optical performance level prediction results, formulate root cause improvement recommendations;

[0043] S534: Integrate the sample optical performance level prediction results, root cause impact report and root cause improvement suggestions to obtain performance analysis conclusions.

[0044] According to the optical performance testing method for an optical darkroom based on a smart sensor provided by the present invention, step S6, the step of calculating the comprehensive performance score and classifying the grades, includes:

[0045] S61: Based on the performance analysis conclusions, preset the weights of each performance indicator;

[0046] S62: Score each performance indicator and calculate the overall performance score using a weighted summation method;

[0047] S63: Performance levels are divided based on comprehensive performance scores, and the scope of use and limitations corresponding to each performance level are also indicated.

[0048] S64: Generate a standardized test report based on the comprehensive performance score, performance level, and performance analysis conclusions.

[0049] The present invention also provides an optical performance testing system for an optical darkroom based on intelligent sensors, comprising:

[0050] The sample screening module is used to complete the macroscopic visual screening of samples based on the low light assisted environment of the dark room, record appearance defects and uniformity data, and establish a sample initial classification dataset.

[0051] The defect localization module is used to generate a defect spatial coordinate dataset based on the three-dimensional morphology and internal structure data of the sample in the initial sample classification dataset, and to construct a defect localization map and a core database of optical performance; it also marks abnormal wavelength ranges and fluorescence signals to establish an abnormal correlation dataset.

[0052] The anomaly correlation module is used to match anomaly features with component information based on the anomaly correlation dataset, clarify the correlation between pollutants and optical anomalies, and obtain a comprehensive dataset.

[0053] The performance level prediction module is used to construct a comprehensive feature matrix based on a comprehensive dataset, complete performance level prediction and defect root cause analysis through a deep learning model, and output performance analysis conclusions.

[0054] The rating module is used to calculate the comprehensive performance score based on the performance analysis results, according to preset weights, and to classify the scores and generate a standardized test report.

[0055] The optical performance testing method and system based on intelligent sensors in an optical darkroom provided by this invention improves the scientific nature of the analysis by constructing a three-dimensional correlation system of "defect features-optical parameters-composition information", combining multiple algorithms to quantify the correlation and accurately locate the root cause of defects; the fully automated operation replaces manual steps, improves the efficiency of detection and grading with the help of optimized models, reduces reliance on manual labor and subjective errors, and is suitable for batch detection scenarios. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the optical performance detection method for an optical darkroom based on a smart sensor provided in an embodiment of the present invention;

[0058] Figure 2 This is a flowchart of establishing an abnormal correlation dataset provided in an embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of the optical performance testing system for an optical darkroom based on a smart sensor, provided in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0061] Example:

[0062] The following is combined with Figures 1-3 This invention describes a method and system for detecting the optical performance of an optical darkroom based on intelligent sensors.

[0063] like Figures 1-2 As shown in the embodiment of the present invention, the optical performance detection method for an optical darkroom based on a smart sensor includes:

[0064] S1: Perform macroscopic visual screening of samples in a darkroom with low-light assisted environment, record appearance defects and uniformity data, and establish a preliminary sample grading dataset. The low-light assisted environment refers to providing uniform auxiliary illumination in a darkroom using an adjustable low-illuminance LED light source (illuminance range 5-50 lx), which meets visual imaging requirements without altering the inherent optical properties of the sample. Appearance defects include visible surface defects such as scratches, cracks, stains, deformation, and edge damage. Uniformity data refers to the macroscopic uniformity parameters of sample surface color and thickness. The steps for establishing the preliminary sample grading dataset are as follows:

[0065] The sample to be tested is fixed on a special stage in a darkroom with horizontal adjustment and positioning functions, and the surface dust is gently wiped away with a fibrous wiping cloth and anhydrous ethanol.

[0066] Turn on the darkroom environment control system. After the temperature, humidity, and vibration parameters reach the preset standards and stabilize for 30 minutes, start the low-light auxiliary light source, adjust the illuminance to the preset value, and calibrate the uniformity of the light source.

