Mini LED display screen optical packaging method and system based on AI algorithm optimization

Abstract

The invention provides a Mini LED display screen optical packaging method and system based on AI algorithm optimization. Belongs to the technical field of display screens. The method comprises the following steps: acquiring various data in a MiniLED chip packaging process in real time; historical packaging data of the MiniLED chip are obtained, and an AI algorithm model is constructed based on a machine learning algorithm; in the packaging process, the AI algorithm processing system optimizes packaging parameters in real time according to data collected in real time in combination with the trained AI algorithm model; based on a feedback control system, packaging parameters optimized by the AI algorithm are fed back to packaging equipment in real time; and detecting the packaged display screen in real time through a feedback control system. Different types of data are collected through multiple sensors, multi-dimensional capture of information is achieved through feature extraction and fusion, and the understanding and precision of the system on the optical packaging process are enhanced.

Description

Technical Field

[0001] This invention proposes an optical packaging method and system for MiniLED displays based on AI algorithm optimization, belonging to the field of display technology. Background Technology

[0002] With the continuous development of display technology, MiniLED displays, with their advantages of high brightness, high contrast, and high color saturation, have been widely used in high-end televisions, commercial advertising, and sporting events. However, existing MiniLED display optical packaging methods still have some shortcomings. The stability of traditional wireless transmission systems is generally poor, especially in scenarios where multiple data sources simultaneously collect and transmit data, easily leading to signal interference, data loss, or delays. Furthermore, traditional technologies typically lack load-balancing transmission schemes, making them prone to data transmission delays or failures when network congestion or device overload occurs, affecting the efficiency and stability of the entire system. Summary of the Invention

[0003] This invention provides an AI algorithm-optimized optical packaging method and system for MiniLED displays to solve the problems mentioned in the background section.

[0004] This invention proposes an optical packaging method for MiniLED displays based on AI algorithm optimization, the method comprising:

[0005] S1. Real-time acquisition of various data during the MiniLED chip packaging process;

[0006] S2. Obtain historical packaging data of MiniLED chips and build an AI algorithm model based on machine learning algorithms;

[0007] S3. During the packaging process, the AI ​​algorithm processing system optimizes the packaging parameters in real time based on the data collected in real time and the trained AI algorithm model.

[0008] S4. Based on the feedback control system, the packaging parameters optimized by the AI ​​algorithm are fed back to the packaging equipment in real time;

[0009] S5. Real-time monitoring of the packaged display screen is performed through a feedback control system.

[0010] The present invention proposes an optical packaging system for a MiniLED display screen for AI algorithm optimization, comprising a memory, a processor, and a computer program stored in the memory and executable on the memory. The processor executes the program to implement the MiniLED display screen optical packaging method based on AI algorithm optimization as described above.

[0011] The beneficial effects of this invention are as follows: By collecting different types of data through multiple sensors and achieving multi-dimensional information capture through feature extraction and fusion, the system's understanding and accuracy of the optical packaging process are enhanced; based on timestamp and spatial alignment technology, the accuracy and consistency of multi-source data are ensured, improving data integration efficiency and reducing errors caused by data inconsistency or spatiotemporal misalignment; when processing multi-source data, compression algorithms are used for effective data compression, saving storage resources and accelerating data transmission speed; encryption algorithms ensure data security, guaranteeing the security and integrity of data during transmission and avoiding potential data leakage risks; and the establishment of a model for temperature and wireless transmission performance enables prediction and adjustment of wireless signal strength under different temperature conditions. The transmission parameters ensure the stability and reliability of wireless transmission; the data transmission path is dynamically adjusted through a load balancing algorithm, which can respond to network load fluctuations in real time and effectively reduce transmission delays caused by signal interference or congestion; based on historical encapsulation data, machine learning and deep learning algorithms are used for encapsulation optimization, which can automatically learn from a large amount of data and extract key factors affecting the display effect during the encapsulation process; the AI ​​model can be continuously optimized based on training and verification results, thereby improving encapsulation accuracy and further enhancing the performance and stability of the MiniLED display; through in-depth analysis of optical camera images, spectral analyzer data, and displacement sensor data, every detail of the encapsulation process can be precisely controlled, thereby optimizing the display effect. Attached Figure Description

[0012] Figure 1 This is a diagram illustrating the steps of the method described in this invention. Detailed Implementation

[0013] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0014] One embodiment of the present invention, such as Figure 1 As shown, an optical packaging method for a MiniLED display screen based on AI algorithm optimization is disclosed. The method includes:

[0015] S1. Acquire various data in real time during the MiniLED chip packaging process through a multi-source data acquisition system, and transmit these data to the AI ​​algorithm processing system;

[0016] S2. Obtain historical packaging data of MiniLED chips, build an AI algorithm model based on machine learning algorithms; input historical packaging data, output display performance data of the display screen;

[0017] S3. During the packaging process, the AI ​​algorithm processing system optimizes the packaging parameters in real time based on the data collected in real time and the trained AI algorithm model.

[0018] S4. Based on the feedback control system, the packaging parameters optimized by the AI ​​algorithm are fed back to the packaging equipment in real time, and the packaging equipment automatically adjusts the packaging process according to the feedback parameters.

[0019] S5. The packaged display screen is monitored in real time through the feedback control system. If the display performance is found to be substandard, the packaging parameters are adjusted and secondary optimization is performed.

[0020] The working principle of the above technical solution is as follows: A multi-source data acquisition system acquires various data in real time during the MiniLED chip packaging process, including packaging position, angle, spectral characteristics, and displacement information. An optical camera acquires the packaging position and angle in real time, a spectral analyzer provides spectral characteristic data, and a displacement sensor monitors displacement changes during the packaging process. Based on historical packaging data (including packaging parameters, spectral characteristics, and display performance data), a machine learning algorithm is used to train an AI model. This model can predict the display performance (such as brightness uniformity, color accuracy, and contrast) based on the correlation between different packaging process parameters and display performance. The AI ​​algorithm processing system optimizes the parameters during the packaging process in real time based on the real-time acquired data and the trained AI algorithm model. For example, if the optical camera detects a deviation in the packaging position or angle, the AI ​​system calculates the optimal adjustment parameters and feeds them back to the packaging equipment for adjustment. Meanwhile, the spectral data measured by the spectrometer can be used to optimize the formulation and process of the packaging material, thereby improving the luminous efficacy and color reproduction capabilities of the MiniLED chip. Based on the feedback control system, the AI-optimized packaging parameters are fed back to the packaging equipment in real time. The equipment automatically adjusts itself based on this feedback information to ensure that the packaging process is always in an optimal state. The finished display screen is then monitored in real time by the feedback control system. If the display performance is found to be non-compliant with standards, the system will automatically adjust the packaging parameters and perform secondary optimization until the display performance reaches the predetermined standard.

[0021] The effects of the above technical solution are as follows: Real-time optimization of the packaging process using AI algorithms improves the display performance indicators of MiniLED displays, such as brightness uniformity, color accuracy, and contrast, ensuring higher quality and more stable display effects in the final display. Real-time monitoring and optimization of various parameters during the packaging process reduces the chance of deviations, thereby lowering the proportion of defective displays and improving production efficiency and yield. AI algorithms can accurately calculate the optimal adjustment values ​​for parameters such as packaging position, angle, and spectral characteristics, making the packaging process more refined and improving the light efficiency and color reproduction capabilities of the display. The packaging equipment can automatically adjust in real time according to the AI-optimized parameters, reducing manual intervention and improving the automation and intelligence of the packaging process. Due to the real-time optimization through AI algorithms, the number of defective displays is reduced. The AI-optimized packaging process significantly improves the production and rework of MiniLED displays, reducing production costs and increasing efficiency. By optimizing the formulation and process of packaging materials, the luminous efficacy and color reproduction capabilities of MiniLED chips are significantly enhanced, resulting in better performance in high brightness and wide color gamut. Based on a real-time feedback control system, the packaging equipment can respond quickly and make precise adjustments according to AI-optimized parameters, ensuring accurate control of every detail in the packaging process, thereby improving the precision and stability of the production process. Through continuous secondary optimization and feedback adjustments, the packaging process of each batch of displays is ensured to be stable, reducing the impact of human error and equipment failure on product quality. With the continuous accumulation of training data, the AI ​​model can continuously adjust and optimize its parameters based on historical data, better adapting to the needs of different packaging processes and further improving the production quality of displays.

[0022] In one embodiment of the present invention, S1 includes:

[0023] S11. Select and install a multi-source data acquisition system;

[0024] S12. The multi-source data acquisition system installed collects relevant data respectively;

[0025] S13. Process the data collected by the multi-source data acquisition system and transmit it to the AI ​​algorithm processing system via wireless transmission.

