Data Analysis-Based Risk Early Warning Method for LCD Display Production

Abstract

This invention discloses a data analysis-based method for early warning of production risks in LCD displays, belonging to the field of production risk early warning technology. This data analysis-based method acquires process execution sequence data and adhesive layer reflection signal data for each LCD display; performs joint deconstruction processing on the process execution sequence data to obtain LCD bonding conduction resistance characteristic values; decomposes the adhesive layer reflection signal data to obtain adhesive layer reflection sub-signals at each scale; and analyzes LCD adhesive layer reflection defect characteristic values ​​based on a pre-trained adhesive layer reflection mapping model. This invention uses the LCD bonding conduction resistance characteristic values ​​and LCD adhesive layer reflection defect characteristic values ​​after feature normalization mapping processing to perform joint production risk early warning processing on a set production batch, thereby enabling timely early warning of minor process fluctuations and hidden material defects that may occur during production, ensuring the quality of the final product.

Description

Technical Field

[0001] This invention relates to the field of production risk early warning technology, specifically to a data analysis-based method for early warning of production risks in LCD displays. Background Technology

[0002] As a widely used display technology in modern electronic devices, LCD displays involve precise control of multiple stages in their production process, especially the inspection of the adhesive lamination process. In these processes, any minor quality defects may affect the performance and stability of the final product. In particular, for high-quality displays, the quality of the adhesive layer directly determines the brightness uniformity, color accuracy, and reliability of the display.

[0003] In recent years, data analysis has gradually become a hot topic in research and application. By mining data in production, it is possible to monitor and analyze the production quality of LCD displays in real time during the production process and take timely early warning measures, thereby effectively avoiding potential quality problems.

[0004] The limitations of existing technologies include at least the following problems: Existing technologies struggle to conduct in-depth analysis of the bonding process and the internal quality of the adhesive layer during LCD display production. This makes it difficult for production quality monitoring to make timely and accurate judgments when faced with complex process fluctuations and hidden material defects, which can easily lead to batch quality risks. For example, existing technologies cannot express minor inconsistencies in temperature or pressure and interface contact during the bonding process. If production continues, it will lead to insufficient adhesive bonding, causing display abnormalities or even delamination in subsequent testing or use. In addition, existing technologies also struggle to detect micro-defects such as microcracks and bubbles within the adhesive layer, which are related to the final production risks. This can lead to defective products flowing into the next process or leaving the factory, thus affecting the quality of the entire batch of products. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a data analysis-based method for early warning of production risks in LCD displays. This method solves the problem that existing technologies struggle to analyze the bonding process and adhesive layer quality in conjunction with each other, leading to the failure to detect minute defects in a timely manner and thus affecting product quality.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data analysis-based LCD display production risk early warning method, comprising the following steps: within a set period, acquiring the process operation sequence data and adhesive layer reflection signal data of each LCD display in a set production batch; performing joint deconstruction processing on the process operation sequence data of each LCD display in the set production batch to obtain the LCD lamination conduction hindrance characteristic value of the set production batch; performing multi-scale decomposition processing on the adhesive layer reflection signal data of each LCD display in the set production batch to obtain the adhesive layer reflection sub-signal at each scale of the corresponding LCD display; analyzing the LCD adhesive layer reflection defect characteristic value of the set production batch based on a pre-trained adhesive layer reflection mapping model and combined with the adhesive layer reflection sub-signal at each scale of each LCD display in the set production batch; performing feature normalization mapping processing on the LCD lamination conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value of each LCD display in the set production batch; and performing joint production risk early warning processing on the set production batch based on the LCD lamination conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value after feature normalization mapping processing.

[0007] Further, the process operation sequence data includes the pressing temperature value, pressing pressure value, interface contact resistance value, heat flux density value, adhesive layer stress feedback value, and electrostatic potential distribution value at each time point. The specific steps to obtain the LCD pressing conduction hindrance characteristic value in the set production batch are as follows: Read the process operation sequence data of each LCD display in the set production batch and perform preprocessing; Based on the preprocessed process operation sequence data of each LCD display in the set production batch, analyze the pressing response timing characteristic set of the corresponding LCD display, including the pressing synchronization balance characteristic value and interface response coordination characteristic value at each time point; Perform timing evolution processing on the pressing response characteristic set of each LCD display in the set production batch to obtain the LCD pressing conduction hindrance characteristic value in the set production batch.

[0008] Furthermore, the specific steps for analyzing the pressing response timing characteristic set of each LCD display in the production batch are as follows: Based on the pressing temperature value, pressing pressure value, and heat flux density value of each LCD display in the production batch at each time point, analyze the pressing synchronization balance characteristic value at the corresponding time point; Based on the interface contact impedance value, adhesive layer stress feedback value, and electrostatic potential distribution value of each LCD display in the production batch at each time point, analyze the interface response coordination characteristic value at the corresponding time point.

[0009] Furthermore, the specific steps of the timing evolution processing are as follows: Based on the timing feature set of the pressing response of each LCD display in the set production batch, analyze the pressing conduction feature value at the corresponding time point; based on the pressing conduction feature value of each LCD display in the set production batch at each time point, extract the conduction extension feature value and conduction fluctuation aggregation feature value of the corresponding LCD display, and analyze the LCD pressing conduction stagnation feature value of the corresponding LCD display.

[0010] Further, the specific steps for analyzing the pressing conduction characteristic values ​​of each LCD display screen at each time point in the production batch are as follows: Read the pressing response timing feature set of each LCD display screen in the production batch, extract the average pressing synchronization balance feature and the average interface response coordination feature of the corresponding LCD display screen, and sum them to obtain the conduction sum value of the corresponding LCD display screen; Ratio the average pressing synchronization balance feature and the average interface response coordination feature of each LCD display screen in the production batch to the conduction sum value to obtain the synchronization balance adjustment coefficient and the interface coordination adjustment coefficient of the corresponding LCD display screen; Based on the synchronization balance adjustment coefficient, the interface coordination adjustment coefficient, and the pressing response timing feature set of each LCD display screen in the production batch, analyze the pressing conduction characteristic values ​​at the corresponding time points.