[0067] Using a high-definition industrial camera paired with a macro lens, macroscopic images are acquired from six directions: top, bottom, left, right, front, and back of the sample. Simultaneously, thickness data for each area of ​​the sample is acquired using a laser thickness gauge.

[0068] A machine vision recognition algorithm based on the YOLOv8 model is used to analyze the acquired images, automatically identifying the type and size of appearance defects. Combined with thickness data, uniformity deviation is calculated, and samples with no defects, minor defects, moderate defects, and severe defects are labeled as Level 1, Level 2, Level 3, and Level 4, respectively. An initial classification dataset is constructed, containing sample numbers, defect information, uniformity data, and classification results. The steps for analyzing the acquired images using the YOLOv8 model-based machine vision recognition algorithm include:

[0069] The acquired raw images are processed by grayscale conversion, Gaussian blur denoising, and image enhancement to improve image clarity and defect identification.

[0070] Load the pre-trained YOLOv8 model and fine-tune it based on an optical sample appearance defect dataset. Set the confidence threshold to 0.6 and the IOU threshold to 0.5, and initialize the model inference parameters. IOU stands for Intersection over Union, which is a core metric in the field of object detection used to measure the degree of overlap between two bounding boxes, such as the defect box predicted by the algorithm and the real defect box. Essentially, it is the ratio of the intersection area to the union area of ​​the two boxes.

[0071] The preprocessed image is input into the fine-tuned YOLOv8 model. The model extracts image features through the backbone network, fuses multi-scale features through the neck network, and finally outputs the bounding box coordinates, category, and confidence of the defect through the head network.

[0072] The length, width, and area of ​​the defect are calculated based on the bounding box coordinates, and then converted into actual dimensions using the image acquisition scale.

[0073] Suspected defects with a confidence level below 0.6 are eliminated, and non-maximum suppression algorithm is used to remove duplicates from overlapping bounding boxes to ensure the uniqueness of the detection results.

[0074] S2: Based on the three-dimensional morphology and internal structure data of the samples in the initial sample grading dataset, generate a defect spatial coordinate dataset and construct a defect location map. The steps include:

[0075] S21: Analyze internal defects, including internal porosity, impurity distribution, and interlayer bonding state, from 3D morphology and internal structure data. 3D morphology data refers to spatial data reflecting the three-dimensional contour of the sample surface, acquired using a 3D laser scanning microscope. Internal structure data refers to invisible structural information such as the internal interlayer structure, pore distribution, and impurity content, acquired through micro-CT scanning. Low-dose radiation is used during the scanning process to avoid damaging the sample's optical properties. Specific implementation steps include: denoising the 3D laser scanning and micro-CT scanning data to remove noise interference; segmenting the preprocessed data using a threshold segmentation algorithm to identify the location and distribution of internal pores and impurities; analyzing the interlayer bonding state using an edge detection algorithm to determine the presence of defects such as interlayer separation and bubbles; classifying internal defects into pore, impurity, and interlayer defect categories, and recording the preliminary size and distribution range of each type of defect.

[0076] S22: Establish a three-dimensional Cartesian coordinate system with the geometric center of the sample as the origin. Map external and internal defects to this system, generating a defect spatial coordinate dataset containing defect type, size, and location coordinates. The defect spatial coordinate dataset refers to structured data containing defect type, size, and spatial location after mapping various defects to a unified coordinate system. The steps for generating the defect spatial coordinate dataset include:

[0077] Import the 3D topographic data of the sample into 3D modeling software, determine the geometric center of the sample, and establish a 3D Cartesian coordinate system of X, Y, and Z, where the X-axis is along the sample length, the Y-axis is along the sample width, and the Z-axis is along the sample thickness. The geometric center of the sample is determined as follows: Let the coordinates of any n feature points in the 3D space of the sample be (x1, y1, z1), (x2, y2, z2), ..., (x... n ,y n ,z n Geometric center coordinates (O) X O Y, O Z The calculation formula is expressed as follows:

[0078]

[0079]

[0080]

[0081] Where n is the total number of feature points selected in the three-dimensional space of the sample, i is the index of the feature point, and O X O Y O Z These are the coordinates of the geometric center of the sample on the X, Y, and Z axes in a three-dimensional Cartesian coordinate system.