[0026] The working principle of the above technical solution is as follows: First, multiple high-precision measurement devices are selected and installed, specifically including optical cameras, spectrometers, and displacement sensors. The selection and installation of these devices ensure that multi-dimensional data related to MiniLED chip packaging can be acquired efficiently and accurately. The optical camera is used to acquire images of the MiniLED chip packaging position and angle. Through high-resolution and high-frame-rate imaging equipment, it is ensured that each frame of the image can be accurately recorded to analyze any minute changes during the packaging process. The spectrometer is responsible for measuring the spectral characteristics of the chip to ensure that the packaged chip meets the standards in terms of optical performance, especially in terms of accuracy in parameters such as emission wavelength range and full width at half maximum (FWHM). The displacement sensor is used to monitor the displacement changes of the chip in real time during the packaging process, accurate to the micrometer level, to ensure precise position control of the chip during the packaging process and avoid errors. The acquisition of optical data, spectral data, and displacement data is carried out simultaneously. An optical camera acquires images in real time at a set frame rate, ensuring accurate capture of visual information at every moment during the packaging process. A spectral analyzer analyzes the chip's spectrum immediately after packaging to ensure its luminescence characteristics meet design requirements. A displacement sensor continuously monitors chip displacement changes, providing real-time feedback on the chip's movement during packaging. This system allows for the simultaneous acquisition of three different types of data, providing comprehensive information for subsequent analysis. All acquired data is processed in real time through a multi-source data acquisition system. This data includes not only optical images and spectral data but also displacement measurement data, forming a complementary multi-dimensional information set. The processed data is then wirelessly transmitted to an AI algorithm processing system. Wireless transmission ensures the system's flexibility and efficiency, avoiding the complexity of physical wiring, and also makes data processing more efficient and real-time. The AI ​​algorithm system receives and processes this data, typically using technologies such as image recognition, spectral analysis, and displacement data fusion to automatically analyze and judge the quality of chip packaging and whether there are any deviations or defects. Finally, through the processing of AI algorithms, the system can output high-precision packaging quality assessment results, ensuring that the packaging of each chip meets design requirements and adjusting deviations in the packaging process in a timely manner.

[0027] The effects of the above technical solution are as follows: By selecting high-precision optical cameras, spectrometers, and displacement sensors, the packaging process of MiniLED chips can be effectively monitored, ensuring high-precision control of parameters such as packaging position, angle, and spectral characteristics. Especially when chip size and packaging precision requirements are very stringent, this solution helps ensure product quality and stability. Employing a multi-source data acquisition system, combining optical, spectral, and displacement sensing technologies, enables real-time data acquisition. This provides immediate feedback to each stage of the packaging process, facilitating timely detection and correction of deviations, and reducing the generation of defective products. Due to the high frame rate and high precision of the acquisition equipment, the speed and accuracy of data acquisition can be greatly improved, especially on fast production lines, enabling the acquisition of large amounts of high-quality measurement data in a short time to meet the needs of industrial production. Synchronous data acquisition with displacement sensors, optical cameras, and spectrometers ensures accurate data exchange between different sources, improving the effectiveness of subsequent data processing and analysis. Synchronous data acquisition makes the monitoring of various parameter changes during the packaging process more consistent, thereby optimizing production control. Wireless transmission technology transmits data to the AI ​​algorithm processing system, enabling real-time data analysis and optimization, reducing the pressure of manual intervention and improving automation levels. Meanwhile, AI algorithms can learn and adjust themselves based on collected data, further optimizing the packaging process and improving production efficiency. Through high-precision spectral analysis and displacement measurement, potential optical characteristic deviations and displacement deviations during the packaging process can be detected and resolved promptly, ensuring the optical performance, reliability, and consistency of MiniLED chips. Because the equipment is adjustable and the system can adapt to different production requirements, the flexible deployment of optical cameras, spectral analyzers, and displacement sensors can meet the needs of different production lines and processes, enhancing the system's versatility. Real-time monitoring and precise control can reduce the scrap rate caused by poor packaging, thereby lowering rework and repair costs. Automated processing systems not only improve efficiency but also reduce labor costs by minimizing manual intervention.

[0028] In one embodiment of the present invention, S13 includes:

[0029] S131. Preprocess the data;

[0030] S132. By establishing a spatial coordinate system, spatial alignment of data collected by different sensors is performed based on feature point matching, and the data are represented in the same coordinate system.

[0031] S133. Based on timestamp technology, a timestamp is added to each data point during the data acquisition process, and the data from different sensors are aligned using the timestamps;

[0032] S134. Extract different features from data from different sources, and fuse the features from different sources to form a comprehensive feature vector;

[0033] S135. Different compression algorithms are used to compress different types of data;

[0034] S136. Before data transmission, the data is encrypted using an encryption algorithm and key to convert the original data into ciphertext data; at the receiving end, the ciphertext data is decrypted using the corresponding decryption algorithm and key to restore the original data.

[0035] S137. Encrypted data is transmitted through a multi-channel transmission protocol. During data transmission, the data transmission path is adjusted in real time through a load balancing algorithm, and the temperature of the transmission environment is monitored in real time. The wireless transmission parameters are adjusted according to the temperature changes.

[0036] S138. Based on data analysis, establish a model between temperature and wireless transmission performance, predict transmission performance at different temperatures, and take corresponding measures to optimize in advance.

[0037] The working principle of the above technical solution is as follows: data preprocessing, the preprocessing including:

[0038] Image denoising: For image data acquired by optical cameras, algorithms such as median filtering and Gaussian filtering are used to remove noise from the image. Median filtering can effectively remove salt-and-pepper noise, while Gaussian filtering has a good suppression effect on Gaussian noise. By selecting appropriate filtering parameters, the edge and detail information of the image can be preserved as much as possible while removing noise.

[0039] Spectral data calibration: Since the spectrometer may be affected by environmental factors (such as temperature and humidity) during the measurement process, the measurement data may be biased. Therefore, it is necessary to calibrate the spectral data. Regular calibration is performed using a standard light source, a calibration curve is established, and the measurement data is corrected according to the calibration curve to ensure the accuracy of the spectral data.

[0040] Displacement data filtering: The displacement data collected by the displacement sensor may be affected by factors such as mechanical vibration and electromagnetic interference, resulting in high-frequency noise. A low-pass filter is used to filter the displacement data to remove high-frequency noise and extract effective displacement change information.

[0041] Feature point matching aligns data from different sensors to a unified spatial coordinate system, ensuring clear relationships between sensor data. Timestamp technology marks the time of each data point during acquisition, ensuring precise temporal alignment and avoiding errors caused by time differences. Important features are extracted from different data sources, such as edge features from images and wavelength features from spectral data, and then fused to form a comprehensive feature vector. Different compression algorithms are used for different types of data, including:

[0042] Image compression: Optical cameras capture large amounts of image data. To reduce bandwidth requirements for data transmission, image compression algorithms such as JPEG and PNG are used to compress the images. While maintaining image quality, an appropriate compression ratio is selected to reduce image data storage space and transmission time.

[0043] Spectral data compression: Spectral data is typically high-dimensional and sparsity. Compression algorithms such as Principal Component Analysis (PCA) and wavelet transform are used to compress spectral data. Dimensionality reduction removes redundant information from the spectral data, retains the main spectral features, and reduces the data volume.

[0044] Displacement data compression: Displacement data often exhibits certain regularities. Compression algorithms such as differential coding and predictive coding are used to compress displacement data. By recording the amount of change in displacement data instead of the original data, the amount of data is reduced, and transmission efficiency is improved.

[0045] Before data transmission, the data is encrypted using an encryption algorithm and key, converting the original data into ciphertext. At the receiving end, the ciphertext is decrypted using a corresponding decryption algorithm and key to recover the original data. The encrypted data is transmitted via a multi-channel transmission protocol. During data transmission, a load balancing algorithm adjusts the data transmission path in real time, and the ambient temperature is monitored. Wireless transmission parameters, such as transmit power and modulation scheme, are adjusted based on temperature changes. When the temperature rises, wireless signal attenuation increases; in this case, the transmit power can be appropriately increased to enhance signal strength. When the temperature drops, the transmit power can be appropriately reduced to save energy. Simultaneously, a suitable modulation scheme is selected based on temperature changes; in high-temperature environments, a modulation scheme with strong anti-interference capabilities is chosen to improve data transmission reliability. Based on data analysis, a model is established between temperature and wireless transmission performance to predict transmission performance at different temperatures, allowing for proactive optimization measures. For example, in high-temperature environments, the number of signal repeaters can be increased to expand signal coverage; in low-temperature environments, the antenna direction and angle can be adjusted to improve signal reception quality.