[0011] Furthermore, the mucosal reflector signal is specifically the mucosal acoustic wave reflection intensity value at each time point of each scale, and the mucosal reflection mapping model includes a signal input layer, a feature parsing layer, and a mapping output layer.

[0012] Further, the specific steps for analyzing the characteristic values ​​of LCD adhesive layer reflection defects in the production batch are as follows: The adhesive layer reflection sub-signal at each scale of each LCD display in the production batch is input into a pre-trained adhesive layer reflection mapping model. The internal reflection sensing feature set at the corresponding scale is extracted, including reflection propagation distortion feature values, reflection frequency offset distortion feature values, and adhesive layer reflection mismatch feature values. Based on the adhesive layer reflection sub-signal at each scale of each LCD display in the production batch, the reflection weight contribution value at the corresponding scale is analyzed. Based on the reflection weight contribution value at each scale of each LCD display in the production batch and the internal reflection sensing feature set, the characteristic values ​​of LCD adhesive layer reflection defects in the production batch are analyzed.

[0013] Furthermore, the specific steps for extracting the internal reflection sensing feature set of each LCD display at each scale in a given production batch are as follows: In the signal input layer of the capillary reflection mapping model, the capillary reflection sub-signal at each scale of each LCD display in the given production batch is received and preprocessed; in the feature analysis layer of the capillary reflection mapping model, the reflection time sequence feature vector at the corresponding scale is analyzed based on the preprocessed capillary reflection sub-signal at each scale of each LCD display in the given production batch; in the signal input layer of the capillary reflection mapping model, the internal reflection sensing feature set at the corresponding scale is output based on the reflection time sequence feature vector at each scale of each LCD display in the given production batch.

[0014] Furthermore, the specific steps for analyzing the reflection weight contribution value of each LCD display screen in the production batch at each scale are as follows: Based on the adhesive layer reflective sub-signal of each LCD display screen in the production batch at each scale, analyze the adhesive layer reflection entropy value at the corresponding scale; based on the adhesive layer reflection entropy value of each LCD display screen in the production batch at each scale, analyze the reflection weight contribution value at the corresponding scale.

[0015] Furthermore, the specific steps for joint production risk early warning processing of a set production batch based on the LCD bonding conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value after feature normalization mapping are as follows: the LCD bonding conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value in the set production batch after feature normalization mapping are judged and analyzed with several preset risk early warning intervals respectively; based on the judgment and analysis results, corresponding production risk early warning measures are taken for the set production batch.

[0016] The present invention has the following beneficial effects:

[0017] (1) The data analysis-based LCD display production risk early warning method can achieve accurate production process monitoring by jointly analyzing process operation sequence data and adhesive layer reflection signal data. The process operation sequence data is deconstructed and processed to extract LCD bonding conduction resistance characteristic values. The adhesive layer reflectivity signal is processed by multi-scale decomposition technology to obtain adhesive layer reflection defect characteristic values. After normalization mapping, it is judged with the risk early warning interval, thereby implementing joint production risk early warning for LCD displays. This enables the early detection and timely warning of possible small process fluctuations and hidden material defects in the production process, thereby reducing the risk of defective products flowing into the next process or leaving the factory, and thus ensuring the quality of the final product.

[0018] (2) The data analysis-based LCD display production risk early warning method extracts the pressing response timing feature set by deeply analyzing the process running sequence data, and performs timing processing based on this to extract the conduction extension feature value and conduction fluctuation aggregation feature value. This helps to identify hysteresis and stagnation in the pressing energy transfer process, timely discover potential risks such as uneven pressing or poor interface contact, and thus provide timely early warning. This improves the controllability of the production process and ensures that not only the generation of defective products can be reduced, but also the product qualification rate can be significantly improved.

[0019] (3) The data analysis-based LCD display production risk warning method introduces a pre-trained adhesive layer reflection mapping model and combines multi-scale analysis and reflection weight contribution value to accurately analyze the internal defect state of the adhesive layer. After the adhesive layer reflection signal is processed by multi-scale decomposition, it can accurately capture the reflection characteristics at each scale. By extracting the reflection perception feature set, it can identify small defects inside the adhesive layer such as reflection propagation distortion and frequency offset distortion. At the same time, the LCD adhesive layer reflection defect feature value generated by the analysis of reflection weight contribution value can significantly improve the defect identification rate in LCD display production, make quality control more refined, and reduce quality problems caused by failure to detect defects in time, thereby improving the overall production efficiency.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This is a flowchart of the LCD display production risk early warning method based on data analysis according to the present invention.

[0022] Figure 2 This is a flowchart illustrating the specific steps involved in obtaining the LCD pressing conduction resistance characteristic value in a set production batch within the data analysis-based LCD display production risk early warning method of the present invention.

[0023] Figure 3 This is a flowchart illustrating the specific steps involved in analyzing and setting characteristic values ​​of LCD adhesive layer reflection defects in a production batch within the data analysis-based LCD display production risk early warning method of the present invention. Detailed Implementation

[0024] Please see Figure 1This invention provides a technical solution: a data analysis-based method for early warning of production risks in LCD displays, comprising the following steps: within a set period, acquiring process operation sequence data and adhesive layer reflection signal data for each LCD display in a set production batch (random sampling is performed from the production batch to extract several LCD displays). The adhesive layer reflection signal data is specifically the adhesive layer acoustic wave reflection intensity value at each time point, which can be periodically received by an embedded ultrasonic transducer during the pressing process, i.e., the adhesive layer acoustic wave reflection intensity value at each time point.