[0082] For each external and internal defect, the center point of the defect is selected as the feature point. The coordinates (x, y, z) of the feature point in the three-dimensional coordinate system are read by a coordinate measurement tool, and the length, width, height and other dimensional parameters of the defect are recorded.

[0083] Enter the defect number, type, size parameters, spatial coordinates (x, y, z), and classification information into a structured table to generate a defect spatial coordinate dataset.

[0084] S23: Based on the defect spatial coordinate dataset, draw two-dimensional planar distribution maps and three-dimensional solid distribution maps, label the confidence level and associated regions of each defect, and construct a defect location map. The defect location map is a visual map that intuitively displays the spatial distribution and relationships of defects.

[0085] Using visualization tools, based on the spatial coordinate data of defects, two-dimensional defect distribution maps of the XY, XZ, and YZ planes are drawn, and a three-dimensional distribution map is constructed. Different types of defects are marked with different colors: red for appearance defects, blue for porosity defects, green for impurities, and yellow for interlayer defects.

[0086] By calculating the accuracy of the defect identification algorithm and combining it with the results of manual review, the confidence level C of each defect is determined, expressed by the formula:

[0087]

[0088] Where α is the probability of correct recognition by the algorithm, and γ is the manual verification confirmation coefficient, which is 1 or 0.5, where 1 indicates correct recognition and 0.5 indicates pending confirmation.

[0089] Based on the spatial distance method, the distance D between any two defect center points is calculated. When D ≤ a preset threshold (set according to sample size, usually 5mm), it is determined that a spatial correlation exists, the correlated area is marked, and a correlation line is drawn. The formula for calculating the distance between any two defect center points is expressed as:

[0090]

[0091] Where (x1,y1,z1) and (x2,y2,z2) are the coordinates of any two defect center points.

[0092] By integrating two-dimensional distribution maps, three-dimensional distribution maps, confidence level annotations, and related region information, a defect location map is generated.

[0093] S3 constructs a core database of optical performance based on defect location maps. It marks anomalous wavelength ranges and fluorescence signals, establishing an anomalous correlation dataset. The steps include:

[0094] S31: Based on the defect location map, select defect areas and corresponding defect-free control areas as detection targets, acquire optical spectral data within the target wavelength range, and construct a core database of optical performance including parameters such as transmittance, reflectance, and absorptivity. Specifically, this includes:

[0095] In the defect localization map, for each defect region, three feature points, including the defect center and its surrounding area, are selected as detection targets. At the same time, control targets of the same number and location are selected in the defect-free area of ​​the sample to ensure the symmetry of the target selection.

[0096] Align the spectrophotometer probe with the target point, set the target wavelength range (depending on the sample type, such as the 400-760nm visible light band for optical lenses), and collect the transmittance and reflectance data of each target point at different wavelengths.

[0097] According to the law of conservation of energy, the absorbance at each wavelength can be calculated using the following formula:

[0098]

[0099] Where A is the absorptivity at each wavelength, i.e., the ratio of absorbed light intensity to incident light intensity, ranging from 0 to 1. T is the transmittance, i.e., the ratio of transmitted light intensity to incident light intensity, ranging from 0 to 1. R is the reflectivity, i.e., the ratio of reflected light intensity to incident light intensity, ranging from 0 to 1.

[0100] The target number, corresponding defect information, wavelength value, T, R, and A parameters are entered into the database and stored using a MySQL database. Data can be retrieved by defect type and wavelength range.

