Mini / micro led industry chain quality monitoring method and system based on large model

Abstract

The application discloses a kind of Mini / MicroLED industry chain quality condition monitoring method and system based on large model, it is related to industrial big data technology field, first by multi-source sensor network and automated detection equipment collect the historical production environment data of each link of industry chain, then utilize time series feature extraction algorithm and correlation analysis algorithm to extract intra-link feature and inter-link correlation feature, then based on deep learning technology and graph neural network constructs quality evaluation model, real-time monitoring the quality state of each link of industry chain, when detecting quality anomaly, identify quality abnormal link and inter-link quality transmission path by link correlation tracing algorithm, finally according to quality anomaly trace information, generate quality improvement scheme by link collaborative optimization algorithm.It improves the efficiency of quality anomaly trace.

Description

Technical Field

[0001] This invention relates to the field of industrial big data technology, specifically a method and system for monitoring the quality status of the Mini / MicroLED industry chain based on a large model. Background Technology

[0002] Mini / MicroLED, as a next-generation display technology, boasts advantages such as high brightness, low power consumption, and long lifespan, demonstrating enormous potential in high-end display applications. However, the Mini / MicroLED industry chain involves multiple complex stages, including chip fabrication, mass transfer, driver integration, and packaging. The quality status of each stage not only affects the quality of its own products but also influences the production efficiency of downstream stages through complex interrelationships, forming a quality transfer chain.

[0003] However, traditional quality monitoring methods often employ a single-stage, independent monitoring model, lacking a systematic analysis of the quality correlations between different stages of the industrial chain, making it difficult to trace the source of quality problems. For example, temperature fluctuations in the chip fabrication stage may affect the alignment accuracy of the mass transfer stage through changes in material properties, but existing monitoring systems cannot establish such cross-stage quality transfer relationships.

[0004] Therefore, this invention proposes a method and system for monitoring the quality status of the Mini / MicroLED industry chain based on a large model. Summary of the Invention

[0005] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a method and system for monitoring the quality status of the Mini / MicroLED industry chain based on a large model, thereby improving the efficiency of tracing the source of quality anomalies.

[0006] To achieve the above objectives, a method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model is proposed, including the following steps:

[0007] Step 1: Obtain historical production environment data for each link in the industry chain, including chip fabrication, mass transfer, driver integration, and packaging.

[0008] Step 2: Extract quality features from the historical production environment data to obtain the quality features of the historical production environment data. The quality features include intra-process features and inter-process correlation features.

[0009] Step 3: Construct a quality assessment model based on the aforementioned quality characteristics. The quality assessment model includes a process quality assessment sub-model and a process correlation assessment sub-model.

[0010] Step 4: Use the quality assessment model to monitor the real-time production data of each link in the industrial chain in real time, and obtain the quality monitoring results;

[0011] Step 5: Determine whether the quality monitoring results meet the preset quality threshold conditions;

[0012] Step 6: If the quality monitoring results do not meet the preset quality threshold conditions, the quality monitoring results are analyzed through a process association tracing algorithm to identify the abnormal quality process and the quality transfer path between processes, and to obtain the quality abnormality source tracing information.

[0013] Step 7: Based on the quality anomaly tracing information, generate a quality improvement plan through a process collaborative optimization algorithm. The quality improvement plan includes intra-process optimization measures and inter-process collaborative optimization measures.

[0014] The process of obtaining historical production environment data for each link in the industrial chain includes the following steps:

[0015] Step 11: Collect historical production environment data for the chip fabrication process through a multi-source sensor network, including temperature, humidity, air pressure, cleanliness, and process parameters;

[0016] Step 12: Collect historical production environment data for the mass transfer process using automated testing equipment, including transfer accuracy, alignment error, yield, and environmental parameters;

[0017] Step 13: Collect historical production environment data of the driver integration process through the integration test platform, including driver circuit parameters, integration yield and environmental conditions;

[0018] Step 14: Collect historical production environment data of the packaging process through the packaging testing system, including packaging material performance, packaging process parameters and environmental conditions.

[0019] The process of extracting quality features from the historical production environment data includes the following steps:

[0020] Step 21: Extract the intra-process features of the historical production environment data using a time-series feature extraction algorithm, including statistical features, trend features, and anomaly features;

[0021] Step 22: Extract the inter-process correlation features of the historical production environment data through correlation analysis algorithms, including temporal correlation features, causal correlation features, and quality transfer features;

[0022] Step 23: Integrate the intra-stage features and inter-stage correlation features using a feature fusion algorithm to form a complete quality feature set;

[0023] The process of constructing a quality assessment model based on the quality characteristics includes the following steps:

[0024] Step 31: Construct a process quality assessment sub-model using deep learning technology to evaluate the quality status within each process.