[0046] The effects of the above technical solutions are as follows: By effectively removing noise from images through methods such as median filtering and Gaussian filtering, not only are image details and edges preserved, but the accuracy of subsequent image analysis is also improved. Noise reduction processing enhances the visual quality of optical images in practical applications, reducing erroneous analyses caused by noise. Calibration of spectral data ensures measurement accuracy and reduces deviations caused by environmental factors (such as temperature and humidity changes). Regular calibration using a standard light source improves the reliability of spectral data, providing more accurate data support for scientific research or industrial testing. Low-pass filters are used to filter displacement data, removing high-frequency noise caused by mechanical vibration, electromagnetic interference, and other factors, extracting more accurate displacement change information. This helps improve the monitoring accuracy of machinery and equipment. By establishing a spatial coordinate system and matching feature points, spatial alignment of data from different sensors is achieved, allowing for the comprehensive representation of multidimensional data from various sensors within the same coordinate system. Data fusion can improve the overall performance of a system and reduce the potential errors of a single sensor. Aligning data from different sensors using timestamp technology ensures the synchronization of various data types over time, effectively improving the accuracy and timeliness of data processing. Extracting useful features from various sensor data and fusing them into a comprehensive feature vector provides more comprehensive information for subsequent data analysis, machine learning modeling, and other tasks, helping to improve the accuracy of decision support systems. For different types of data (images, spectra, displacement), appropriate compression algorithms can be used to reduce storage space and transmission bandwidth requirements. Compressing spectral data using methods such as PCA and wavelet transform, and compressing displacement data using methods such as differential coding, can effectively reduce storage and transmission pressure and improve the overall efficiency of the system. Encryption algorithms are used during data transmission to ensure data security and prevent data leakage or tampering. This measure ensures data confidentiality and integrity, adapting to application scenarios with high data security requirements. Through multi-channel transmission protocols and load balancing algorithms, it adjusts data transmission paths in real time and regulates wireless transmission parameters (such as transmit power and modulation methods) based on temperature changes, improving both data transmission efficiency and reliability. Since wireless signal attenuation increases with temperature, adjusting transmit power according to temperature also reduces energy waste. By establishing a model between temperature and wireless transmission performance, and through real-time prediction and optimization measures (such as adding repeaters and adjusting antenna direction), it ensures optimal wireless transmission performance under different temperature conditions, thereby guaranteeing data transmission stability and efficiency.

[0047] In one embodiment of the present invention, S132 includes:

[0048] Construct a global three-dimensional Cartesian coordinate system; and establish a local coordinate system for each sensor;

[0049] Based on the transformation relationship between the global coordinate system and various local coordinate systems, a coordinate system transformation model is established;

[0050] Select feature point types for different types of data;

[0051] Feature points were extracted using a multi-scale analysis method;

[0052] A hierarchical matching strategy is adopted: coarse matching of feature points is performed at low resolution to quickly filter out possible matching feature point pairs; fine matching is performed on the coarse matching results at high resolution.

[0053] Hardware acceleration technology is used to parallelize the feature point extraction and matching algorithms, and computing resources are dynamically allocated according to the real-time nature of data acquisition and the priority of computing tasks.

[0054] A compensation model is established between temperature and sensor performance to compensate for temperature in data collected by different sensors; the data is corrected based on the compensation model by real-time monitoring of ambient temperature.

[0055] During the data fusion process, the impact of temperature on data from different sensors is considered, and the data fusion algorithm is optimized for different temperatures.

[0056] The working principle of the above technical solution is as follows: Based on the installation positions and initial calibration parameters of devices such as optical cameras, spectrometers, and displacement sensors, a global three-dimensional Cartesian coordinate system is constructed. This coordinate system takes a fixed point (such as the center point of the equipment mounting platform) as its origin, and the X, Y, and Z axes correspond to the physical installation direction of the equipment, providing a reference for the spatial positioning of subsequent data from different sensors. A local coordinate system is established for each sensor; for example, for an optical camera, a local camera coordinate system is established with the camera's optical center as its origin based on its imaging principle and lens parameters; for a spectrometer, a local spectral coordinate system is established based on its spectral acquisition window and optical path design. The establishment of local coordinate systems helps to more accurately describe the spatial characteristics of sensor data. Based on the transformation relationship between the global coordinate system and various local coordinate systems, a coordinate system transformation model is established. This model should consider factors such as sensor installation errors and mechanical deformation, and uses rigid body transformations (including translation and rotation) to describe the transformation between different coordinate systems. Through an accurate coordinate system transformation model, data collected by different sensors are accurately transformed to the global coordinate system. For different types of data, feature point types are selected. For optical images, feature point detection algorithms with rotation invariance and scale invariance (such as SIFT, SURF, etc.) are used to extract features such as corner points and edge points in the image. For spectral data, features are extracted based on the shape of the spectral curve. For spectral data, feature points such as spectral peaks, valleys, and inflection points are extracted. For displacement data, features such as extreme points and turning points are extracted based on the trend and periodicity of displacement changes. Considering the differences in resolution and scale of data from different sensors, a multi-scale analysis method is used to extract feature points. In optical images, feature points are detected at different scales by constructing an image pyramid to adapt to targets of different sizes and shapes. In spectral data, the spectral curves are smoothed and filtered at different scales to extract feature points at different scales. In displacement data, feature points are extracted by analyzing the displacement data at different time scales based on the temporal resolution and sampling frequency.A hierarchical matching strategy is adopted. First, coarse matching of feature points is performed at low resolution to quickly filter out potentially matching feature point pairs. Then, fine matching is performed on the coarse matching results at high resolution to improve matching accuracy. This hierarchical matching strategy reduces computational overhead and increases matching speed while maintaining matching accuracy. Hardware acceleration technology is used to parallelize the feature point extraction and matching algorithms, dynamically allocating computing resources based on the real-time nature of data acquisition and the priority of computational tasks. For example, when the data acquisition frequency is high and real-time requirements are high, computing resources are prioritized for feature point extraction and matching; when the computational task is lighter, some resources can be allocated to feature point extraction and matching. Allocate computing resources to other data processing tasks to improve the overall efficiency of the system; establish a compensation model between temperature and sensor performance to perform temperature compensation on data collected by different sensors; correct the data according to the compensation model by monitoring the ambient temperature in real time; consider the impact of temperature on data from different sensors during the data fusion process and optimize the data fusion algorithm for different temperatures; for example, when the temperature is high, increase the weight of optical image data and decrease the weight of spectral data and displacement data, because optical images are relatively more stable at high temperatures; when the temperature is low, appropriately adjust the weights of each data source to improve the robustness of data fusion.

[0057] The effects of the above technical solution are as follows: By establishing a transformation model between the global and local coordinate systems, accurate spatial positioning and fusion of data from different types of sensors can be achieved. Precise coordinate system transformation can eliminate positioning deviations caused by factors such as equipment installation errors and mechanical deformation, thereby improving the overall measurement accuracy of the system. For different types of data, feature point extraction and multi-scale analysis methods can capture key features in the data at different scales. By introducing algorithms such as SIFT and SURF, rotation- and scale-invariant features can be effectively extracted, improving the robustness of matching. A hierarchical matching strategy, combined with low-resolution and high-resolution feature point matching, can significantly improve data processing efficiency while ensuring accuracy. Furthermore, parallel processing achieved through hardware acceleration technology can dynamically allocate computing resources according to the needs of real-time data acquisition, improving the system's response speed and overall performance. By monitoring the ambient temperature in real time and correcting sensor data according to a temperature compensation model, the impact of temperature changes on the performance of different sensors can be effectively eliminated, ensuring that the system can still provide reliable measurement results under different temperature conditions. This temperature compensation strategy improves the system's adaptability and robustness. By considering the impact of temperature on different sensor data, the data fusion algorithm is optimized. Adjusting the weights of various data sources at different temperatures allows for adaptive adjustments to the system's output based on environmental changes, resulting in a more stable and accurate data fusion process. Introducing temperature compensation not only addresses real-time performance and data overhead issues but also further improves data accuracy, consistency, and fusion precision. By establishing a compensation model between temperature and sensor performance, the data processing process is optimized, enhancing the system's robustness and reliability. Furthermore, temperature compensation effectively reduces computational redundancy and optimizes computational resource allocation, enabling the system to operate stably and efficiently under varying temperature conditions.

[0058] In one embodiment of the present invention, S133 includes:

[0059] The data acquisition frequencies of different light sensors are analyzed; timestamps are added at the sensor hardware level; and the hardware clock is calibrated by synchronizing with a high-precision time source.

[0060] Based on a multi-sensor time synchronization protocol, the timestamps of different sensors are aligned to the same time base; for data collected at different frequencies, an interpolation algorithm is used for time alignment.

[0061] Machine learning algorithms are used to model the time alignment problem of different data types; the time relationship between different data types is learned through training datasets to build a time alignment model; new data is input into the model and time alignment is performed automatically.

[0062] Data is processed in layers according to the characteristics and processing requirements of different data types;

[0063] A compensation model is established between light intensity and the imaging quality of the optical camera and the measurement results of the spectral analyzer; the optical image data and spectral data are corrected according to the compensation model by real-time monitoring of light intensity.