[0101] S32: Based on optical parameter data of defect-free standard samples of the same type in the core database of optical performance, a statistical analysis method is used to set the normal threshold range for optical parameters. The tested samples are compared with the normal threshold range to screen for abnormal wavelength ranges exceeding the threshold range. The excitation wavelength is set (based on the sample composition; for example, 365nm ultraviolet excitation is typically used for organic impurities), and the excitation intensity is 100μW / cm². 2It captures the fluorescence signal characteristics of a sample under a preset specific excitation light, including characteristic parameters such as fluorescence emission wavelength, fluorescence intensity, and fluorescence lifetime.

[0102] S33: Associate and match the abnormal wavelength range, fluorescence signal characteristics with the location and type information of the corresponding defects, record the variation range of abnormal parameters, the correspondence between fluorescence signals and defects, use correlation analysis algorithms to remove interference data with no obvious correlation, and establish an abnormal association dataset.

[0103] S4: Based on the anomaly association dataset, match the anomaly features with the component information to clarify the association between pollutants and optical anomalies, and obtain a comprehensive dataset.

[0104] Step S4, the steps of matching abnormal features with component information, include:

[0105] S41: Analyze the intrinsic properties of the sample surface and interior based on the abnormal regions and corresponding features marked in the abnormal correlation dataset.

[0106] S42: Establish a mapping model between abnormal features and component information, match the changes in optical parameters, fluorescence signal features and intrinsic attribute information corresponding to the abnormal wavelength range in the abnormal correlation dataset, and obtain the correlation information between attribute features and the degree of optical anomaly.

[0107] The steps to establish a mapping model include:

[0108] S421: Normalize the abnormal features and intrinsic attribute information, outputting standard abnormal features and standard intrinsic attribute information. Since the abnormal features and intrinsic attribute information have different dimensions, a min-max normalization algorithm is used to map the data to the [0,1] interval to eliminate the influence of dimensions. The min-max normalization formula is expressed as x'=(xx min ) / (x max -x min ), where x is the original data, x min x is the minimum value of this type of data. max x' represents the maximum value of this type of data, and x' represents the normalized standard data.

[0109] S422: Based on standard anomaly features and standard intrinsic attribute information, a mapping model framework is constructed to obtain the initial mapping model. Specifically, a neural network model is used to construct the mapping framework. The input layer consists of standard anomaly features (the dimension is determined according to the number of features, usually 8-12 dimensions), and there are 3 hidden layers (each with 64, 32, and 16 neurons respectively). The activation function is the ReLU function f(x)=max(0,x), and the output layer is the predicted value corresponding to the standard intrinsic attribute information, thus constructing the initial mapping model.

[0110] S423: Train the initial mapping model based on historical detection data to obtain the mapping model. Collect historical detection data of similar samples, dividing them into a training set (70%), a validation set (15%), and a test set (15%). Input the training set into the initial mapping model, using mean squared error as the loss function, and iteratively train using the gradient descent algorithm, with 1000 iterations. After each iteration, evaluate the model performance using the validation set. When the loss function value on the validation set tends to stabilize, i.e., the loss change is ≤0.0001 over 20 consecutive iterations, stop training, adjust model parameters (number of hidden layer neurons, learning rate, etc.), and optimize model accuracy.

[0111] S424: Input the data to be matched into the mapping model, the model outputs the corresponding attribute feature prediction value, and combined with the true value of the attribute feature, outputs the correlation coefficient between the attribute feature and the degree of optical anomaly.

[0112] S43: Based on correlation information and the correlation coefficient matrix, clarify the influence mechanism of intrinsic attribute information on optical performance, such as specific impurity elements leading to a decrease in transmittance, and organic pollutants producing characteristic fluorescence signals. Irrelevant attribute interference is eliminated, and core attribute characteristics are retained. The matching results, correlation analysis, and influencing mechanisms are integrated to obtain a comprehensive dataset.