[0025] Step 32: Construct a sub-model for evaluating the correlation between links using graph neural network technology to assess the quality correlation status between links;

[0026] Step 33: Integrate the process quality assessment sub-model and the process correlation assessment sub-model using an ensemble learning method to form a unified quality assessment model;

[0027] The real-time monitoring of production data at each stage of the industrial chain using the quality assessment model includes the following steps:

[0028] Step 41: Acquire real-time production data of each link in the industrial chain through the data acquisition interface and perform preprocessing;

[0029] Step 42: Extract real-time features that match the quality features from the preprocessed real-time production data using the feature extraction module;

[0030] Step 43: Input the real-time features into the quality assessment model through the model inference engine to generate real-time quality monitoring results;

[0031] The process of determining whether the quality monitoring result meets the preset quality threshold condition includes the following steps:

[0032] Step 51: Set and maintain the preset quality threshold conditions through the threshold management module, including the process quality threshold and the process-related quality threshold;

[0033] Step 52: Compare the quality monitoring results with the preset quality threshold conditions using a threshold comparison algorithm to generate a comparison result;

[0034] Step 53: The decision logic module determines whether the quality monitoring result meets the preset quality threshold condition based on the comparison result, and generates a judgment result;

[0035] If the quality monitoring result does not meet the preset quality threshold condition, the analysis of the quality monitoring result through the process correlation traceability algorithm includes the following steps:

[0036] Step 61: Analyze the quality monitoring results using the anomaly localization module to identify the quality anomaly process;

[0037] Step 62: Identify the quality transfer path between the links based on the output of the link association evaluation sub-model using the association analysis module;

[0038] Step 63: Integrate the information on the quality anomaly links and the quality transfer path between the links through the root cause analysis module to generate the quality anomaly tracing information;

[0039] The step of generating a quality improvement plan based on the quality anomaly tracing information and using a process collaborative optimization algorithm includes the following steps:

[0040] Step 71: Based on the quality anomaly tracing information, the optimization target setting module determines the optimization target and constraints for quality improvement;

[0041] Step 72: Generate intra-process optimization measures for the quality anomaly process using the intra-process optimization module;

[0042] Step 73: Generate inter-process collaborative optimization measures based on the inter-process quality transfer path through the inter-process collaboration module;

[0043] Step 74: Integrate the intra-stage optimization measures and inter-stage collaborative optimization measures into a complete quality improvement plan through the solution integration module.

[0044] A quality monitoring system for the Mini / MicroLED industry chain based on a large model is proposed, including a data acquisition module, a quality feature extraction module, a model building module, a real-time monitoring module, a threshold judgment module, an anomaly tracing module, and a quality optimization module; wherein, the modules are connected to each other by electrical means.

[0045] The data acquisition module acquires historical production environment data for each link in the industrial chain, including chip fabrication, mass transfer, driver integration, and packaging, and sends the historical production environment data to the quality feature extraction module.

[0046] The quality feature extraction module extracts quality features from the historical production environment data to obtain the quality features of the historical production environment data. The quality features include intra-process features and inter-process correlation features. The quality features are then sent to the model building module.

[0047] The model building module constructs a quality assessment model based on the quality characteristics. The quality assessment model includes a process quality assessment sub-model and a process correlation assessment sub-model, and sends the quality assessment model to the real-time monitoring module.

[0048] The real-time monitoring module uses the quality assessment model to monitor the real-time production data of each link in the industrial chain, obtains the quality monitoring results, and sends the quality monitoring results to the threshold judgment module.

[0049] The threshold judgment module determines whether the quality monitoring result meets the preset quality threshold condition. If it does, the process ends; otherwise, the quality monitoring result is sent to the anomaly tracing module.

[0050] The anomaly tracing module analyzes the quality monitoring results through a process association tracing algorithm, identifies the quality anomaly process and the quality transfer path between processes, obtains the quality anomaly tracing information, and sends the quality anomaly tracing information to the quality optimization module.

[0051] The quality optimization module generates a quality improvement plan based on the quality anomaly tracing information using a process collaborative optimization algorithm. The quality improvement plan includes intra-process optimization measures and inter-process collaborative optimization measures.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] First, historical production environment data for each link in the industry chain is collected through a multi-source sensor network and automated detection equipment. Then, time-series feature extraction algorithms and correlation analysis algorithms are used to extract intra-link features and inter-link correlation features. Next, a quality assessment model is built based on deep learning technology and graph neural networks to monitor the quality status of each link in the industry chain in real time. When a quality anomaly is detected, a link correlation tracing algorithm is used to identify the link with the quality anomaly and the quality transmission path between links. Finally, based on the quality anomaly tracing information, a link collaborative optimization algorithm is used to generate a quality improvement plan. This improves the efficiency of quality anomaly tracing in the Mini / MicroLED industry chain. Attached Figure Description

[0054] Figure 1 This is a flowchart of a method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model, as described in Embodiment 1 of the present invention.

[0055] Figure 2 This is a module connection diagram of a Mini / MicroLED industry chain quality monitoring system based on a large model, as shown in Embodiment 2 of the present invention. Detailed Implementation

[0056] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0057] Example 1

[0058] like Figure 1As shown, a method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model includes the following steps:

[0059] Step 1: Obtain historical production environment data for each link in the industry chain, including chip fabrication, mass transfer, driver integration, and packaging.

[0060] Step 2: Extract quality features from the historical production environment data to obtain the quality features of the historical production environment data. The quality features include intra-process features and inter-process correlation features.

[0061] Step 3: Construct a quality assessment model based on the aforementioned quality characteristics. The quality assessment model includes a process quality assessment sub-model and a process correlation assessment sub-model.

[0062] Step 4: Use the quality assessment model to monitor the real-time production data of each link in the industrial chain in real time, and obtain the quality monitoring results;

[0063] Step 5: Determine whether the quality monitoring results meet the preset quality threshold conditions;

[0064] Step 6: If the quality monitoring results do not meet the preset quality threshold conditions, the quality monitoring results are analyzed through a process association tracing algorithm to identify the abnormal quality process and the quality transfer path between processes, and to obtain the quality abnormality source tracing information.

[0065] Step 7: Based on the quality anomaly tracing information, generate a quality improvement plan through a process collaborative optimization algorithm. The quality improvement plan includes intra-process optimization measures and inter-process collaborative optimization measures.