[0064] Based on the time alignment accuracy evaluation index system, the time alignment accuracy evaluation index is calculated by comparing the aligned data with the real time, thereby quantifying the effect of time alignment.

[0065] The working principle of the above technical solution is as follows: Analysis is performed on the data acquisition frequencies of different sensors, such as optical cameras, spectrometers, and displacement sensors. Optical cameras may acquire image data at high frame rates (e.g., tens or even hundreds of frames per second), while spectrometers have relatively low acquisition frequencies (e.g., a few times per second), and the acquisition frequency of displacement sensors depends on their specific application scenario. Understanding these differences is fundamental for time alignment; and the delay time from data generation to acquisition by the acquisition system is analyzed. For example, optical cameras may have image sensor readout delays and data transmission delays; the spectral acquisition and data processing of spectrometers also introduce certain delays; and the signal processing and transmission of displacement sensors also involve delays. Accurate measurement of these delay times is crucial for precise time alignment. This involves unifying the timestamp format of all sensors to a single standard format; implementing timestamp addition at the sensor hardware level; embedding the current timestamp into the image data simultaneously with the image sensor's output data for optical cameras; adding timestamp generation modules to the data acquisition circuits of spectral analyzers and displacement sensors to ensure that timestamps are generated simultaneously with the data for spectral analysis; calibrating the hardware clock by synchronizing with high-precision time sources (such as GPS time, atomic clocks, etc.); ensuring that timestamps from different sensors are on the same time base based on a multi-sensor time synchronization protocol; using interpolation algorithms for time alignment of data acquired at different frequencies; for example, estimating the corresponding optical image data at the time point of the spectral data using linear interpolation, spline interpolation, etc., to achieve time alignment between the two types of data for high-frequency optical image data and low-frequency spectral data; and using feature matching algorithms to time-align the feature differences between different data types; for example, extracting edge features of the chip in the optical image and specific spectral peak features in the spectral data, and using feature matching algorithms to find the time points of corresponding features in different data to achieve time alignment. This paper utilizes machine learning algorithms to model the time alignment problem for different data types. It learns the temporal relationships between different data types using a training dataset to establish a time alignment model. New data is then input into the model for automatic time alignment. Data is processed in layers according to the characteristics and processing requirements of different data types. For example, optical image data is divided into edge feature layers and texture feature layers; spectral data into absorption peak feature layers and emission peak feature layers; and displacement data into distance and angle change layers. Time alignment and subsequent processing are performed at different levels to reduce mutual interference between different data types. Processing parameters are dynamically adjusted to address the differences in processing for different data types. For example, in image processing, the parameters of median filtering and Gaussian filtering are dynamically adjusted based on the noise level and edge feature intensity of the optical image; in spectral data processing, the dimensionality reduction parameters of principal component analysis (PCA) are dynamically adjusted based on the sparsity and high dimensionality of the spectral data.A compensation model is established between light intensity and the imaging quality of the optical camera and the measurement results of the spectral analyzer. By monitoring the light intensity in real time, the optical image data and spectral data are corrected according to the compensation model. Based on the time alignment accuracy evaluation index system, including time deviation and time synchronization error, the time alignment accuracy evaluation index is calculated by comparing the aligned data with the real time, and the effect of time alignment is quantified.

[0066] The effects of the above technical solution are as follows: By analyzing the data acquisition frequency and latency of different sensors, the accuracy of time alignment can be precisely measured, ensuring that data can be accurately compared and analyzed under the synchronization of multiple sensors. This is particularly important for multi-sensor fusion systems; by converting the timestamp formats of different sensors into a unified standard format, the problem of time synchronization between different sensors can be effectively solved, providing a reliable time reference for subsequent data processing and analysis; by implementing the addition of timestamps at the hardware level and synchronizing with a high-precision time source, the accuracy of time synchronization can be significantly improved, providing strong support for the time alignment of sensor data; and by combining interpolation algorithms, feature matching algorithms, and machine learning algorithms, the time alignment problem of different types of sensor data can be flexibly handled. For example, interpolation methods for optical images and spectral data, and feature matching algorithms for aligning different data types, can achieve efficient time alignment in various application scenarios. By processing features of different data types hierarchically, interference between different data types can be avoided, and dynamic parameter adjustments based on the characteristics of different data types can make the data from different sensors more accurate and stable during processing. Establishing a compensation model between illumination intensity and optical images and spectral data can effectively reduce the impact of illumination changes on sensor data quality, thereby improving data accuracy and consistency. By establishing a time alignment accuracy evaluation index system, the effect of time alignment can be quantitatively evaluated, helping to continuously optimize time synchronization schemes and algorithms. By introducing an illumination intensity compensation model, not only is the impact of illumination changes on the quality of optical images and spectral data resolved, but time alignment accuracy, data consistency, and the accuracy of multi-sensor fusion are also improved. By correcting the interference of illumination intensity on data in real time, the system can more accurately synchronize time and correct data, enhancing the adaptability and stability of the entire system under different environmental conditions.

[0067] In one embodiment of the present invention, S134 includes:

[0068] The chip image captured by the optical camera is processed using an edge detection algorithm; the shape descriptor method is used to quantify the shape of the chip.

[0069] Wavelength and intensity analysis is performed on the spectral data acquired by the spectrometer; and the full width at half maximum (FWHM) and peak area of ​​the spectral peaks are calculated.

[0070] Based on the displacement data collected by the displacement sensor, the chip's moving distance and speed at different time points are calculated; the speed information is obtained by differentiating the displacement data using the differential method.

[0071] For chips with rotational motion, analyze their angular changes; use angle sensors or calculate angular changes through displacement data, such as using trigonometric relationships to calculate the angular increment based on displacement changes;

[0072] Analysis of cross-modal feature mismatch problem;

[0073] PCA is used to reduce the dimensionality of high-dimensional optical image features, extracting the main components and reducing the feature dimensionality.

[0074] A feature selection algorithm is used to select the most representative features from spectral and displacement features; a semantic mapping model is constructed to map features from different modalities to the same semantic space.

[0075] By using knowledge graphs, features of different modalities can be associated with existing knowledge;

[0076] A model relating pressure to different modal characteristics was established. Optical images, spectral data, and displacement data of the chip under different pressures were measured experimentally, and a mathematical model between pressure and characteristics was established using regression analysis. Based on the model, the extracted characteristics were corrected. When the pressure value during the measurement process was known, the model was used to compensate for the characteristics and eliminate the influence of pressure on the characteristics.

[0077] The working principle of the above technical solution is as follows: An optical camera acquires chip images, and edge detection algorithms are used to process the images to extract the chip's edge and shape features; shape descriptor methods (such as Hu moments, Zernike moments, etc.) are used to quantify the chip's shape features, enabling subsequent analysis and comparison; a spectrometer acquires the chip's spectral data. Through wavelength and intensity analysis, characteristic wavelengths (absorption peaks, emission peaks, etc.) and their corresponding intensity values ​​in the spectrum are identified. Furthermore, by calculating the peak's half-width and peak area, detailed spectral information is obtained. The half-width reflects the peak's width, and the peak area is related to the chip's material content. Numerical integration methods (such as the trapezoidal rule, Simpson's rule, etc.) are used to accurately calculate the peak area; a displacement sensor acquires the chip's displacement data, and after differential processing, the chip's moving speed and displacement changes can be obtained. For rotational motion, angle changes are further calculated, and angle increments are obtained through trigonometric functions or angle sensors; data from different modalities (optical images, spectra, displacement) exhibit differences in feature dimensions and semantics. Optical images typically possess high-dimensional and complex pixel information, while spectral features are relatively low-dimensional but contain physical and chemical information, and displacement features are relatively simple, usually reflecting parameters such as position and velocity. These differences complicate feature matching and fusion, especially the differences in semantic meaning, which increase the difficulty of data fusion. Furthermore, noise and uncertainty are also major factors leading to cross-modal feature mismatch. Optical images may be affected by ambient lighting and noise, while spectral data may be affected by instrument accuracy and environmental factors, and displacement data may also be subject to noise from mechanical vibration and electromagnetic interference. For high-dimensional optical image features, PCA dimensionality reduction is employed to extract the main components, reducing dimensionality while retaining most of the information. By calculating the contribution rate of eigenvalues, the most representative principal components are selected and retained. Feature selection algorithms are applied to spectral and displacement features to select the most representative features for further analysis. The establishment of a semantic mapping model helps map features from different modalities to a shared semantic space, improving the relationships and matching degrees between them. By establishing associations between different modal features and existing knowledge through knowledge graphs, and leveraging entities and relationships in the knowledge graphs, semantic interpretations are further provided for different modal features, thereby enhancing the correlation between cross-modal features. Wavelet denoising methods are used to remove noise from different modal data, improving the accuracy of feature extraction. By experimentally measuring chip data under different pressures, mathematical models between pressure and different modal features are established using regression analysis or neural networks. Once the relationship model between pressure and features is established, features can be corrected using the model when the pressure value is known, eliminating the influence of pressure on features and thus improving the accuracy of the data.