[0113] S5: Based on a comprehensive dataset, construct a comprehensive feature matrix, and use a deep learning model to predict performance levels and analyze root causes of defects, outputting performance analysis conclusions. The comprehensive feature matrix refers to transforming the core indicators in the comprehensive dataset into structured data in matrix form.

[0114] The steps involved in using deep learning models to predict performance levels and analyze root causes of defects include:

[0115] S51: Based on the comprehensive dataset, extract core indicators including defect features, optical parameter features, and component features. Convert categorical indicators into numerical indicators through one-hot encoding, and retain the original values ​​of continuous indicators. Arrange the quantified core indicators by row to construct a comprehensive feature matrix.

[0116] S52: Based on the comprehensive feature matrix, a hybrid prediction model is used to predict the optical performance level of the sample, and the predicted optical performance level of the sample is output. The hybrid prediction model refers to a deep learning model that combines CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory Network).

[0117] The steps for predicting the optical performance level of a sample using a hybrid prediction model include:

[0118] S521: Divide the comprehensive feature matrix into a training set, a validation set, and a test set according to a preset ratio. The ratio is 70% for the training set, 15% for the validation set, and 15% for the test set.

[0119] S522: Spatial features are extracted from the comprehensive feature matrix through the CNN module, and the spatial features output by the CNN module are input into the LSTM module to extract temporal features.

[0120] S523: Set the model hyperparameters. Train the hybrid prediction model based on the training set, and dynamically adjust the hyperparameters using the validation set to obtain the trained hybrid prediction model. Initial hyperparameters include learning rate η = 0.001, batch size = 32, and number of iterations (epochs) = 200. Input the training set into the hybrid prediction model, use the cross-entropy loss function as the classification loss, and iteratively train using the Adam optimizer. After each iteration, evaluate the model's prediction accuracy using the validation set, and dynamically adjust hyperparameters such as the learning rate and batch size until the model performance stabilizes.

[0121] S524: Input the test set into the trained hybrid prediction model and calculate the prediction accuracy Acc = (Number of correctly predicted samples / Total number of samples) × 100%. When the prediction accuracy reaches the preset accuracy level, the model is deemed qualified, and the prediction results of the sample optical performance level are output. The sample optical performance level is divided into four levels: excellent, good, qualified, and unqualified. The model outputs the performance level and corresponding confidence level for each sample.

[0122] S53: Combining the sample optical performance level prediction results with the anomaly correlation dataset, analyze the root cause of defects using a causal reasoning algorithm, and output performance analysis conclusions. The steps include:

[0123] S531: A causal reasoning algorithm based on Bayesian networks as its core, which combines the defect location, type, optical parameter anomaly features and component information of the anomaly association dataset. Specifically, the defect type and component information are used as cause nodes, and the optical parameter anomaly and performance level are used as result nodes. Based on the anomaly association dataset, the conditional probability between each node is calculated to construct a causal relationship network between defects and optical performance anomalies.

[0124] S532: Quantify the weights of each influencing factor in the causal relationship network, prioritize the identification of core root causes whose weight exceeds the preset percentage, and output a root cause impact report.

[0125] The steps for quantifying the weights of various influencing factors in a causal network include:

[0126] Clearly define the target layer, the criteria layer, and the solution layer to form a clear hierarchical relationship.

[0127] For the criterion layer and the solution layer, the importance of any two factors is compared pairwise using the 1-9 scaling method, generating a judgment matrix S with dimensions k×k (k being the number of factors at the corresponding level), satisfying S=[s ab ]k×k , where s ab Let represent the importance scale of factor a relative to factor b, and s ab =1 / s ba s aa =1. In the 1-9 scale, 1 indicates that the two factors are equally important, 3 indicates that the former is slightly more important than the latter, 5 indicates that the former is significantly more important than the latter, 7 indicates that the former is strongly more important than the latter, 9 indicates that the former is extremely more important than the latter, 2, 4, 6, and 8 are intermediate transitional scales, and the reciprocal indicates the opposite contrast relationship.