[0066] The process of obtaining historical production environment data for each link in the industrial chain includes the following steps:

[0067] Step 11: Collect historical production environment data for the chip fabrication process through a multi-source sensor network, including information such as temperature, humidity, air pressure, cleanliness, and process parameters.

[0068] Specifically, the multi-source sensor network consists of temperature sensors, humidity sensors, pressure sensors, particle counters, and process parameter recorders distributed throughout the chip fabrication workshop. The temperature sensors are PT100 platinum resistance temperature sensors with a measurement accuracy of ±0.1℃ and a sampling frequency of 10 seconds / time. The humidity sensors are capacitive humidity sensors with a measurement range of 0-100%RH and an accuracy of ±2%RH. The pressure sensors are piezoresistive pressure sensors with a measurement range of 80-110kPa and an accuracy of ±0.1kPa. The particle counter can detect particle concentrations at four particle size levels: 0.3μm, 0.5μm, 1μm, and 5μm, with a sampling flow rate of 2.83L / min. The process parameter recorder collects key process parameters such as epitaxial growth temperature, gas flow rate, chamber pressure, and growth rate. All sensor data is aggregated to a data acquisition server via an industrial IoT gateway, forming a time-series database. The data storage format is a structured format of timestamp-parameter name-parameter value.

[0069] Step 12: Collect historical production environment data of the mass transfer process through automated testing equipment, including information such as transfer accuracy, alignment error, yield and environmental parameters.

[0070] Specifically, the automated inspection equipment includes a high-precision vision inspection system, a laser alignment system, and an environmental monitoring system. The high-precision vision inspection system uses a 4K resolution industrial camera equipped with a macro lens, achieving a pixel accuracy of 1μm, to detect the transfer position accuracy of Mini / MicroLED chips. The laser alignment system employs infrared laser ranging technology with a measurement accuracy of ±0.5μm, used to monitor alignment errors during the transfer process in real time. The environmental monitoring system includes temperature and humidity sensors, an electrostatic discharge detector, and a vibration detector, used to monitor the stability of the transfer environment. The automated inspection equipment transmits the collected data to a central database via industrial Ethernet. Data is collected once per transfer batch, with each batch containing 1000-10000 Mini / MicroLED units. The data format includes batch number, timestamp, transfer accuracy distribution, alignment error statistics, yield data, and environmental parameter values.

[0071] Step 13: Collect historical production environment data of the driver integration process through the integration test platform, including information such as driver circuit parameters, integration yield and environmental conditions.

[0072] Specifically, the integrated testing platform consists of an electrical parameter tester, a functional testing system, and an environmental monitoring unit. The electrical parameter tester uses a high-precision digital multimeter and oscilloscope to measure the voltage, current, power, and timing parameters of the drive circuit. The voltage measurement accuracy is ±0.01V, the current measurement accuracy is ±0.1mA, and the timing measurement accuracy is ±1ns. The functional testing system uses automated test equipment (ATE), equipped with dedicated test fixtures and test programs, to test the integrated functionality and performance of the driver IC and the Mini / MicroLED chip. The environmental monitoring unit monitors environmental parameters such as temperature, humidity, static electricity, and airflow during the integration process. The integrated testing platform uploads test data to the quality management system in real time, with a data acquisition cycle of once per hour. Each test sample consists of 50-100 integrated units, and the data includes electrical parameter statistics, functional test pass rate, environmental parameter records, and integration yield.

[0073] Step 14: Collect historical production environment data of the packaging process through the packaging testing system, including information such as packaging material performance, packaging process parameters and environmental conditions.

[0074] Specifically, the packaging testing system includes a material performance analyzer, a process parameter monitor, and an environmental monitoring device. The material performance analyzer employs an infrared spectrometer, a thermogravimetric analyzer, and a mechanical property tester to detect the chemical composition, thermal stability, and mechanical strength of the packaging materials. The process parameter monitor records process parameters such as temperature curves, pressure changes, time control, and gas flow rates during the packaging process. The environmental monitoring device monitors environmental conditions in the packaging workshop, including temperature, humidity, cleanliness, and airflow velocity. The packaging testing system stores data in the production management system using dedicated data acquisition software. Data is acquired once per packaging batch, with each batch containing 100-500 packaging units. The data includes material performance indicators, process parameter records, environmental condition data, and packaging yield information.

[0075] The process of extracting quality features from the historical production environment data includes the following steps:

[0076] Step 21: Extract the intra-process features of the historical production environment data using a time-series feature extraction algorithm, including statistical features, trend features, and anomaly features.

[0077] Specifically, the time-series feature extraction algorithm first preprocesses the historical production environment data for each stage, including missing value imputation, outlier detection, and data standardization. Missing value imputation uses moving average and interpolation methods; outlier detection uses the 3σ criterion and box plot method; and data standardization uses Z-score standardization. Then, the algorithm calculates statistical features, including mean, standard deviation, skewness, kurtosis, quantiles, and coefficient of variation; extracts trend features, including linear trend coefficients, periodic indicators, autocorrelation coefficients, and seasonality indices; and identifies anomalous features, including abrupt change points, anomalous duration, and anomalous magnitude. The algorithm employs a sliding window technique with a 24-hour window size and a 1-hour sliding step, extracting features from the key parameters of each stage to form a feature vector within that stage.

[0078] Step 22: Extract the inter-process correlation features of the historical production environment data through the correlation analysis algorithm, including time-series correlation features, causal correlation features, and quality transfer features.