[0025] The process timing data of each LCD display in a set production batch are jointly deconstructed to obtain the LCD lamination conduction hysteresis characteristic value in the set production batch; the adhesive layer reflection signal data of each LCD display in the set production batch are decomposed into multi-scale data to obtain the adhesive layer reflection sub-signal of the corresponding LCD display at each scale (i.e., wavelet transform is performed based on a preset wavelet basis function; during the wavelet transform, the resolution of the signal in the frequency domain and time domain is controlled by adjusting the wavelet scale parameter and time shift parameter, respectively, where the scale parameter is used to adjust the wavelet scaling ratio to determine the analysis frequency band, and the time shift parameter is used to control the movement position of the analysis window on the time axis; by expanding the wavelet function at different scales, the adhesive layer reflectivity signal is subjected to layered filtering, and the original complex signal is decomposed into several sub-signals with different frequency bands).

[0026] Based on a pre-trained adhesive layer reflection mapping model, and combined with adhesive layer reflection sub-signals at each scale for each LCD display in a given production batch, the characteristic values ​​of LCD adhesive layer reflection defects in the given production batch are analyzed. The LCD bonding conduction hindrance characteristic values ​​and LCD adhesive layer reflection defect characteristic values ​​for each LCD display in the given production batch are then subjected to feature normalization mapping processing (e.g., normalization maps the corresponding values ​​to between 0 and 1). Based on the LCD bonding conduction hindrance characteristic values ​​and LCD adhesive layer reflection defect characteristic values ​​after feature normalization mapping processing, a joint production risk early warning processing is performed on the given production batch.

[0027] The specific steps for joint production risk early warning processing of a set production batch based on the LCD bonding conduction resistance characteristic value and LCD adhesive layer reflection defect characteristic value after feature normalization mapping are as follows: The LCD bonding conduction resistance characteristic value and LCD adhesive layer reflection defect characteristic value in the set production batch after feature normalization mapping are compared with several preset risk early warning intervals. Each risk early warning interval includes one conduction resistance early warning interval and one adhesive layer reflection defect early warning interval, and each risk early warning interval corresponds to one production risk early warning measure. Based on the judgment and analysis results, corresponding production risk early warning measures are taken for the set production batch. That is, based on the production risk early warning measures corresponding to the LCD bonding conduction resistance characteristic value and LCD adhesive layer reflection defect characteristic value of each LCD display in the set production batch after feature normalization mapping being within the preset risk early warning interval, production risk early warning processing is performed on each LCD display in the set production batch, including but not limited to the following examples:

[0028] Level 1 Warning (Emergency Risk - Red):

[0029] Early warning range for conduction hysteresis: 0.9-1.0;

[0030] Early warning range for adhesive layer reflection defects: 0.9-1.0;

[0031] Early warning measures: Trigger an audible and visual alarm, send an emergency notification to production, quality, and equipment engineers, and prepare to implement contingency plans such as cleaning the pressing mold, calibrating the pressure sensor, and replacing the adhesive batch;

[0032] Level 2 Warning (High Risk - Orange):

[0033] Early warning range for conduction hysteresis: 0.7-0.89;

[0034] Early warning range for adhesive layer reflection defects: 0.7-0.89;

[0035] Early warning measures: Send alarm information to workstation operators and team leaders, automatically increase the sampling frequency of online inspection (AOI) for this production batch to 100%; and prompt operators to make minor compensatory adjustments to the pressing pressure and verify the adhesive viscosity parameters.

[0036] Level 3 Warning (Moderate Risk - Yellow):

[0037] Early warning range for conduction hysteresis: 0.5-0.69;

[0038] Early warning range for adhesive layer reflection defects: 0.5-0.69;

[0039] Early warning measures: Display early warning prompts on the workstation control panel, increase the sampling frequency to twice the normal rate, generate a work order, and recommend cleaning and maintaining the pressing rollers after the end of this batch, and checking the adhesive supply pipeline.

[0040] Level 4 Warning (Mild Risk - Blue):

[0041] Early warning range for conduction hysteresis: 0.3-0.49;

[0042] Early warning range for adhesive layer reflection defects: 0.3-0.49;

[0043] Early warning measures: Automatically record events where the characteristic value exceeds the limit and include them in the history of the equipment or production line. At the same time, give a clear prompt on the workstation operation interface to remind the operator to pay attention to the parameter. If the characteristic value is continuously in the fourth warning range or higher for three consecutive production cycles, the warning level will be automatically raised to the third level (medium risk - yellow) and trigger the associated early warning measures (such as increasing the sampling frequency).

[0044] Level 5 (Normal Operation - Green):

[0045] Early warning range for conduction hysteresis: 0.0-0.29;

[0046] Early warning range for adhesive layer reflection defects: 0.0-0.29;

[0047] At this point, no warning is issued, the production process is under control, existing parameters are maintained, and production is carried out according to standard procedures.

[0048] Specifically, such as Figure 2 As shown, the process execution sequence data includes the pressing temperature, pressing pressure, interface contact resistance, heat flux density, adhesive layer stress feedback, and electrostatic potential distribution at each time point. The specific steps to obtain the LCD pressing conduction hindrance characteristic value in the set production batch are as follows: Read the process execution sequence data of each LCD display in the set production batch and perform preprocessing; Based on the preprocessed process execution sequence data of each LCD display in the set production batch, analyze the pressing response timing characteristic set of the corresponding LCD display, including the pressing synchronization balance characteristic value and interface response coordination characteristic value at each time point; Perform timing evolution processing on the pressing response characteristic set of each LCD display in the set production batch to obtain the LCD pressing conduction hindrance characteristic value in the set production batch.

[0049] The pressing temperature value is the temperature of the heating head during the pressing stage, which can be collected in real time by a thermocouple installed on the inner surface of the pressing head.

[0050] The pressing pressure value is the actual pressure applied by the pressure head to the LCD panel. It can be collected by setting a force sensor or pressure sensor (such as piezoelectric or strain gauge type) under the pressure head or in the cylinder circuit.