[0078] The effects of the above technical solution are as follows: By integrating optical images, spectral data, and displacement data, the chip's state can be comprehensively and accurately reflected, including its shape, chemical composition, physical properties, and motion state. Utilizing cross-modal feature fusion enables deeper chip analysis, compensating for the deficiencies of single-modal information and improving the accuracy of detection and diagnosis. Through PCA dimensionality reduction and feature selection algorithms (such as ReliefF and Information Gain algorithms), the dimensionality of the feature space can be reduced, thereby lowering computational complexity while retaining key information. This is particularly important for high-dimensional data (such as optical image features), helping to accelerate subsequent analysis processes and improve real-time performance. By establishing a semantic mapping model and utilizing knowledge graphs, the semantic differences between cross-modal features are resolved. This method maps features from different modalities to the same semantic space, allowing data from different modalities to be compared and analyzed within a unified framework, enhancing the correlation between data. Employing wavelet denoising methods effectively reduces noise from data from different modalities, improving the accuracy of feature extraction. This is particularly important for sensor data such as optical images, spectral data, and displacement data, which are easily affected by noise, helping to reduce errors and improve the reliability of analysis results. By establishing a relationship model between pressure and various modal features, chip features under different pressures can be corrected, eliminating the influence of environmental pressure on features. This compensation mechanism ensures the accuracy of feature data, making chip detection results more stable and reliable. This technical solution combines multiple data processing and analysis methods, such as shape descriptors, spectral peak analysis, displacement velocity calculation, and feature selection, and has strong automated analysis capabilities. Through a data-driven approach, manual intervention can be reduced, improving the efficiency and accuracy of chip detection and analysis processes. Introducing pressure compensation not only solves the problem of cross-modal feature mismatch but also effectively handles the inconsistencies in optical images, spectral data, and displacement data caused by pressure changes. By establishing a relationship model between pressure and features, the influence of pressure on features can be eliminated, improving the accuracy, stability, and consistency of data, and ensuring high-quality fusion of multimodal data under different pressure conditions.

[0079] In one embodiment of the present invention, S137 includes:

[0080] During data transmission, the load of each transmission channel is monitored in real time; the server load status is assessed; and the data transmission path is dynamically adjusted based on the load status, allocating data to channels with lower loads. The load status is assessed in the following ways:

[0081]

[0082] Among them, w cpu (t) represents the CPU weight coefficient at time t; N cpuThis represents the normalized value of CPU load; w MEM (t) represents the memory weight coefficient at time t; N MEM This represents the normalized value of memory load; w NET (t) represents the network weight coefficients at time t; N NET Let represent the normalized value of the network load; α(t) represents the dynamic adjustment factor at time t; ΔL(t) represents the rate of change of load at time t; L avg (t) represents the average load rate at time t;

[0083] Establish a temperature data acquisition system to collect data from temperature sensors regularly; transmit the temperature data to the data processing center using wired or wireless methods.

[0084] The wireless signal transmission power is adjusted in real time according to temperature changes;

[0085] Select the appropriate modulation method based on temperature changes; and determine different adjustment strategies based on dynamic parameter adjustment rules, according to the temperature change range and rate of change.

[0086] Collect wireless transmission performance data at different temperatures; preprocess the collected data and establish a model between temperature and wireless transmission performance based on machine learning algorithms;

[0087] Validate the trained model by evaluating its performance using data not used in training; optimize the model based on the validation results.

[0088] A dynamic adjustment mechanism is established to dynamically adjust the wireless transmission parameters based on the real-time prediction results of the model. At the same time, the adjusted transmission performance data is collected and fed back to the model for further optimization, forming a closed-loop optimization system.

[0089] The working principle of the above technical solution is as follows: During data transmission, the system monitors the load of each transmission channel in real time, collecting indicators such as server CPU utilization, memory utilization, and network bandwidth to assess the current load status. When the load is high, the system automatically adjusts the data transmission path, prioritizing channels with lower loads to ensure system stability and transmission efficiency. The system establishes a temperature data acquisition system, periodically collecting temperature sensor data via wired or wireless means and transmitting the temperature data to the data processing center. This temperature data is used for subsequent wireless signal adjustment to optimize transmission quality. By monitoring temperature changes, when the temperature rises, the attenuation of the wireless signal intensifies; therefore, the system automatically increases the transmission power to enhance signal strength. Optimal transmission power values ​​at different temperatures are obtained through experiments, and a mapping relationship between temperature and transmission power is established to achieve rapid dynamic adjustment. Temperature changes affect the transmission quality of wireless signals. In high-temperature environments, the system selects modulation methods with strong anti-interference capabilities (such as QPSK and 16QAM) to improve transmission reliability; in low-temperature environments, more efficient modulation methods (such as BPSK and QAM) are selected to improve transmission efficiency. The adjustment strategy is selected based on the rate of temperature change. A more aggressive strategy is adopted when the temperature rises rapidly in a short period of time, while a conservative strategy is used when the temperature changes gradually. The system sets different thresholds for temperature changes (e.g., 30°C). When the temperature exceeds these thresholds, corresponding adjustment measures are automatically triggered, such as increasing the transmission power or adjusting other wireless transmission parameters. The system collects wireless transmission performance data (such as transmission rate, bit error rate, signal strength, etc.) at different temperatures and uses machine learning algorithms to build a model between temperature and wireless transmission performance. This model is trained after data preprocessing to improve the prediction accuracy of wireless transmission performance. The trained model is then validated using data not used in the training process to evaluate its performance. If the verification results are unsatisfactory, the system will optimize the model, such as adjusting the structure or adding new feature variables to improve the model's generalization ability. Based on the model's prediction results, in high-temperature environments, the system will increase the number of repeaters, adjust the antenna direction and angle, etc., to expand the signal coverage and ensure signal reception quality. In low-temperature environments, the system will adjust parameters such as antenna and transmit power to optimize the signal transmission path and avoid unnecessary energy waste. Through real-time monitoring and model prediction, the system can dynamically adjust wireless transmission parameters and collect transmission performance data after adjustment to feed back to the model. This forms a closed-loop optimization system that continuously improves transmission efficiency and quality.

[0090] The above technical solution achieves the following effects: By monitoring the load of each transmission channel in real time and dynamically adjusting the data transmission path, data congestion can be avoided, improving data transmission efficiency. Simultaneously, adjusting the wireless signal transmission power and selecting appropriate modulation methods based on temperature changes effectively enhances transmission reliability under different environmental conditions. Dynamically adjusting wireless signal parameters according to temperature changes ensures good signal transmission performance even in high or low temperature environments. For example, increasing transmission power at high temperatures and selecting a high-efficiency modulation method at low temperatures addresses the impact of temperature variations. By employing machine learning algorithms to establish a model between temperature and wireless transmission performance, and training and optimizing the model, the system gains enhanced intelligence and adaptability. The system can automatically adjust the number of signal repeaters, antenna angle, and transmission parameters based on real-time prediction results, thus automatically responding to environmental changes; dynamically adjusting transmission power and signal parameters allows the system to maximize energy savings without affecting performance. For example, reducing transmission power in low-temperature environments and optimizing signal coverage by adding repeaters in high-temperature environments avoids energy waste. By optimizing signal transmission performance under different environments, the system can expand signal coverage and improve signal reception quality. Especially in high or low temperature conditions, flexibly adjusting repeaters and antenna orientation can effectively improve overall signal quality. Based on a dynamic adjustment mechanism using real-time monitoring data, the system can continuously optimize according to actual transmission performance, forming a closed-loop feedback system that continuously improves system stability and performance, ensuring that wireless transmission reaches its optimal state in various environments. The above formula not only considers real-time data on CPU, memory, and network load but also introduces dynamic adjustment factors and load change rates. This multi-dimensional evaluation can more comprehensively and accurately reflect the system's load status, avoiding evaluation bias caused by excessive focus on a single resource. By monitoring the load status of each transmission channel in real time, the formula can flexibly adjust the data transmission path. Depending on the load, data can be allocated to channels with lower loads, thereby improving system transmission efficiency, avoiding congestion, and reducing latency. The dynamic adjustment factor and load change rate in the formula can respond to load fluctuations in real time, ensuring that the system can adapt to different load changes. For example, when the load on a transmission channel suddenly increases, the system can switch paths in a timely manner to ensure that data transmission is not affected. By introducing weight coefficients for different resources (such as CPU, memory, and network), the formula can adjust the weight of the load importance of different resources according to the actual situation. This weighted mechanism can ensure that the system makes optimal decisions based on the current workload. The ratio of the load change rate to the average load in the formula can help the system detect load fluctuation trends and respond in advance.For example, if the load changes too rapidly, it may indicate that a resource is about to reach its bottleneck, and the system can automatically take preventative measures to avoid performance degradation. By comprehensively considering the average load and the rate of change, the formula can more accurately optimize load allocation, thereby improving data transmission efficiency and reducing resource waste. Especially in high-concurrency environments with large load fluctuations, the formula can ensure that the utilization rate of each resource is optimal. This formula enables the system to adaptively adjust under unstable or changing workload conditions. This intelligent adjustment reduces the need for manual intervention and improves the automation and stability of the system. By assessing the load of different transmission channels and dynamically adjusting the transmission path, the formula can effectively achieve load balancing. For complex system architectures, load balancing not only optimizes data flow but also avoids resource bottlenecks, improving the overall system availability and fault tolerance. In the process of combining temperature data and wireless transmission performance optimization, the formula provides more optimization dimensions for wireless communication. By adjusting the signal transmission power and modulation method according to temperature changes, the formula can effectively cope with wireless performance fluctuations under different environmental conditions, ensuring the reliability and stability of transmission.