[0079] Specifically, the correlation analysis algorithm first constructs a time alignment matrix for parameters between stages, matching data from different stages based on product batch numbers and timestamps. Then, the algorithm calculates time-series correlation features, using the cross-correlation function (CCF) to analyze the time lag relationship between parameters in different stages, and calculating Pearson correlation coefficient, Spearman rank correlation coefficient, and mutual information value to quantify the correlation strength between parameters. It extracts causal correlation features, using Granger causality tests and transfer entropy methods to identify the direction and strength of causal relationships between stages. Finally, it identifies quality transfer features by tracking the changes in quality indicators between stages, establishing a quality transfer matrix to quantify the impact of upstream stage parameter changes on downstream stage quality indicators. The correlation analysis algorithm performs pairwise analysis on key parameters between all stages, forming an inter-stage correlation feature matrix.

[0080] Step 23: Integrate the intra-stage features and inter-stage correlation features using a feature fusion algorithm to form a complete quality feature set.

[0081] Specifically, the feature fusion algorithm employs a multi-level fusion strategy, including data-level fusion, feature-level fusion, and decision-level fusion. In data-level fusion, the original data from different stages are aligned by time and batch to form a unified data view. In feature-level fusion, Principal Component Analysis (PCA) and an autoencoder are used to reduce the dimensionality of features within each stage, and tensor decomposition is used to compress inter-stage related features. Then, feature concatenation is used to integrate the two types of features. In decision-level fusion, a weighted voting mechanism is used to integrate decision results based on different feature subsets. The feature fusion algorithm also includes a feature selection module, which uses the L1 regularized LASSO algorithm and a tree-model-based feature importance assessment method to select the feature subset with the greatest influence on quality assessment, ultimately forming a quality feature vector of dimension d, where d is typically between 50 and 200, depending on the complexity of the industry chain.

[0082] The process of constructing a quality assessment model based on the quality characteristics includes the following steps:

[0083] Step 31: Construct a process quality assessment sub-model using deep learning technology to evaluate the quality status within each process.

[0084] Specifically, the stage quality assessment sub-model employs a hybrid architecture of Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). The LSTM network, used to capture long-term dependencies in temporal features, consists of three LSTM layers, each containing 128 neurons, using the tanh activation function and a dropout rate of 0.3 for regularization. The CNN network, used to extract local feature patterns, consists of two convolutional layers, each using 64 3×3 convolutional kernels, employing the ReLU activation function and max pooling. The input to the stage quality assessment sub-model is the in-stage feature vector of each stage, and the output is the stage quality score (a real number between 0 and 1) and the probability distribution of quality anomalies. Model training uses the Adam optimizer with a learning rate of 0.001, a batch size of 64, and 200 training epochs, employing an early stopping strategy to avoid overfitting. The training and validation datasets are divided in an 8:2 ratio, and a weighted combination of mean squared error (MSE) and cross-entropy loss functions is used as the optimization objective. For the four stages of chip fabrication, mass transfer, driver integration, and packaging, independent sub-models for stage quality assessment are constructed.

[0085] Step 32: Construct a sub-model for evaluating the correlation between links using graph neural network technology to assess the quality correlation status between links.

[0086] Specifically, the sub-model for evaluating link correlation adopts a Graph Attention Network (GAT) architecture, treating each link in the industry chain as a node in a graph, and the quality correlation between links as edges. The GAT network contains three graph attention layers, each with eight attention heads, a hidden layer dimension of 64, and employs the LeakyReLU activation function (negative slope of 0.2) and a residual connection structure. The input to the sub-model is the inter-link correlation feature matrix and link node features; the output is the inter-link quality transfer strength matrix and correlation anomaly detection results. Model training uses the AdamW optimizer with a learning rate of 0.0005, weight decay of 0.01, a batch size of 32, and 150 training epochs. Graph data augmentation techniques, including edge masking and feature perturbation, are employed during training to enhance the model's generalization ability. The sub-model can adaptively learn the quality transfer patterns between links, identifying key quality transfer paths and potential quality risk points.

[0087] Step 33: Integrate the process quality assessment sub-model and the process correlation assessment sub-model using an ensemble learning method to form a unified quality assessment model.

[0088] Specifically, the ensemble learning method employs a stacking architecture, using the process quality assessment sub-model and the process correlation assessment sub-model as base learners, and integrating the prediction results of each base learner through a meta-learner. The meta-learner uses a gradient boosting decision tree (GBDT) model, containing 100 decision trees with a maximum depth of 5, a learning rate of 0.1, and mean squared error as the loss function. The training process of the ensemble learning method uses K-fold cross-validation (K=5) to avoid information leakage. The final output of the quality assessment model includes: quality scores for each process, the strength of quality correlations between processes, an overall quality status assessment of the supply chain, and a quality anomaly warning signal. The quality assessment model also includes an uncertainty estimation module, employing the Monte Carlo dropout method to calculate the confidence interval of the prediction results through multiple forward propagations, providing a reliability assessment for the quality monitoring results.

[0089] The real-time monitoring of production data at each stage of the industrial chain using the quality assessment model includes the following steps:

[0090] Step 41: Acquire real-time production data of each link in the industrial chain through the data acquisition interface and perform preprocessing.