[0051] The interface contact impedance value is the tightness of the bonding interface and can reflect the presence of bubbles or inclusions. It can be obtained by setting a microelectrode conductive bonding sensor between the pressure head and the substrate to form a closed electrical measurement circuit. During the pressing process, the change in circuit resistance is detected by low amplitude AC current (such as 1 kHz) to obtain the interface contact impedance value in real time.

[0052] The heat flux density is the intensity of heat transfer per unit area through the adhesive layer. It can be achieved by embedding a thin heat flow sensor on the back of the pressing platform or the pressure head. This sensor consists of two layers of thermally conductive material with a thermocouple diaphragm sandwiched in between. When the pressing process occurs, a small temperature difference is generated between the upper and lower layers. The thermocouple diaphragm outputs a millivolt-level electrical signal that is proportional to the temperature difference. According to the calibration coefficient of the sensor, this electrical signal is converted into the heat transfer rate per unit area, i.e., the heat flux density.

[0053] The stress feedback value of the adhesive layer is the stress value inside the adhesive layer during the pressing process. It can be collected in real time by arranging strain gauges or piezoelectric stress sensors at the bottom of the press head or on the substrate support platform.

[0054] The electrostatic potential distribution value is the electrostatic charge state of the bonding area surface. It can be obtained by arranging several electrostatic induction probes above and around the bonding station along the edge of the bonding area. Each electrostatic induction probe senses the electrostatic potential changes in the bonding area surface and the surrounding air layer in real time, and outputs an electrical signal proportional to the average potential of the sensed area. Combined with the spatial position of each electrostatic induction probe in the bonding plane coordinate system, spatial interpolation is performed on the potential data of multiple points at the same time to obtain the electrostatic potential distribution map of the bonding area at the corresponding time. The potential values ​​in the electrostatic potential distribution map are statistically processed to calculate the average potential value, which is then used as the electrostatic potential distribution value at the current time.

[0055] The specific steps for analyzing the pressing response timing characteristic set of each LCD display in the production batch are as follows: Based on the pressing temperature, pressing pressure, and heat flux density values ​​at each time point for each LCD display in the production batch, analyze the pressing synchronization balance characteristic values ​​at the corresponding time points. Specifically, normalize the pressing temperature, pressing pressure, and heat flux density values ​​at each time point (i.e., remove units and convert to dimensionless values), and average the normalized pressing temperature, pressing pressure, and heat flux density values ​​at each time point to obtain the value for each... The average pressing value at each time point was calculated, and the normalized pressing temperature, pressing pressure, and heat flux density values ​​at each time point were compared with the average pressing value (absolute value was taken). The results were then weighted and summed. The weighted sum was transformed using the reciprocal suppression mapping function f(x)=1 / (1+x), such as 1 / (1+the weighted sum), to obtain the pressing synchronization balance characteristic value at each time point. This characteristic value is used to characterize the coordinated balance state of heat input, mechanical loading, and energy conduction at that time point, i.e., the instantaneous stability of the equipment driving process.

[0056] Based on the interface contact impedance value, adhesive layer stress feedback value, and electrostatic potential distribution value of each LCD display screen in a production batch at each time point, the interface response coordination characteristic value at the corresponding time point is analyzed. Specifically, the interface contact impedance value, adhesive layer stress feedback value, and electrostatic potential distribution value at each time point are normalized (i.e., converted to dimensionless values ​​by removing units) and the average value is taken to obtain the interface response benchmark value at each time point (reflecting the overall feedback level of the interface at that moment). The interface contact impedance value, adhesive layer stress feedback value, and electrostatic potential distribution value at each time point after normalization are respectively compared with the interface response benchmark value (the absolute value is taken), and then weighted summation is performed. The weighted summation result is transformed using the reciprocal suppression mapping function f(x)=1 / (1+x) to obtain the interface response coordination characteristic value at each time point, which is used to characterize the overall feedback coordination degree of the bonding interface at that time point.

[0057] The specific steps of the time-series evolution processing are as follows: Based on the pressing response time-series feature set of each LCD display in the set production batch, analyze the pressing conduction feature value at the corresponding time point; based on the pressing conduction feature value of each LCD display in the set production batch at each time point, extract the conduction extension feature value and conduction fluctuation aggregation feature value of the corresponding LCD display, and analyze the LCD pressing conduction stagnation feature value of the corresponding LCD display. Specifically, extract the difference of pressing conduction feature value at any adjacent time point, and count the difference between adjacent pressing conduction feature value differences, take the absolute value, normalize the result and calculate the average value to obtain the average change trend amplitude. Transform the average change trend amplitude through the reciprocal mapping function to obtain the conduction extension feature value (used to characterize the continuous transmission of pressing energy and stress in the time series. When this feature value is low, it indicates that there is a sudden change in pressing conduction, energy conduction is blocked or interface reaction is delayed, and the LCD display is prone to defects such as local bubble residue, interface delamination, bright spots or indentations).

[0058] Read the compression conduction characteristic value at each time point, and extract the mean and standard deviation of the compression conduction characteristic. Use these to set the fluctuation threshold, which is the sum of the mean and standard deviation of the compression conduction characteristic multiplied by a preset multiplier (which can be between 1.0 and 2.5). Count the number of time points where the compression conduction characteristic value is higher than the fluctuation threshold, and perform a ratio with the total number of time points in the cycle to extract the conduction fluctuation aggregation characteristic value (used to characterize the non-uniformity of local energy input; when this characteristic value is high, it indicates that the compression energy is concentrated in a certain area in a short period of time, causing heat accumulation or mechanical stress overload, and the LCD screen is prone to defects such as bright spots, local overpressure, thermal damage, or uneven interface stress).