[0128] First, calculate the largest eigenvalue λ of the judgment matrix. max Then calculate the consistency index CI=(λ) max -k) / (k-1), combined with the average random consistency index RI, we get the consistency ratio CR=CI / RI. When CR≤0.1, the consistency of the judgment matrix is ​​deemed qualified.

[0129] The eigenvector method is used to find the largest eigenvalue λ corresponding to the judgment matrix S. max The eigenvectors are normalized to obtain the weights of each factor, ensuring that the sum of all factor weights is 1.

[0130] S533: Based on the root cause impact report and combined with the sample optical performance level prediction results, formulate root cause improvement recommendations, including recommendations for optimizing raw material screening processes and improving the cleanliness of the production environment for impurity-related root causes; recommendations for adjusting lamination temperature and pressure parameters for interlayer defect-related root causes; and specify the implementation steps, expected effects, and testing and verification methods for the improvement measures.

[0131] S534: Integrate the sample optical performance level prediction results, root cause impact report and root cause improvement suggestions, supplement the sample basic information, test conditions, data reliability description, and obtain performance analysis conclusions.

[0132] S6: Based on the performance analysis conclusions, calculate the comprehensive performance score according to preset weights and classify the grades, generating a standardized test report. The standardized test report must conform to the industry standard (such as GB / T 26828-2011) and include test information, data results, analysis conclusions, improvement suggestions, etc.

[0133] Step S6, which involves calculating the overall performance score and classifying the levels, includes:

[0134] S61: Based on the performance analysis conclusions, preset the weights of each performance index. The total weight for optical parameters is 0.5, for defect indices it is 0.3, for component purity it is 0.2, and the total weight is 1. The weights can be dynamically adjusted according to the sample type; for example, for high-precision optical lenses, the weight of optical parameters can be increased to 0.6.

[0135] S62: Each performance indicator is scored, and a weighted summation is used to calculate the overall performance score S, expressed by the formula:

[0136]

[0137] Wherein, S1 is the score of optical parameter index, S2 is the score of defect index, S3 is the score of component purity index, and w1, w2, and w3 are the weights of the corresponding indexes.

[0138] S63: Performance levels are categorized based on comprehensive performance scores, including: Level A (S≥90 points); Level B (80≤S<90 points); Level C (60≤S<80 points); and Level D (S<60 points). The applicable scope and limitations for each performance level are also indicated. Applicable scenarios are specified for different levels; for example, Level A is suitable for high-end optical equipment, Level B for general optical equipment, Level C for low-precision optical equipment, and Level D is prohibited.

[0139] S64: Generate a standardized test report based on the comprehensive performance score, performance level, and performance analysis conclusions.

[0140] like Figure 3 As shown, the present invention also provides an optical performance testing system for an optical darkroom based on intelligent sensors, comprising: a sample screening module, a defect location module, an anomaly correlation module, a performance level prediction module, and a level classification module.

[0141] The sample screening module is used to complete macroscopic visual screening of samples in a darkroom low-light assisted environment, record appearance defects and uniformity data, and establish a sample initial classification dataset.

[0142] The defect localization module generates a defect spatial coordinate dataset based on the three-dimensional morphology and internal structure data of the sample in the initial sample classification dataset, and constructs a defect localization map and a core database of optical performance. It also marks abnormal wavelength ranges and fluorescence signals to establish an abnormal correlation dataset.

[0143] The anomaly association module is used to match anomaly features with component information based on the anomaly association dataset, clarify the association between pollutants and optical anomalies, and obtain a comprehensive dataset.

[0144] The performance level prediction module is used to construct a comprehensive feature matrix based on the comprehensive dataset, complete performance level prediction and defect root cause analysis through a deep learning model, and output performance analysis conclusions.

[0145] The grading module is used to calculate the comprehensive performance score and grade it according to the performance analysis conclusions and preset weights, and generate a standardized test report.