[0091] Specifically, the data acquisition interface uses Industrial Internet of Things (IIoT) protocols (such as OPC UA, MQTT, Modbus TCP) to establish real-time communication connections with production equipment and sensor networks at each stage. The data acquisition frequency is set according to the process characteristics of different stages: 10 seconds / time for chip fabrication, 1 minute / time for mass transfer, 5 minutes / time for driver integration, and 10 minutes / time for packaging. The preprocessing process includes data cleaning, outlier handling, and data standardization. Data cleaning uses a moving median filter algorithm to remove noise; outlier handling uses a modified Z-score method to identify and replace outliers; and data standardization uses a Min-Max normalization method to map each parameter value to the [0,1] interval. The preprocessing module also includes a data integrity check mechanism. When data is missing, forward padding or interpolation estimation methods are used to fill in the missing data according to the proportion of missing data, ensuring the continuity and integrity of the real-time data stream.

[0092] Step 42: Extract real-time features that match the quality features from the preprocessed real-time production data using the feature extraction module.

[0093] Specifically, the feature extraction module employs the same feature extraction algorithm as in step two, but it is optimized for real-time scenarios, including an incremental calculation strategy and a feature caching mechanism. The incremental calculation strategy uses a sliding window technique for statistical features, with a window size of 4 hours, calculating only the impact of newly added data on feature values ​​each time to avoid redundant calculations; it uses an online learning algorithm for trend features, updating trend model parameters in real time; and it uses real-time anomaly detection algorithms for abnormal features, including Local Outlier Factor (LOF) and Isolation Forest. The feature caching mechanism maintains the feature calculation results for the most recent 24 hours for extracting inter-stage correlation features. The real-time feature vector output by the feature extraction module maintains the same dimension and semantics as the input features of the quality assessment model, ensuring the accuracy of model inference.

[0094] Step 43: Input the real-time features into the quality assessment model through the model inference engine to generate real-time quality monitoring results.

[0095] Specifically, the model inference engine employs a lightweight deep learning framework, supporting model quantization and acceleration techniques to ensure real-time inference. The model inference process first inputs real-time features into the process quality assessment sub-model to calculate the quality score and anomaly probability of each process. Then, it inputs process features and inter-process correlation features into the process correlation assessment sub-model to calculate the inter-process quality transfer strength and correlation anomaly detection results. Finally, a meta-learner integrates the outputs of each sub-model to generate a comprehensive quality monitoring result. The quality monitoring result includes: real-time quality scores (real numbers between 0 and 1) for each process, quality anomaly probability (real numbers between 0 and 1), an inter-process quality correlation strength matrix (n×n matrix, where n is the number of processes), an overall supply chain quality status assessment (four levels: excellent, good, medium, and poor), and quality anomaly warning signals (four levels: normal, slightly abnormal, moderately abnormal, and severely abnormal). The model inference engine operates at an inference frequency of 1 minute per iteration, ensuring timely detection of quality anomalies.

[0096] The process of determining whether the quality monitoring result meets the preset quality threshold condition includes the following steps:

[0097] Step 51: Set and maintain the preset quality threshold conditions through the threshold management module, including the process quality threshold and the process-related quality threshold.

[0098] Specifically, the threshold management module constructs a multi-level threshold system based on historical quality data and expert knowledge. The process quality thresholds include: a lower limit threshold for quality scoring (usually set to 0.7), an upper limit threshold for the probability of quality anomalies (usually set to 0.3), and a threshold for the fluctuation range of key parameters (set according to the process requirements of different parameters); the process-related quality thresholds include: a threshold for changes in the intensity of quality correlation (usually set to ±20%) and a threshold for the confidence level of correlation anomaly detection (usually set to 0.8).

[0099] Step 52: Compare the quality monitoring results with the preset quality threshold conditions using a threshold comparison algorithm to generate a comparison result.

[0100] Specifically, the threshold comparison algorithm employs a multi-condition logical judgment method to compare each dimension of the quality monitoring results with the corresponding threshold conditions. For the process quality score, it determines whether it is below the lower limit threshold; for the quality anomaly probability, it determines whether it is above the upper limit threshold; for key parameter fluctuations, it determines whether they exceed the fluctuation range threshold; for changes in quality correlation strength, it determines whether they exceed the change threshold; and for the confidence level of correlation anomaly detection, it determines whether it is above the confidence level threshold. The threshold comparison algorithm uses a weighted scoring mechanism, assigning different weights to the comparison results of different dimensions to calculate a comprehensive comparison score.

[0101] Step 53: The decision logic module determines whether the quality monitoring result meets the preset quality threshold condition based on the comparison result, and generates a judgment result.

[0102] Specifically, the decision logic module adopts a decision tree structure. Based on the comprehensive score of the comparison results and the specific comparison situation of each dimension, it determines whether the quality monitoring results meet the preset quality threshold conditions. When the comprehensive score is lower than 0.6 or the comparison result of any key dimension shows a significant excess of the threshold (exceeding 50%), it is determined that the preset quality threshold conditions are not met; when the comprehensive score is higher than 0.8 and the comparison results of all key dimensions are within the threshold range, it is determined that the preset quality threshold conditions are met; for cases in between,

[0103] If the quality monitoring result does not meet the preset quality threshold condition, the analysis of the quality monitoring result through the process correlation traceability algorithm includes the following steps:

[0104] Step 61: Analyze the quality monitoring results using the anomaly localization module to identify the quality anomaly process.