[0059] The conduction extension characteristic value and conduction fluctuation aggregation characteristic value of each LCD display in the set production batch are weighted. In this weighting process, the conduction extension characteristic value is transformed by the reciprocal suppression mapping function f(x)=1 / (1+x), such as 1 / (1+conduction extension characteristic value), and the average value is taken to obtain the LCD pressing conduction hindrance characteristic value in the set production batch. This characteristic value is used to characterize the degree of hysteresis and hindrance in the pressing process of energy and stress transmission. When the characteristic value is high, it indicates that the pressing energy is difficult to be smoothly transferred to the interface layer or released to the outside, resulting in local heat retention and stress accumulation. The LCD display is prone to defects such as interface delamination, adhesive layer wrinkles or poor adhesion.

[0060] In this implementation plan, by conducting in-depth analysis of process timing data, the factors affecting the quality of LCD displays during production can be comprehensively captured, thereby improving the ability to identify potential defects. Secondly, by performing time-series analysis on the pressing response timing feature set, the extracted conduction extension feature value and conduction fluctuation aggregation feature value can be used to monitor energy transfer and stress distribution during the pressing process in real time and trigger early warnings in a timely manner. This allows for the early detection of problems in production, improves production efficiency, and reduces resource waste. Finally, through precise quality monitoring, the quality risks in LCD display production can be significantly reduced, especially defects such as local bubble residue and interface delamination, improving product stability and ultimately enhancing product quality.

[0061] Specifically, the steps for analyzing the pressing conduction characteristic value of each LCD display screen at each time point in the production batch are as follows: Read the pressing response timing characteristic set of each LCD display screen in the production batch (i.e., the pressing synchronization balance characteristic value and interface response coordination characteristic value at each time point), extract the average pressing synchronization balance characteristic value and the average interface response coordination characteristic value of the corresponding LCD display screen respectively, and perform summation to obtain the conduction sum value of the corresponding LCD display screen.

[0062] The average values ​​of the pressing synchronization balance characteristics and interface response coordination characteristics of each LCD display in the set production batch are compared with the conduction sum value to obtain the corresponding LCD display's synchronization balance adjustment coefficient and interface coordination adjustment coefficient. Based on the synchronization balance adjustment coefficient, interface coordination adjustment coefficient, and pressing response timing feature set of each LCD display in the set production batch, the pressing conduction characteristic value at the corresponding time point is analyzed. Specifically, the pressing synchronization balance characteristic value and interface response coordination characteristic value of each LCD display in the set production batch at each time point are weighted. In the weighting process, the synchronization balance adjustment coefficient is used as the weight coefficient of the pressing synchronization balance characteristic value, and the interface coordination adjustment coefficient is used as the weight coefficient of the interface response coordination characteristic value, so as to obtain the pressing conduction characteristic value of each LCD display in the set production batch at each time point, which is used to characterize the smoothness of energy and stress transmission between the pressing execution end and the interface feedback end at that time point.

[0063] In this implementation scheme, by summing the mean values ​​of the pressing synchronization balance characteristics and the interface response coordination characteristics, and combining them with a weighted method, it not only considers the pressing synchronization and interface response at each time point, but also makes the feature analysis more flexible by adjusting the coefficients, which can dynamically reflect changes in the production process. Secondly, by combining the synchronization balance adjustment coefficient and the interface coordination adjustment coefficient with their respective feature values, the production process of each LCD display can be precisely adjusted, thereby improving the stability of production. Finally, through precise time-series analysis and the introduction of adjustment coefficients, potential production risks can be identified in a timely manner, especially the detection of minor defects in the adhesive layer interface or the poor energy and stress transmission during the pressing process, and accurate early warnings can be given, thereby effectively reducing the proportion of defective products in production.

[0064] Specifically, the mucosal reflector signal is the mucosal acoustic wave reflection intensity value at each scale time point (it should be noted that the interval between two adjacent scale time points is not the same at each scale, depending on the scale parameter selected at the corresponding scale). The mucosal reflection mapping model includes a signal input layer, a feature parsing layer, and a mapping output layer.

[0065] like Figure 3 As shown, the specific steps for analyzing the characteristic values ​​of LCD adhesive layer reflection defects in a set production batch are as follows: Input the adhesive layer reflection sub-signal at each scale of each LCD display in the set production batch into the pre-trained adhesive layer reflection mapping model, and extract the internal reflection perception feature set at the corresponding scale, including reflection propagation distortion feature value, reflection frequency offset distortion feature value, and adhesive layer reflection mismatch feature value; Based on the adhesive layer reflection sub-signal at each scale of each LCD display in the set production batch, analyze the reflection weight contribution value at the corresponding scale.

[0066] Based on the reflection weight contribution value and internal reflection perception feature set of each LCD display screen in the set production batch at each scale, the characteristic value of LCD adhesive layer reflection defect in the set production batch is analyzed. Specifically, the reflection propagation distortion characteristic value, reflection frequency offset distortion characteristic value, and adhesive layer reflection mismatch characteristic value of each LCD display screen in the set production batch at each scale are weighted and averaged with the reflection weight contribution value at the corresponding scale to obtain the comprehensive reflection propagation distortion characteristic value, comprehensive reflection frequency offset distortion characteristic value, and comprehensive adhesive layer reflection mismatch characteristic value of each LCD display screen in the set production batch. The values ​​are then weighted to obtain the reflection defect characteristic value of each LCD display screen, and the average value is taken to obtain the LCD adhesive layer reflection defect characteristic value in the set production batch.

[0067] It should be noted that in this implementation example, the weighting coefficients of each parameter in the weighted processing can be obtained using sample entropy weighting. Taking the weighted processing of obtaining reflection defect feature values ​​as an example, the comprehensive reflection propagation distortion feature value, comprehensive reflection frequency offset distortion feature value, and comprehensive adhesive layer reflection mismatch feature value of each LCD display are obtained, and their corresponding information entropy values ​​are extracted respectively. Then, their corresponding information entropy values ​​are transformed using the reciprocal suppression mapping function f(x)=1 / (1+x), such as 1 / (1+information entropy value of comprehensive reflection propagation distortion feature value), and summed to obtain the information entropy sum value. The corresponding transformed information entropy values ​​are then compared with the information entropy sum value to obtain the weighting coefficients corresponding to each parameter.