[0091] In one embodiment of the present invention, S2 includes:

[0092] S21. Collect a large amount of historical packaging data for MiniLED chips; preprocess the collected historical packaging data;

[0093] S22. Based on the characteristics of the data and the needs of encapsulation optimization, select machine learning algorithms and use deep learning frameworks to build AI algorithm models;

[0094] S23. Divide the preprocessed historical data into a training set (70%), a validation set (15%), and a test set (15%). The training set is used for training the model, the validation set is used to tune the model's hyperparameters, and the test set is used to evaluate the model's performance.

[0095] S24. Use the training set to train the AI ​​algorithm model, and continuously adjust the model's weights and biases through the backpropagation algorithm so that the model's output gradually approaches the actual display performance data.

[0096] S25. Use the validation set to evaluate the trained model, adjust the model's hyperparameters based on the evaluation results, and repeat the training and evaluation process until the model's performance reaches its optimal level.

[0097] The working principle of the above technical solution is as follows: Collect a large amount of historical data related to MiniLED chip packaging, including various parameters of the packaging process (such as temperature, pressure, time, etc.), and information such as packaging position and angle under these parameters. Also, record performance data related to the display screen, such as brightness uniformity, color accuracy, and contrast. Preprocess this data, which may include data denoising, standardization, normalization, and handling missing values, to prepare for subsequent machine learning and deep learning model training. Based on the characteristics of the data, select a suitable deep learning model, specifically a model combining convolutional neural networks (CNN) and recurrent neural networks (RNN). CNNs are used to process image data, extracting features such as the shape and edges of MiniLED chips; while RNNs are used to process time-series data (such as spectral data and displacement data), capturing the temporal dependencies between data. The AI ​​algorithm model is constructed, containing multiple layers of convolutional layers (such as 3x3 or 5x5 convolutional kernels), pooling layers (such as max pooling), recurrent layers (such as LSTM layers), and fully connected layers. Finally, the processed information is mapped to display performance data through the output layer. Preprocessed historical data is divided into training, validation, and test sets for model training, hyperparameter tuning, and performance evaluation, respectively. The training set accounts for 70%, the validation set for 15%, and the test set for 15%. The model is trained using the training set, and the weights and biases are adjusted using the backpropagation algorithm. As training progresses, the model gradually adjusts to make its output more closely resemble actual display performance data. After training, the model's performance is evaluated using the validation set, with evaluation metrics including mean squared error (MSE) and mean absolute error (MAE). Based on the evaluation results, further adjust the model's hyperparameters, such as learning rate and batch size, to optimize model performance; repeatedly train and evaluate until the model's prediction accuracy reaches its optimal level, ensuring that the model can accurately predict the display performance data of the screen.

[0098] The effects of the above technical solution are as follows: By applying deep learning models, especially the combination of CNN and RNN, subtle changes in the packaging process can be captured more accurately. This allows for accurate modeling of the relationship between packaging parameters (such as temperature, pressure, and time) and display performance, providing a scientific basis for optimizing the packaging process. Furthermore, by training and applying AI models, automated packaging process optimization can be achieved. This data-driven approach can predict display performance based on input packaging parameters and historical data, reducing manual intervention, improving production efficiency, and ensuring that each MiniLED display meets the expected performance standards. By precisely controlling each parameter in the packaging process, problems such as uneven brightness, color difference, and contrast in MiniLED displays can be effectively reduced, thereby improving the quality and stability of the display. In particular, by controlling indicators such as brightness uniformity, color accuracy, and contrast, the consistency of screen performance can be ensured, meeting the quality requirements of high-end displays. Utilizing historical packaging data and deep learning models can reduce the development time of traditional trial-and-error methods and quickly find the optimal combination of packaging process parameters. This shortens the R&D and production cycle of MiniLED displays, thereby accelerating the market launch of new products. Deep learning models have strong generalization capabilities, can process large amounts of complex data, and make predictions based on data patterns. This technology allows companies to build accurate performance prediction models using historical data, enabling new product packaging processes to rely less on excessive experimentation and human experience, and instead on data-driven and scientific computing-based decision-making. During production, process parameters may vary due to different equipment, materials, and environmental factors. By adjusting the predictions of deep learning models in real time, the flexibility of the production line can be improved to adapt to production needs under various conditions, while ensuring efficient and stable production output. Combining CNNs and RNNs allows for the simultaneous processing of static data (such as chip appearance) and dynamic data (such as spectral changes and displacement), meaning the model can handle more complex process optimization problems, thus solving challenges that traditional methods cannot address. The advantages of deep learning are particularly pronounced when dealing with large-scale, high-dimensional data.

[0099] In one embodiment of the present invention, S3 includes:

[0100] S31. During the packaging process, the packaging position, angle, spectral characteristics and displacement data collected in real time by the multi-source data acquisition system are transmitted to the AI ​​algorithm processing system as input to the AI ​​algorithm model.

[0101] S32, the AI ​​algorithm processing system inputs real-time data into the trained AI algorithm model, the model performs inference calculations, and outputs the display performance prediction result under the current packaging parameters; wherein, the display performance is predicted in the following way:

[0102]

[0103] Where k(t) represents the saturation coefficient; ω i φ represents the feature weight of the i-th feature; i γ represents the normalized eigenvalue of the i-th feature; θ represents a fixed feature threshold used to adjust the center position of the Sigmoid function and control the output offset; j This represents the weight of the j-th feature combination; and Indicates the index range of feature combinations; θ combo (t) represents the threshold parameter that changes over time; C comp Indicates the compensation factor;

[0104] S33. Based on the prediction results, the AI ​​algorithm uses an optimization algorithm to optimize the encapsulation parameters;

[0105] S34. The AI ​​algorithm outputs the optimized packaging parameters in the form of digital signals and transmits them to the packaging equipment. The output parameters include specific values ​​such as packaging temperature, pressure, and time, as well as the adjustment range and trend of the parameters.

[0106] The working principle of the above technical solution is as follows: During the packaging process, the system acquires various data related to packaging in real time through a multi-source data acquisition system; including:

[0107] Package location data: indicates the position of the LED chip or other components during the packaging process.

[0108] Packaging angle data: Record the installation angle of the LED during the packaging process to ensure accurate positioning.

[0109] Spectral characteristic data: Capture spectral changes after packaging to evaluate the brightness and color accuracy of the display.

[0110] Displacement data: reflects minute displacements that occur during the packaging process, ensuring packaging accuracy.

[0111] This data is transmitted in real time to the AI ​​algorithm processing system as input for subsequent inference calculations. The real-time data collected is then input into a pre-trained AI algorithm model. The AI ​​model analyzes and infers from this data, calculating and outputting performance predictions for the display screen based on known packaging processes and performance relationships. These predictions include:

[0112] Brightness uniformity: This refers to whether the brightness distribution of the display screen is even.

[0113] Color accuracy: The trueness and accuracy of the colors displayed on the screen.

[0114] Contrast ratio: The degree of contrast between different brightness ranges of a display screen.

[0115] These predictions will be used to guide the optimization of subsequent packaging parameters.

[0116] Based on the predictions of the AI ​​model, the system adjusts key parameters in the packaging process using optimization algorithms (such as genetic algorithms or particle swarm optimization). For example, if brightness uniformity is insufficient, the AI ​​algorithm evaluates and adjusts parameters in the packaging process, such as temperature and pressure, to improve brightness uniformity; if color accuracy deviation is large, the algorithm adjusts spectral characteristics, angle, and temperature to optimize color performance. The optimization algorithm explores multiple possible parameter combinations to find the optimal packaging parameter configuration; the optimized packaging parameters are then transmitted to the packaging device in the form of digital signals. These parameters include:

[0117] Package temperature: Which temperature range should be set to ensure optimal performance?