[0105] Specifically, the anomaly localization module employs a contribution analysis method and anomaly pattern matching technology to identify quality anomalies. The contribution analysis method calculates the contribution of each stage's quality score to the overall quality status, identifying stages with significantly reduced contributions. The anomaly pattern matching technology matches the current quality monitoring results with a predefined anomaly pattern library, identifying the most similar anomaly pattern and its corresponding anomaly stage. The anomaly localization module also employs a confidence assessment mechanism, assigning a confidence score (a real number between 0 and 1) to each identified anomaly stage to reflect the reliability of the anomaly judgment. When multiple stages are simultaneously identified as anomalies, the anomaly localization module sorts the anomaly stages according to their severity and confidence level, determining the primary and secondary anomalies. The anomaly localization module outputs a list of anomaly stages, including stage name, anomaly severity, confidence level, and anomaly characteristic description.

[0106] Step 62: Identify the quality transfer path between the links based on the output of the link association evaluation sub-model through the association analysis module.

[0107] Specifically, the correlation analysis module employs causal inference methods and path analysis techniques to identify quality transfer paths between links. The causal inference method, based on the quality correlation strength matrix output by the link correlation assessment sub-model, applies Bayesian networks and structural equation modeling (SEM) to infer the direction and strength of causal relationships between links. The path analysis technique treats the supply chain as a directed graph, using shortest path algorithms and critical path analysis methods to identify the quality transfer path from the source link of the anomaly to the link exhibiting the anomaly. The correlation analysis module supports path analysis in multi-source anomaly scenarios; when multiple anomaly links are detected, it can identify their mutual influence relationships and common influence paths. The correlation analysis module outputs a quality transfer path graph, including path nodes (links), path edges (quality transfer relationships), transfer strength, and transfer delay.

[0108] Step 63: Integrate the information on the quality anomaly links and the quality transfer path between the links through the root cause analysis module to generate the quality anomaly tracing information.

[0109] Specifically, the root cause analysis module employs a multidimensional analysis framework and an expert knowledge base for tracing the origins of quality anomalies. The multidimensional analysis framework analyzes anomalies from three dimensions: time, space, and parameters. The time dimension analyzes the time point of occurrence, duration, and historical similar cases of the anomaly; the spatial dimension analyzes the specific location, impact range, and spatial distribution characteristics of the anomaly; and the parameter dimension analyzes key parameters related to the anomaly, parameter change trends, and interactions between parameters. The expert knowledge base includes historical quality problem cases, process expert experience rules, and an equipment failure mode library, which assist in root cause analysis through case reasoning and rule matching. The root cause analysis module outputs quality anomaly tracing information, including: the root cause of the anomaly, the specific reason for the anomaly, the anomaly propagation path, impact assessment, and the chain of evidence. This quality anomaly tracing information is in a structured format, facilitating the generation and execution of subsequent quality improvement plans.

[0110] The step of generating a quality improvement plan based on the quality anomaly tracing information and using a process collaborative optimization algorithm includes the following steps:

[0111] Step 71: Based on the quality anomaly tracing information, the optimization target setting module determines the optimization target and constraints for quality improvement.

[0112] Specifically, the optimization goal setting module sets hierarchical optimization goals based on the severity, scope of impact, and urgency of the quality anomaly tracing information. These optimization goals include: short-term goals (resolving urgent anomalies within 24 hours), medium-term goals (restoring normal production within 7 days), and long-term goals (improving overall quality levels within 30 days). The constraints include: resource constraints (available equipment, manpower, and materials), time constraints (production plans and delivery deadlines), cost constraints (optimizing the investment budget), and technical constraints (the range of process parameter adjustments). The optimization goal setting module uses the Analytic Hierarchy Process (AHP) to determine the priority and weight of each goal and employs fuzzy comprehensive evaluation to assess the stringency of the constraints. The optimization goals and constraints are output in a structured form as input parameters for subsequent optimization algorithms.

[0113] Step 72: Generate the intra-process optimization measures for the quality abnormality process through the intra-process optimization module.

[0114] Specifically, the in-process optimization module employs a knowledge graph-based reasoning system and parameter optimization algorithms to generate in-process optimization measures. The knowledge graph-based reasoning system includes a process knowledge graph, an equipment knowledge graph, and a quality problem knowledge graph. Through semantic reasoning and similar case retrieval, it matches optimization measure templates suitable for the current abnormal situation. The parameter optimization algorithm uses a Bayesian optimization method to search for the optimal parameter combination in the process parameter space and predict the quality improvement effect of different parameter adjustment schemes. The optimization measures generated by the in-process optimization module include: process parameter adjustment schemes (specific parameter names, adjustment direction, and adjustment range), equipment maintenance schemes (maintenance items, maintenance cycle, and maintenance standards), material replacement schemes (replacement material specifications and usage conditions), and operating procedure optimization schemes (adjustment of operating steps and key control points). These in-process optimization measures are prioritized according to implementation requirements and include expected improvement effects and risk assessment information.

[0115] Step 73: Generate inter-process collaborative optimization measures based on the inter-process quality transfer path through the inter-process collaboration module.

[0116] Specifically, the inter-process collaboration module employs a multi-agent collaborative optimization algorithm and a game theory model to generate inter-process collaborative optimization measures. The multi-agent collaborative optimization algorithm treats each process as an agent with autonomous decision-making capabilities, seeking the global optimal solution through information sharing and negotiation mechanisms. The game theory model analyzes the interests and strategy choices of each process during quality improvement, designing incentive mechanisms to promote inter-process collaboration. The inter-process collaborative optimization measures include: a cross-process parameter collaborative adjustment scheme (coordinated change strategies for parameters in upstream and downstream processes), a quality information sharing mechanism (real-time transmission rules for key quality data), a quality standard collaborative optimization scheme (consistent adjustment of quality standards across processes), and a quality responsibility collaborative mechanism (joint prevention and handling processes for quality problems). These inter-process collaborative optimization measures emphasize systemic thinking, improving the overall quality collaboration level of the industrial chain by optimizing interfaces and interactions between processes.