[0068] The specific steps for analyzing the reflection weight contribution value of each LCD display in the production batch at each scale are as follows: Based on the adhesive layer reflector signal at each scale of each LCD display in the production batch, analyze the adhesive layer reflection entropy value at the corresponding scale. Specifically, calculate the entropy value of the adhesive layer reflector signal at each scale based on Shannon information entropy to extract the adhesive layer reflection entropy value at the corresponding scale (used to measure the signal complexity of the adhesive layer reflector signal at each scale; the higher the value, the more complex the signal and the greater the information content, which may indicate defects such as microcracks, voids, bubbles or structural inhomogeneity at the adhesive layer interface).

[0069] Based on the adhesive layer reflection entropy value of each LCD display at each scale in the set production batch, the reflection weight contribution value at the corresponding scale is analyzed. Specifically, the adhesive layer reflection entropy value of each LCD display at each scale in the set production batch is summed to obtain the adhesive layer reflection entropy sum value of each LCD display in the set production batch. The adhesive layer reflection entropy value of each LCD display at each scale in the set production batch is then compared with the adhesive layer reflection entropy sum value, and the corresponding result is used as the reflection weight contribution value at the corresponding scale.

[0070] In this implementation scheme, by extracting and analyzing the internal reflection sensing feature set of the adhesive layer reflector signal at each scale, and combining it with the reflection weight contribution value, multiple changes in the signal can be comprehensively considered, thereby accurately capturing potential problems. Secondly, by calculating the adhesive layer reflection entropy value of the adhesive layer reflector signal using Shannon information entropy, it is possible to effectively identify minute anomalies or complex patterns in the signal. These may be caused by defects such as microcracks or bubbles at the adhesive layer interface. High entropy values ​​indicate greater signal complexity, suggesting the existence of potential interface defects, allowing for early identification and effective warning. Finally, by calculating the ratio of reflection entropy value to total entropy value, the reflection weight contribution value at each scale can be adjusted. This allows for adjustment of its influence on the corresponding features at each scale according to the complexity of the signal, enabling adaptive adjustment based on the signal performance at different scales, improving the accuracy of the analysis, ensuring that defects can be detected in a timely manner, and thus enabling timely capture of quality fluctuations in the LCD display production process, thereby providing early warning to prevent the problem from spreading and ensuring the stability of product quality.

[0071] Specifically, the steps for extracting the internal reflection sensing feature set of each LCD display at each scale in a set production batch are as follows: In the signal input layer of the viscous reflection mapping model, the viscous reflection sub-signal of each LCD display at each scale in the set production batch is received and the signal is preprocessed, such as: normalizing the viscous reflection signal data of each LCD display in the set production batch, mapping its value to between 0 and 1, and performing noise suppression processing, which can be done by using wavelet threshold denoising or median filtering methods to smooth and suppress random high-frequency noise and isolated spikes in the reflection signal;

[0072] In the feature parsing layer of the delamination reflection mapping model, based on the delamination reflection sub-signals at each scale of each LCD display in the preprocessed production batch, the reflection time-series feature vectors at the corresponding scales are analyzed. Specifically, this layer uses LSTM. The LSTM network captures the long-term dependencies between each time step through time dependency learning, extracting the changing trends of the signal at different scale time points. Through the recurrent neural network (RNN) mechanism, it updates its hidden state for each time step, thus remembering historical information in the time-series data. Based on this information, the time-series features of the delamination reflection signal are extracted. This process can process the delamination reflection sub-signals at each scale to provide reflection time-series feature vectors that reflect the time-varying patterns, such as:

[0073] A Hilbert transform (such as the Discrete Hilbert Transform algorithm) is performed on the mucosal reflector signal at each scale to calculate its analytic signal. Based on this analytic signal, the instantaneous phase value can be obtained. This extracts the instantaneous phase value of the reflection at each time point of each scale (i.e., the instantaneous phase value of the mucosal reflector signal at each scale's time point). The difference (absolute value) between the instantaneous phase values ​​at two adjacent time points at each scale is then calculated, and the average is taken to obtain the reflection at each scale. The phase nonlinearity value (the degree of phase distortion of the signal during propagation) is calculated. Based on the Hilbert transform, the reflection amplitude envelope at each time point of each scale is extracted. The difference (absolute value) of the reflection amplitude envelopes of two adjacent time points at each scale is processed and the mean value is taken to obtain the reflection amplitude nonlinearity at each scale. This value is then weighted with the reflection phase nonlinearity value at the corresponding scale to extract the reflection propagation distortion features at each scale. This feature is used to characterize the degree of signal propagation distortion caused by defects in the adhesive layer (such as microcracks, bubbles, etc.) during signal propagation.

[0074] For each scale, the cohesive reflector signal is segmented using a sliding time window (e.g., the length of the sliding time window is set to 5-10 scale time points at the corresponding scale, and the window sliding step size is set to one scale time point at the corresponding scale). For each sliding time window at each scale, the dominant frequency (i.e., the frequency component with the largest amplitude in the power spectrum of the reflector signal obtained after FFT) is extracted using Fast Fourier Transform (FFT), thus obtaining the local dominant frequency value corresponding to that sliding time window; the overall cohesive reflector signal at each scale is then processed. The dominant frequency (the frequency corresponding to the maximum amplitude value calculated by FFT over the entire signal length) is used as the reference frequency. The absolute value of the difference between the local dominant frequency and the reference frequency for each sliding time window is calculated to obtain the instantaneous frequency offset of that window. Statistical features of the instantaneous frequency offsets of all sliding time windows are extracted, including the maximum value, mean, and variance of the frequency offsets, and weighted to extract the reflection frequency offset distortion features, which are used to characterize the non-uniformity of the local interface state of the adhesive layer. The larger this feature is, the more serious the adhesive layer bonding quality defects of the fixed LCD display are, and the higher the production risk of subsequent display abnormalities and reliability failures.