[0146] In summary, the optical performance testing method and system based on intelligent sensors in an optical darkroom provided by this invention improves the scientific nature of the analysis by constructing a three-dimensional correlation system of "defect features-optical parameters-composition information", combining multiple algorithms to quantify the correlation and accurately locate the root cause of defects; the fully automated operation replaces manual steps, improves the efficiency of detection and grading with the help of optimized models, reduces reliance on manual labor and subjective errors, and is suitable for batch detection scenarios.

[0147] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the optical performance of an optical darkroom based on intelligent sensors, characterized in that, include: S1: Macroscopic visual screening of samples was completed in a dark room with low light assisted environment, and appearance defects and uniformity data were recorded to establish a sample initial classification dataset. S2: Based on the three-dimensional morphology and internal structure data of the sample in the initial sample classification dataset, generate a defect spatial coordinate dataset and construct a defect location map; S3: Based on the aforementioned defect location map, construct a core database of optical performance; mark abnormal wavelength ranges and fluorescence signals, and establish an abnormal correlation dataset; S4: Based on the aforementioned anomaly association dataset, match the anomaly features with the component information to clarify the association between pollutants and optical anomalies, and obtain a comprehensive dataset; S5: Based on the comprehensive dataset, construct a comprehensive feature matrix, use a deep learning model to complete performance level prediction and defect root cause analysis, and output performance analysis conclusions; S6: Based on the performance analysis conclusions, calculate the comprehensive performance score according to the preset weights and classify the grades, and generate a standardized test report.

2. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 1, characterized in that, Step S2, the steps for constructing the defect location map include: S21: Analyze the internal defects in the three-dimensional morphology and internal structure data, including internal pores, impurity distribution, and interlayer bonding state; S22: Establish a three-dimensional rectangular coordinate system with the geometric center of the sample as the origin, and map the appearance defects and the internal defects to the three-dimensional rectangular coordinate system to generate a defect space coordinate dataset containing defect type, size, and position coordinates; S23: Based on the defect spatial coordinate dataset, draw a two-dimensional planar distribution map and a three-dimensional solid distribution map, mark the confidence level and associated region of each defect, and construct the defect location map.

3. The optical performance testing method for an optical darkroom based on intelligent sensors according to claim 1, characterized in that, Step S3, the steps for establishing the anomaly correlation dataset include: S31: Based on the defect location map, select the defect area and the corresponding defect-free control area as the detection target point, obtain the optical spectral data of the target wavelength range, and construct a core database of optical performance containing parameters of transmittance, reflectance and absorptivity. S32: Based on the core database of optical performance, set the normal threshold range of optical parameters, screen out abnormal wavelength ranges that exceed the threshold range, and capture the fluorescence signal characteristics of the sample under a preset specific excitation light; S33: Associate and match the abnormal wavelength range, the fluorescence signal characteristics, and the location and type information of the corresponding defects; record the variation range of abnormal parameters, the correspondence between fluorescence signals and defects; remove interference data with no obvious correlation; and establish the abnormal association dataset.

4. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 1, characterized in that, Step S4, the steps of matching abnormal features with component information, include: S41: For the abnormal regions marked in the abnormal correlation dataset, analyze the intrinsic attribute information of the sample surface and interior; S42: Establish a mapping model between the abnormal features and component information, and match the changes in optical parameters and fluorescence signal features corresponding to the abnormal wavelength range in the abnormal correlation dataset with the features in the intrinsic attribute information to obtain the correlation information between the attribute features and the degree of optical anomaly. S43: Based on the correlation information, clarify the influence mechanism of the intrinsic attribute information on optical performance, eliminate irrelevant attribute interference, and integrate the matching results, correlation analysis and influence mechanism to obtain a comprehensive dataset.

5. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 4, characterized in that, Step S42, the steps for establishing the mapping model include: S421: Normalize the abnormal features and the intrinsic attribute information to output standard abnormal features and standard intrinsic attribute information; S422: Based on the standard anomaly features and the standard intrinsic attribute information, construct a mapping model framework to obtain an initial mapping model; S423: Train the initial mapping model based on historical detection data to obtain the mapping model; S424: Input the data to be matched into the mapping model and output the correlation coefficient between the attribute features and the degree of optical anomaly.

6. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 1, characterized in that, Step S5, which involves using a deep learning model to predict performance levels and analyze the root causes of defects, includes the following steps: S51: Based on the comprehensive dataset, extract core indicators including defect features, optical parameter features, and component features, and construct a comprehensive feature matrix; S52: Based on the comprehensive feature matrix, a hybrid prediction model is used to predict the optical performance level of the sample, and the prediction result of the optical performance level of the sample is output. S53: Combining the predicted optical performance level of the sample with the abnormal correlation dataset, the root cause of the defect is analyzed using a causal reasoning algorithm, and the performance analysis conclusion is output.

7. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 6, characterized in that, Step S52, which involves using a hybrid prediction model to predict the optical performance level of the sample, includes: S521: Divide the comprehensive feature matrix into a training set, a validation set, and a test set according to a preset ratio; S522: Extract spatial features from the comprehensive feature matrix using a CNN module, and extract temporal features using an LSTM module; S523: Set the model hyperparameters, train the hybrid prediction model based on the training set, and dynamically adjust the hyperparameters through the validation set to obtain the trained hybrid prediction model; S524: Input the test set into the trained hybrid prediction model. When the prediction accuracy reaches the preset accuracy level, the model is deemed qualified, and the prediction result of the optical performance level of the sample is output.

8. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 6, characterized in that, Step S53, the step of outputting the performance analysis conclusion, includes: S531: Based on the causal reasoning algorithm, and combining the defect location, type, optical parameter anomaly characteristics and component information of the aforementioned abnormal association dataset, construct a causal relationship network between defects and optical performance anomalies. S532: Quantify the weights of each influencing factor in the causal network, prioritize the identification of core root causes whose weight exceeds the preset weight, and output a root cause impact report. S533: Based on the root cause impact report and the predicted results of the sample optical performance level, formulate root cause improvement recommendations; S534: Integrate the sample optical performance level prediction results, root cause impact report and root cause improvement suggestions to obtain performance analysis conclusions.

9. The optical performance testing method for an optical darkroom based on a smart sensor according to claim 1, characterized in that, Step S6, which involves calculating the overall performance score and classifying the levels, includes: S61: Based on the performance analysis conclusions, preset the weights of each performance index; S62: Score each performance indicator and calculate the overall performance score using a weighted summation method; S63: Based on the comprehensive performance score, classify the performance level and indicate the scope of use and limitations corresponding to the performance level; S64: Based on the comprehensive performance score, the performance level, and the performance analysis conclusion, generate a standardized test report.

10. An optical performance testing system for an optical darkroom based on intelligent sensors, which employs the optical performance testing method for an optical darkroom based on intelligent sensors as described in any one of claims 1 to 9, characterized in that, include: The sample screening module is used to complete macroscopic visual screening of samples in a dark room with low light assistance, record appearance defects and uniformity data, and establish a sample initial classification dataset. The defect localization module is used to generate a defect spatial coordinate dataset based on the three-dimensional morphology and internal structure data of the sample in the initial sample classification dataset, and to construct a defect localization map and a core database of optical performance; it also marks abnormal wavelength ranges and fluorescence signals to establish an abnormal correlation dataset. An anomaly association module is used to match anomaly features with component information based on the anomaly association dataset, clarify the association between pollutants and optical anomalies, and obtain a comprehensive dataset. The performance level prediction module is used to construct a comprehensive feature matrix based on the comprehensive dataset, complete performance level prediction and defect root cause analysis through a deep learning model, and output performance analysis conclusions. The rating module is used to calculate the comprehensive performance score and classify the rating based on the performance analysis conclusions according to preset weights, and generate a standardized test report.

Images