[0117] Step 74: Integrate the intra-stage optimization measures and inter-stage collaborative optimization measures into a complete quality improvement plan through the solution integration module.

[0118] Specifically, the solution integration module employs a hierarchical organizational structure and a time-series scheduling strategy to integrate and optimize measures. The hierarchical organizational structure divides optimization measures into an overall layer (overall quality improvement direction), a tactical layer (inter-process coordination strategies), and an operational layer (specific implementation measures). The time-series scheduling strategy, based on the urgency, implementation difficulty, and interdependencies of the measures, formulates short-term action plans (within 24 hours), medium-term improvement plans (within 7 days), and long-term optimization plans (within 30 days). The quality improvement solution includes: a description of the problem background (abnormal phenomena and root cause analysis), a description of optimization goals (expected improvement effects and time requirements), specific optimization measures (intra-process and inter-process measures), an implementation roadmap (time nodes and milestones), a resource allocation plan (human resources, equipment, and material requirements), and an effectiveness evaluation method (evaluation indicators and evaluation cycle). The quality improvement solution uses a structured document format, facilitating understanding and execution at each stage, while supporting dynamic adjustment and feedback optimization mechanisms to ensure the effectiveness and adaptability of the solution implementation.

[0119] like Figure 2 As shown, a quality monitoring system for the Mini / MicroLED industry chain based on a large model includes a data acquisition module, a quality feature extraction module, a model building module, a real-time monitoring module, a threshold judgment module, an anomaly tracing module, and a quality optimization module; wherein, the various modules are connected to each other electrically.

[0120] The data acquisition module acquires historical production environment data for each link in the industrial chain, including chip fabrication, mass transfer, driver integration, and packaging, and sends the historical production environment data to the quality feature extraction module.

[0121] The quality feature extraction module extracts quality features from the historical production environment data to obtain the quality features of the historical production environment data. The quality features include intra-process features and inter-process correlation features. The quality features are then sent to the model building module.

[0122] The model building module constructs a quality assessment model based on the quality characteristics. The quality assessment model includes a process quality assessment sub-model and a process correlation assessment sub-model, and sends the quality assessment model to the real-time monitoring module.

[0123] The real-time monitoring module uses the quality assessment model to monitor the real-time production data of each link in the industrial chain, obtains the quality monitoring results, and sends the quality monitoring results to the threshold judgment module.

[0124] The threshold judgment module determines whether the quality monitoring result meets the preset quality threshold condition. If it does, the process ends; otherwise, the quality monitoring result is sent to the anomaly tracing module.

[0125] The anomaly tracing module analyzes the quality monitoring results through a process association tracing algorithm, identifies the quality anomaly process and the quality transfer path between processes, obtains the quality anomaly tracing information, and sends the quality anomaly tracing information to the quality optimization module.

[0126] The quality optimization module generates a quality improvement plan based on the quality anomaly tracing information using a process collaborative optimization algorithm. The quality improvement plan includes intra-process optimization measures and inter-process collaborative optimization measures.

[0127] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model, characterized in that, Includes the following steps: Step 1: Obtain historical production environment data for each link in the industry chain, including chip fabrication, mass transfer, driver integration, and packaging. Step 2: Extract quality features from the historical production environment data to obtain the quality features of the historical production environment data. The quality features include intra-process features and inter-process correlation features. The method for extracting the inter-linkage features is as follows: The correlation analysis algorithm is used to extract the inter-process correlation features of the historical production environment data, including time-series correlation features, causal correlation features, and quality transfer features. The correlation analysis algorithm first constructs a time alignment matrix for parameters between stages, matching data from different stages based on product batch numbers and timestamps. Then, it calculates time-series correlation features, using cross-correlation functions to analyze the time lag relationship between parameters in different stages, and calculating Pearson correlation coefficient, Spearman rank correlation coefficient, and mutual information value to quantify the correlation strength between parameters. It extracts causal correlation features, using Granger causality tests and transfer entropy methods to identify the direction and strength of causal relationships between stages. It identifies quality transfer features by tracking the changing patterns of quality indicators between stages, establishing a quality transfer matrix to quantify the impact of upstream stage parameter changes on downstream stage quality indicators. Finally, the algorithm performs pairwise analysis of key parameters between all stage pairs to form an inter-stage correlation feature matrix. Step 3: Construct a quality assessment model based on the aforementioned quality characteristics. The quality assessment model includes a process quality assessment sub-model and a process correlation assessment sub-model. The process correlation assessment sub-model is constructed using graph neural network technology and is used to assess the quality correlation status between processes. Step 4: Use the quality assessment model to monitor the real-time production data of each link in the industrial chain in real time, and obtain the quality monitoring results; Step 5: Determine whether the quality monitoring results meet the preset quality threshold conditions; Step 6: If the quality monitoring results do not meet the preset quality threshold conditions, the quality monitoring results are analyzed through a process association tracing algorithm to identify the abnormal quality process and the quality transfer path between processes, and to obtain the quality abnormality source tracing information. Step 7: Based on the quality anomaly tracing information, generate a quality improvement plan through a process collaborative optimization algorithm. The quality improvement plan includes intra-process optimization measures and inter-process collaborative optimization measures.

2. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model as described in claim 1, characterized in that, The process of obtaining historical production environment data for each link in the industrial chain includes the following steps: Step 11: Collect historical production environment data for the chip fabrication process through a multi-source sensor network, including temperature, humidity, air pressure, cleanliness, and process parameters; Step 12: Collect historical production environment data for the mass transfer process using automated testing equipment, including transfer accuracy, alignment error, yield, and environmental parameters; Step 13: Collect historical production environment data of the driver integration process through the integration test platform, including driver circuit parameters, integration yield and environmental conditions; Step 14: Collect historical production environment data of the packaging process through the packaging testing system, including packaging material performance, packaging process parameters and environmental conditions.

3. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model as described in claim 2, characterized in that, The process of extracting quality features from the historical production environment data includes the following steps: Step 21: Extract the intra-process features of the historical production environment data using a time-series feature extraction algorithm, including statistical features, trend features, and anomaly features; Step 22: Extract the inter-process correlation features of the historical production environment data through correlation analysis algorithms, including temporal correlation features, causal correlation features, and quality transfer features; Step 23: Integrate the intra-stage features and inter-stage correlation features using a feature fusion algorithm to form a complete quality feature set.

4. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model according to claim 3, characterized in that, The process of constructing a quality assessment model based on the quality characteristics includes the following steps: Step 31: Construct a process quality assessment sub-model using deep learning technology to evaluate the quality status within each process. Step 32: Construct a sub-model for evaluating the correlation between links using graph neural network technology to assess the quality correlation status between links; Step 33: Integrate the process quality assessment sub-model and the process correlation assessment sub-model using an ensemble learning method to form a unified quality assessment model.

5. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model according to claim 4, characterized in that, The real-time monitoring of production data at each stage of the industrial chain using the quality assessment model includes the following steps: Step 41: Acquire real-time production data of each link in the industrial chain through the data acquisition interface and perform preprocessing; Step 42: Extract real-time features that match the quality features from the preprocessed real-time production data using the feature extraction module; Step 43: Input the real-time features into the quality assessment model through the model inference engine to generate real-time quality monitoring results.

6. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model according to claim 5, characterized in that, The process of determining whether the quality monitoring result meets the preset quality threshold condition includes the following steps: Step 51: Set and maintain the preset quality threshold conditions through the threshold management module, including the process quality threshold and the process-related quality threshold; Step 52: Compare the quality monitoring results with the preset quality threshold conditions using a threshold comparison algorithm to generate a comparison result; Step 53: The decision logic module determines whether the quality monitoring result meets the preset quality threshold condition based on the comparison result, and generates a judgment result.

7. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model as described in claim 6, characterized in that, If the quality monitoring result does not meet the preset quality threshold condition, the analysis of the quality monitoring result through the process correlation traceability algorithm includes the following steps: Step 61: Analyze the quality monitoring results using the anomaly localization module to identify the quality anomaly process; Step 62: Identify the quality transfer path between the links based on the output of the link association evaluation sub-model using the association analysis module; Step 63: Integrate the information on the quality anomaly links and the quality transfer path between the links through the root cause analysis module to generate the quality anomaly tracing information.

8. The method for monitoring the quality status of the Mini / MicroLED industry chain based on a large model according to claim 7, characterized in that, The step of generating a quality improvement plan based on the quality anomaly tracing information and using a process collaborative optimization algorithm includes the following steps: Step 71: Based on the quality anomaly tracing information, the optimization target setting module determines the optimization target and constraints for quality improvement; Step 72: Generate intra-process optimization measures for the quality anomaly process using the intra-process optimization module; Step 73: Generate inter-process collaborative optimization measures based on the inter-process quality transfer path through the inter-process collaboration module; Step 74: Integrate the intra-stage optimization measures and inter-stage collaborative optimization measures into a complete quality improvement plan through the solution integration module.

9. A Mini / MicroLED supply chain quality monitoring system based on a large model, used to implement the Mini / MicroLED supply chain quality monitoring method based on a large model as described in any one of claims 1-8, characterized in that, It includes a data acquisition module, a quality feature extraction module, a model building module, a real-time monitoring module, a threshold judgment module, an anomaly tracing module, and a quality optimization module; the modules are connected to each other electrically. The data acquisition module acquires historical production environment data for each link in the industrial chain, including chip fabrication, mass transfer, driver integration, and packaging, and sends the historical production environment data to the quality feature extraction module. The quality feature extraction module extracts quality features from the historical production environment data to obtain the quality features of the historical production environment data. The quality features include intra-process features and inter-process correlation features. The quality features are then sent to the model building module. The model building module constructs a quality assessment model based on the quality characteristics. The quality assessment model includes a process quality assessment sub-model and a process correlation assessment sub-model, and sends the quality assessment model to the real-time monitoring module. The real-time monitoring module uses the quality assessment model to monitor the real-time production data of each link in the industrial chain, obtains the quality monitoring results, and sends the quality monitoring results to the threshold judgment module. The threshold judgment module determines whether the quality monitoring result meets the preset quality threshold condition. If it does, the process ends; otherwise, the quality monitoring result is sent to the anomaly tracing module. The anomaly tracing module analyzes the quality monitoring results through a process association tracing algorithm, identifies the quality anomaly process and the quality transfer path between processes, obtains the quality anomaly tracing information, and sends the quality anomaly tracing information to the quality optimization module. The quality optimization module generates a quality improvement plan based on the quality anomaly tracing information using a process collaborative optimization algorithm. The quality improvement plan includes intra-process optimization measures and inter-process collaborative optimization measures.

Images