[0075] Periodic identification is performed on the mucosal reflector signal at each scale to obtain several mucosal reflection periods at each scale (which can be identified using the zero-crossing method, such as setting a threshold, for example, 0.5, when the mucosal acoustic wave reflection intensity transitions from less than 0.5 to greater than 0.5, or from greater than 0.5 to less than 0.5, it is identified as a crossover point, and the crossover point from the transition from less than 0.5 to greater than 0.5 to the transition from greater than 0.5 to less than 0.5 is taken as the positive half-cycle in the mucosal reflection period, and vice versa).

[0076] For each scale, the acoustic wave reflection intensity values ​​of the adhesive layer at all time points in the positive half-cycle and negative half-cycle of each adhesive layer reflection cycle are integrated to obtain the cumulative reflection intensity of the positive half-cycle and the cumulative reflection intensity of the negative half-cycle in each adhesive layer reflection cycle at each scale. These are then combined using the formula |cumulative reflection intensity of the positive half-cycle - cumulative reflection intensity of the negative half-cycle| / cumulative reflection intensity of the positive half-cycle + cumulative reflection intensity of the negative half-cycle. The average of these results is used to extract the adhesive layer reflection mismatch feature at each scale, which characterizes the inhomogeneity of the adhesive layer interface. A high feature value indicates the possible presence of defects such as microcracks, voids, and impurities at the adhesive layer interface, resulting in reflection waveform mismatch. The reflection propagation distortion feature, reflection frequency offset distortion feature, and adhesive layer reflection mismatch feature are concatenated into a reflection time-series feature vector.

[0077] In the signal input layer of the lamina reflection mapping model, based on the reflection time-series feature vector of each LCD display in the set production batch at each scale, the internal reflection sensing feature set at the corresponding scale is output. Specifically, the reflection propagation distortion feature, reflection frequency offset distortion feature, and lamina reflection mismatch feature in the reflection time-series feature vector of each LCD display in the set production batch are activated by the Sigmoid function to obtain the reflection propagation distortion feature value, reflection frequency offset distortion feature value, and lamina reflection mismatch feature value between 0 and 1.

[0078] The pre-training steps for the mucous layer reflection mapping model are as follows:

[0079] The labeled dataset consists of adhesive reflection signal data and corresponding quality inspection results from several LCD display production batches. The data is labeled by production quality experts based on actual inspection results and quality feedback. Each sample in the labeled dataset includes adhesive reflection sub-signals of each LCD display in the production batch at each scale and corresponding quality truth labels (such as whether adhesive defects exist, defect type, etc.).

[0080] The dataset is preprocessed, including signal normalization (mapping reflectivity signal values ​​to between 0 and 1) and noise suppression (using wavelet thresholding or median filtering). The preprocessed dataset is then divided into training, validation, and test sets. For example, 80% of the data is used for training, 10% for validation, and 10% for testing. All data is arranged according to production batch and time order to ensure that the model can learn the temporal characteristics of the mucosal reflectivity signal during training.

[0081] The mucosal reflection mapping model is trained by inputting the preprocessed mucosal reflection sub-signals at each scale into an LSTM (Long Short-Term Memory) network for temporal feature extraction. The LSTM layer controls the information flow through its internal gating mechanism (input gate, forget gate, and output gate), learning the long-term dependencies between each time step. The LSTM can effectively capture the dynamic changes of the mucosal reflection signal, such as reflection propagation distortion, frequency shift, and reflection mismatch. The LSTM network optimizes the weights through the backpropagation algorithm (BPTT) to minimize prediction errors (such as mean squared error MSE or cross-entropy loss). The LSTM will gradually learn the long-term dependencies and short-term fluctuations of the mucosal reflection features, capturing the change process of the reflection signal.

[0082] During training, optimization algorithms (such as the Adam optimizer) are used to minimize the loss function, and hyperparameters (such as learning rate, number of hidden units in LSTM, batch size, etc.) are adjusted to improve model performance. The model is evaluated using a validation set to monitor its performance on the validation set and prevent overfitting.

[0083] After training, the model's generalization ability is evaluated using a test set to ensure that the model can accurately process unseen mucous reflection signals and extract effective internal reflection sensing feature sets from them.

[0084] In this implementation scheme, by constructing a cohesive layer reflection mapping model and using it to deeply deconstruct the cohesive layer reflective sub-signals at each scale, potential defects in the production process can be captured more comprehensively, especially subtle issues such as microcracks or bubbles. Secondly, by introducing LSTM, the long-term dependence of the signal over time can be captured, and the temporal features of the reflected signal can be extracted, thereby gaining a comprehensive understanding of the signal's changing trend in the production process. This helps to identify abnormal fluctuations or changes that occur in the production process, and generates features such as reflection propagation distortion and reflection frequency offset distortion, thereby effectively identifying cohesive layer interface defects in the LCD display and improving recognition accuracy. Finally, by combining the reflection temporal feature vector with the activation function (Sigmoid), subtle differences in signal changes during the production process can be captured more accurately, and the ability to identify potential defects can be effectively improved, thereby providing early warning to prevent the expansion of defects.

[0085] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0086] 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 data analysis-based method for early warning of risks in LCD display production, characterized in that, Includes the following steps: Within a set period, acquire the process operation sequence data and adhesive layer reflection signal data for each LCD display in a set production batch; The process timing data of each LCD display in a given production batch are jointly deconstructed to obtain the LCD pressing conduction hysteresis characteristic value in the given production batch. The adhesive layer reflection signal data of each LCD screen in the production batch is decomposed into multiple scales to obtain the adhesive layer reflection sub-signal of each scale of the corresponding LCD screen. Based on a pre-trained adhesive layer reflection mapping model, and combined with adhesive layer reflection sub-signals at each scale of each LCD display in a specified production batch, the characteristic values ​​of adhesive layer reflection defects in the LCD in the specified production batch are analyzed. The LCD bonding conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value of each LCD display in the set production batch are subjected to feature normalization mapping processing. Based on the LCD bonding conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value after feature normalization mapping, a joint production risk early warning process is performed on a set production batch.