[0118] Packaging pressure: The pressure value required during the packaging process.

[0119] Packaging time: The required time to ensure the accuracy and stability of the packaging process.

[0120] In addition to specific numerical values, the algorithm also provides the range and trend of parameter adjustments, helping the device to make real-time adjustments and thus automatically optimize the packaging process.

[0121] The effects of the above technical solution are as follows: Through AI algorithm inference, calculation, and optimization, parameters can be adjusted in real time during the packaging process to ensure that the brightness uniformity, color accuracy, and contrast of each display screen reach the optimal level, thereby greatly improving product quality consistency; using intelligent optimization algorithms, real-time optimization adjustments can be made during the packaging process, reducing the need for manual intervention, increasing the automation level of the packaging process, and improving production efficiency and precision. Because AI algorithms can accurately predict and optimize packaging parameters, rework or scrap caused by improper parameter settings is avoided, thus reducing resource and time waste in the production process; AI algorithms can flexibly adjust the packaging process based on real-time collected multi-source data to adapt to the needs of different production conditions, improving the system's adaptability and flexibility; by precisely controlling parameters such as temperature, pressure, and time during the packaging process, unnecessary debugging and testing are avoided, reducing production costs while improving production efficiency; by optimizing packaging parameters, higher quality display effects can be achieved, thereby enhancing the market competitiveness and customer satisfaction of the final product. This AI-based automated packaging optimization not only reduces manual intervention and human error but also enables more intelligent operation of the production line, promoting the progress of intelligent manufacturing and conforming to the future development trend of the manufacturing industry. The above formula, by inputting multi-dimensional data (encapsulation position, angle, spectral characteristics, displacement, etc.) and performing inference calculations through an AI algorithm model, can accurately predict the display performance of a screen. This accuracy provides strong support for optimizing display effects in practical applications, especially during the production process, where parameters can be adjusted in real time to ensure optimal display performance. The formula predicts display performance using multiple features (such as encapsulation position, angle, spectral characteristics, etc.), meaning it considers the comprehensive performance of the screen under different conditions. The weighting of different features and feature combinations by feature weights and combined features helps reveal the degree of influence of each parameter on display performance, ensuring the ability to comprehensively consider all influencing factors. The dynamic parameters in the formula enable the model to adapt to different time and environmental changes. This adaptability ensures that the model can cope with the effects of different production batches, changes in ambient temperature, etc., avoiding the limitations of static models. The prediction of display performance uses a non-linear mapping through the Sigmoid function, which helps to capture the complex relationship between display performance and various features. The Sigmoid function can smoothly map input features to display performance, avoiding complex changes that simple linear models may not be able to accurately describe. By optimizing algorithms and based on predicted display performance results, the AI ​​system can automatically adjust packaging parameters (such as temperature, pressure, and time). This automated optimization not only improves production efficiency but also reduces the risk of errors from human intervention, achieving intelligent production processes. Through accurate prediction and optimization, fluctuations in display performance can be significantly reduced, ensuring consistent display quality. Especially when abnormal fluctuations occur during production, the AI ​​algorithm can adjust parameters promptly to maintain stable display performance.The optimized packaging parameters are output as digital signals, providing detailed information on the adjustment range and trends of each parameter. This transparency offers engineers more intuitive feedback, facilitating further analysis and adjustments to the production process. Compensation factors in the formula help compensate for changes in the external environment or system deviations, ensuring that display performance remains within the expected range. Through this compensation mechanism, the system can self-regulate, reducing the negative impact of external factors on display effects. AI algorithms can monitor display performance in production in real time and automatically optimize parameters, significantly improving production efficiency and product quality. Through continuous system optimization and prediction, problems can be responded to more quickly during production, reducing scrap rates.

[0122] In one embodiment of the present invention, step S4 includes:

[0123] S41. Based on the feedback control system, the packaging parameters optimized by the AI ​​algorithm are fed back to the packaging equipment in real time;

[0124] S42. The packaging equipment automatically adjusts the packaging process based on the feedback packaging parameters.

[0125] The working principle of the above technical solution is as follows: In the initial stage, packaging parameters are optimized through AI algorithms. Based on real-time data during the packaging process (such as temperature, pressure, displacement, etc.), the algorithm analyzes the packaging process and outputs the optimal packaging parameters. These parameters are dynamically adjusted according to real-time production needs and equipment status to ensure the best packaging quality. The optimized packaging parameters are transmitted to the packaging equipment in real time through a feedback control system (such as a PID controller). The PID controller plays a crucial role here. Specifically, the PID controller can automatically adjust the control quantity (e.g., adjust temperature, pressure, etc.) based on the deviation between the target value and the actual measured value. Its basic principle is:

[0126] P (proportion): Adjust the proportion by calculating the difference between the actual parameter and the target value.

[0127] I (Integral): Eliminate long-term small deviations by accumulating errors.

[0128] D (derivative): Predicts future deviation changes and makes adjustments in advance to reduce response delay. Through PID control, packaging equipment can respond promptly to changes in packaging parameters, ensuring the accuracy of the packaging process.

[0129] Once the PID controller adjusts the control input based on feedback information, the packaging equipment will automatically adjust the packaging process based on these feedback parameters. For example:

[0130] Adjust the temperature of the heating device to ensure appropriate heating conditions.

[0131] Adjust the pressure of the pressure device to ensure the sealing and firmness of the seal.

[0132] Adjust the movement distance of the displacement device to ensure that the mechanical parts of the equipment can accurately perform the required packaging operations.

[0133] During these adjustments, the packaging equipment monitors various process parameters in real time, such as temperature and pressure, to ensure that the actual operating conditions are consistent with the target packaging parameters. The system continuously monitors the operating status of the packaging equipment, especially key process parameters such as temperature and pressure, to ensure the stability and safety of the equipment during the adjustment process. This monitoring data is fed back to the control system. If any abnormality is detected, the control system will make corresponding adjustments to avoid equipment damage or unqualified packaging due to improper parameters.

[0134] The above technical solution achieves the following results: By optimizing packaging parameters through AI algorithms and combining them with a real-time feedback control system (such as a PID controller), precise control of packaging parameters can be ensured. The PID controller automatically adjusts the control quantity based on the deviation between the target value and the actual value, thereby achieving precise adjustment of the packaging process. Especially in complex production processes, this automatic adjustment can effectively reduce errors caused by human factors and improve the consistency and accuracy of packaging quality. A key advantage of this technical solution is its ability to provide real-time feedback on packaging parameters and make dynamic adjustments. The packaging equipment monitors key parameters such as temperature and pressure in real time to ensure that the equipment always operates in the optimal state. This real-time feedback not only improves production efficiency but also corrects deviations in a timely manner, avoiding the production of defective products. The automated packaging adjustment process reduces manual intervention and operational errors while optimizing the production process. This efficient production mode can significantly improve production efficiency, reduce downtime and equipment failures, thereby reducing overall production costs. In addition, precise packaging control can reduce material waste and improve resource utilization. By monitoring the operating status of the packaging equipment in real time, especially during process adjustments, the stability and safety of the packaging process are ensured. PID controllers and other control algorithms respond precisely to real-time data when adjusting packaging parameters, preventing equipment damage or safety accidents caused by excessive temperature, pressure, or equipment malfunctions. This intelligent control solution not only ensures that each batch of packaging meets the set standards but also reduces quality fluctuations caused by process instability. This high consistency ensures high-quality final products, enhancing their market competitiveness. The packaging equipment can flexibly adjust based on real-time feedback of packaging parameters to adapt to the packaging needs of different products. This flexibility allows the production process to respond quickly to changes, adapting to the production requirements of various products and improving the system's versatility and adaptability.

[0135] In one embodiment of the present invention, step S5 includes:

[0136] S51. Real-time detection of the packaged display screen is performed using a feedback control system based on a multi-point measurement method.

[0137] S52. Compare the real-time detected display performance data with the preset standard value to evaluate whether the display performance meets the standard.

[0138] S53. If the display performance is found to be substandard, the feedback control system will send the detection data to the AI ​​algorithm processing system; the AI ​​algorithm will then adjust the packaging parameters again based on the feedback data and perform secondary optimization.

[0139] S54. During the secondary optimization process, the Bayesian optimization algorithm is used to continuously optimize the display by combining historical data and real-time detection data, and the above process is repeated until the display performance of the screen meets the requirements.

[0140] The working principle of the above technical solution is as follows: The feedback control system performs real-time detection of the packaged display screen through a multi-point measurement method. The main indicators detected include brightness uniformity, color accuracy, and contrast ratio. To ensure accuracy, professional display testing instruments, such as luminance meters and colorimeters, are used. The luminance meter measures the brightness distribution of the display screen, while the colorimeter measures the color accuracy. This process ensures comprehensive monitoring of the display screen performance; the real-time detected display screen performance data is compared with preset standard values. These standard values ​​can be flexibly set according to different application scenarios or customer needs. For example, for high-end display devices, the standard deviation of brightness uniformity should be less than 3%, and the color accuracy ΔE value should be less than 1.5. If the detected display performance does not meet these standard requirements, subsequent adjustments and optimizations will be performed; if the display performance is substandard, the detection system will feed the real-time detection data back to the AI ​​algorithm processing system. The AI ​​algorithm automatically adjusts the packaging parameters based on the feedback data and performs secondary optimization through process adjustments. This adjustment is based on specific display performance deviations, ensuring that the display screen performance is improved during the adjustment process; in the secondary optimization process, the AI ​​algorithm combines historical data and real-time detection data, using a Bayesian optimization algorithm for continuous optimization. Bayesian optimization is a probabilistic model-based optimization method that guides subsequent experiments by modeling historical experimental results to minimize the number of trials and errors during the optimization process. The AI ​​system will further adjust based on the optimization results, repeating the above optimization process until the display performance meets the predetermined standard requirements.

[0141] The effects of the above technical solution are as follows: Through real-time detection and dynamic feedback adjustment, the display performance of the screen can be ensured to remain within a high standard range. Indicators such as brightness uniformity, color accuracy, and contrast are strictly controlled throughout the production process, avoiding product quality instability. The use of AI algorithms and Bayesian optimization algorithms makes the entire packaging process more intelligent. AI algorithms adjust based on real-time and historical data, quickly identifying and optimizing performance deviations, thus improving production efficiency. Bayesian optimization algorithms also reduce trial-and-error times, shorten optimization cycles, and lower production costs. This solution can adaptively adjust evaluation standards according to different customer needs and application scenarios. For example, for high-end display devices, the system can set stricter requirements for brightness uniformity and color accuracy, while for ordinary devices, the standards can be appropriately relaxed. This flexibility meets the diverse needs of different markets and customers. Since most processes are automatically controlled and optimized by the AI ​​system, human intervention is reduced, lowering the possibility of human error. This not only improves production efficiency but also ensures greater consistency in the quality of the final product. Because this technical solution can produce high-performance, high-quality, consistent displays, the product's competitiveness in the market is greatly enhanced. Consumers have increasingly higher demands for display quality, and displays using this technology can better meet market needs and enhance brand image. This technical solution can promote the intelligent transformation of the display packaging process, using data-driven optimization methods to replace traditional experience-based adjustments. This not only improves production accuracy but also lays the foundation for more intelligent manufacturing technology applications in the future.

[0142] According to one embodiment of the present invention, an AI algorithm-optimized MiniLED display optical packaging system includes a memory, a processor, and a computer program stored in the memory and executable on the memory. The processor executes the program to implement the AI ​​algorithm-optimized MiniLED display optical packaging method as described above.

[0143] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A MiniLED display optical packaging method based on AI algorithm optimization, characterized in that, The method includes: S1. Real-time acquisition of various data during the MiniLED chip packaging process; S2. Obtain historical packaging data of MiniLED chips and build an AI algorithm model based on machine learning algorithms; S3. During the packaging process, the AI ​​algorithm processing system optimizes the packaging parameters in real time based on the data collected in real time and the trained AI algorithm model. S4. Based on the feedback control system, the packaging parameters optimized by the AI ​​algorithm are fed back to the packaging equipment in real time; S5. Real-time monitoring of the packaged display screen is performed through a feedback control system.

2. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 1, characterized in that, S1 includes: S11. Select and deploy a multi-source data acquisition system; S12. The deployed multi-source data acquisition system collects relevant data respectively; S13. Process the data collected by the multi-source data acquisition system and transmit it to the AI ​​algorithm processing system via wireless transmission.

3. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 2, characterized in that, S13 includes: S131. Preprocess the data; S132. By establishing a spatial coordinate system, spatial alignment of data collected by different sensors is performed based on feature point matching, and the data are represented in the same coordinate system. S133. Based on timestamp technology, a timestamp is added to each data point during the data acquisition process, and the data from different sensors are aligned using the timestamps; S134. Extract different features from data from different sources, and fuse the features from different sources to form a comprehensive feature vector; S135. Different compression algorithms are used to compress different types of data; S136. Before data transmission, the data is encrypted using an encryption algorithm and key to convert the original data into ciphertext data; at the receiving end, the ciphertext data is decrypted using the corresponding decryption algorithm and key to restore the original data. S137. Encrypted data is transmitted through a multi-channel transmission protocol. During data transmission, the data transmission path is adjusted in real time through a load balancing algorithm, and the temperature of the transmission environment is monitored in real time. The wireless transmission parameters are adjusted according to the temperature changes. S138. Based on data analysis, establish a model between temperature and wireless transmission performance, predict transmission performance at different temperatures, and take corresponding measures to optimize in advance.

4. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 3, characterized in that, S134 includes: The chip image captured by the optical camera is processed using an edge detection algorithm; the shape descriptor method is used to quantify the shape of the chip. Wavelength and intensity analysis is performed on the spectral data acquired by the spectrometer; and the full width at half maximum (FWHM) and peak area of ​​the spectral peaks are calculated. Based on the displacement data collected by the displacement sensor, the chip's moving distance and speed at different time points are calculated; the speed information is obtained by differentiating the displacement data using the differential method. For chips with rotational motion, analyze their angular changes; use angle sensors or calculate angular changes using displacement data; Analysis of cross-modal feature mismatch problem; PCA dimensionality reduction is performed on high-dimensional optical image features to extract the main components; A feature selection algorithm is used to select the most representative features from spectral features and displacement features; By using knowledge graphs, features of different modalities can be associated with existing knowledge; Establish a mathematical model between stress and features; based on the model, modify the extracted features.

5. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 3, characterized in that, S137 includes: During data transmission, the load of each transmission channel is monitored in real time; based on the load status, the data transmission path is dynamically adjusted to allocate data to channels with lower load. Establish a temperature data acquisition system to collect data from temperature sensors regularly; transmit the temperature data to the data processing center using wired or wireless methods. The wireless signal transmission power is adjusted in real time according to temperature changes; Select the appropriate modulation method based on temperature changes; and determine different adjustment strategies based on dynamic parameter adjustment rules, according to the temperature change range and rate of change. Collect wireless transmission performance data at different temperatures, preprocess the collected data, and establish a model between temperature and wireless transmission performance based on machine learning algorithms; Validate the trained model by evaluating its performance using data not used in training; optimize the model based on the validation results. A dynamic adjustment mechanism is established to dynamically adjust the wireless transmission parameters based on the real-time prediction results of the model. At the same time, the adjusted transmission performance data is collected and fed back to the model for further optimization, forming a closed-loop optimization system.

6. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 1, characterized in that, S2 includes: S21. Collect a large amount of historical packaging data of MiniLED chips and preprocess the collected historical packaging data; S22. Use deep learning frameworks to build AI algorithm models; S23. Divide the preprocessed historical data into training set, validation set and test set; S24. Train the AI ​​algorithm model using the training set; S25. Use the validation set to evaluate the trained model and adjust the model's hyperparameters based on the evaluation results.

7. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 1, characterized in that, The S3 includes: S31. During the encapsulation process, the data collected in real time by the multi-source data acquisition system is transmitted to the AI ​​algorithm processing system; S32. The AI ​​algorithm processing system inputs real-time data into the trained AI algorithm model, and the model performs inference calculations. S33. Based on the prediction results, the AI ​​algorithm optimizes the encapsulation parameters using optimization algorithms; S34. The AI ​​algorithm outputs the optimized packaging parameters in the form of digital signals and transmits them to the packaging device.

8. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 1, characterized in that, The S4 includes: S41. Based on the feedback control system, the packaging parameters optimized by the AI ​​algorithm are fed back to the packaging equipment in real time; S42. The packaging equipment automatically adjusts the packaging process based on the feedback packaging parameters.

9. The MiniLED display optical packaging method based on AI algorithm optimization according to claim 1, characterized in that, The S5 includes: S51. Real-time detection of the packaged display screen is performed using a feedback control system based on a multi-point measurement method. S52. Compare the real-time detected display performance data with the preset standard value to evaluate whether the display performance meets the standard. S53. If the display performance is found to be substandard, the feedback control system will send the detection data to the AI ​​algorithm processing system; the AI ​​algorithm will then adjust the packaging parameters again based on the feedback data and perform secondary optimization. S54. During the secondary optimization process, the Bayesian optimization algorithm is used to continuously optimize the display by combining historical data and real-time detection data, and the above process is repeated until the display performance of the screen meets the requirements.

10. An AI algorithm-optimized optical packaging system for MiniLED displays, characterized in that, The device includes a memory, a processor, and a computer program stored on and executable on the memory, wherein the processor executes the program to implement the AI ​​algorithm-optimized MiniLED display optical packaging method as described in any one of claims 1-9.

Images