2. The LCD display production risk early warning method based on data analysis according to claim 1, characterized in that, The process execution sequence data includes the pressing temperature, pressing pressure, interfacial contact resistance, heat flux density, adhesive layer stress feedback, and electrostatic potential distribution at each time point. The specific steps to obtain the LCD pressing conduction hindrance characteristic value in the set production batch are as follows: Read the process timing data of each LCD display in the set production batch and perform preprocessing; Based on the preprocessed process timing data of each LCD display in the production batch, the pressing response timing feature set of the corresponding LCD display is analyzed, including the pressing synchronization balance feature value and the interface response coordination feature value at each time point. The pressing response feature set of each LCD display in a set production batch is processed by time-series evolution to obtain the pressing conduction hysteresis feature value of the LCD in the set production batch.

3. The LCD display production risk early warning method based on data analysis according to claim 2, characterized in that, The specific steps for analyzing and setting the pressing response timing characteristic set for each LCD display in a production batch are as follows: Based on the pressing temperature, pressing pressure, and heat flux density of each LCD display in the production batch at each time point, the pressing synchronization equilibrium characteristic value at the corresponding time point is analyzed. Based on the interface contact impedance value, adhesive layer stress feedback value, and electrostatic potential distribution value of each LCD display screen in the production batch at each time point, the interface response coordination characteristic value at the corresponding time point is analyzed.

4. The LCD display production risk early warning method based on data analysis according to claim 2, characterized in that, The specific steps of time-series evolution processing are as follows: Based on the pressing response timing characteristic set of each LCD display in the production batch, the pressing conduction characteristic value at the corresponding time point is analyzed. Based on the compression conduction characteristic value of each LCD display screen at each time point in the set production batch, the conduction extension characteristic value and conduction fluctuation aggregation characteristic value of the corresponding LCD display screen are extracted, and the LCD compression conduction stagnation characteristic value of the corresponding LCD display screen is analyzed. The conduction extension characteristic value is used to characterize the continuous transmission of compression energy and stress in the time series.

5. The LCD display production risk early warning method based on data analysis according to claim 4, characterized in that, The specific steps for analyzing and setting the pressure conduction characteristic values ​​of each LCD display screen at each time point in the production batch are as follows: Read the pressing response timing feature set of each LCD display in the set production batch, extract the average pressing synchronization balance feature and the average interface response coordination feature of the corresponding LCD display, and sum them to obtain the conduction sum value of the corresponding LCD display. The average value of the pressing synchronization balance characteristic and the average value of the interface response coordination characteristic of each LCD display in the production batch are compared with the conduction sum value to obtain the synchronization balance adjustment coefficient and interface coordination adjustment coefficient of the corresponding LCD display. Based on the synchronous balance adjustment coefficient, interface coordination adjustment coefficient, and pressing response timing characteristic set of each LCD display in the production batch, the pressing conduction characteristic values ​​at the corresponding time points are analyzed.

6. The LCD display production risk early warning method based on data analysis according to claim 1, characterized in that, The specific mucosal reflector signal is the mucosal acoustic wave reflection intensity value at each time point of each scale. The mucosal reflection mapping model includes a signal input layer, a feature parsing layer, and a mapping output layer.

7. The LCD display production risk early warning method based on data analysis according to claim 6, characterized in that, The specific steps for analyzing and setting the characteristic values ​​of LCD adhesive layer reflection defects in a production batch are as follows: The sticky reflector signal of each LCD display at each scale in the production batch is input into the pre-trained sticky reflector mapping model to extract the internal reflection sensing feature set at the corresponding scale, including reflection propagation distortion feature value, reflection frequency offset distortion feature value, and sticky reflector mismatch feature value. Based on the adhesive reflector signal of each LCD display at each scale in a given production batch, analyze the reflection weight contribution value at the corresponding scale. Based on the reflection weight contribution value and internal reflection perception feature set of each LCD display in the set production batch at each scale, the characteristic value of LCD adhesive layer reflection defect in the set production batch is analyzed.

8. The LCD display production risk early warning method based on data analysis according to claim 7, characterized in that, The specific steps for extracting the internal reflection sensing feature set of each LCD display at each scale in a given production batch are as follows: In the signal input layer of the adhesive layer reflection mapping model, the adhesive layer reflector signal at each scale of each LCD display in the set production batch is received and preprocessed. In the feature analysis layer of the viscous reflection mapping model, based on the viscous reflection sub-signals of each LCD display screen in the set production batch after signal preprocessing, the reflection time sequence feature vectors at the corresponding scales are analyzed. In the signal input layer of the adhesive layer reflection mapping model, based on the reflection time sequence feature vector of each LCD display at each scale in the set production batch, the internal reflection sensing feature set at the corresponding scale is output.

9. The LCD display production risk early warning method based on data analysis according to claim 7, characterized in that, The specific steps for analyzing and setting the reflection weight contribution value for each scale of each LCD display in a production batch are as follows: Based on the adhesive reflector signal of each LCD display at each scale in the production batch, analyze the adhesive reflector entropy value at the corresponding scale. Based on the adhesive layer reflection entropy value of each LCD display at each scale in the set production batch, the reflection weight contribution value at the corresponding scale is analyzed.

10. The LCD display production risk early warning method based on data analysis according to claim 1, characterized in that, The specific steps for joint production risk early warning processing of a set production batch based on the LCD lamination conduction hysteresis characteristic value and the LCD adhesive layer reflection defect characteristic value after feature normalization mapping are as follows: The LCD bonding conduction hindrance characteristic value and LCD adhesive layer reflection defect characteristic value in the set production batch after feature normalization mapping are judged and analyzed with several preset risk warning intervals respectively. Based on the judgment and analysis results, corresponding production risk warning measures are taken for the designated production batches.

Images