UV adhesive detection data performance analysis method and system based on machine learning

Abstract

The invention relates to the technical field of UV adhesive detection and analysis, and discloses a UV adhesive detection data performance analysis method and system based on machine learning. The method comprises the following steps: acquiring optical response time sequence data of a curing process from a multi-dimensional optical sensing array of a UV adhesive production device, and synchronously acquiring real-time working condition parameters of the device; in a multi-layer feature fusion mode, performing noise compensation processing on the optical response time sequence data to obtain a spectral attribute index of a curing reaction, and determining a quality fluctuation period of quality tracking in the continuous curing stage of the UV adhesive according to the spectral attribute index; production line process path information is obtained, defect characteristics are analyzed, interlayer interface defect data in the curing process are obtained, and the defect coupling level is determined in combination with optical response time sequence data; according to the quality fluctuation period and the defect coupling level, early warning is conducted on a dynamic performance attenuation area in the UV glue production process. According to the method, comprehensive monitoring and accurate analysis of the curing process can be realized, and a technical scheme is provided for production quality control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of UV glue detection and analysis, in particular to a UV glue detection data performance analysis method and system based on machine learning. BACKGROUND

[0002] In the production process of UV glue, the curing quality is directly related to the application performance of the final product, and effective monitoring and performance analysis of the curing process are important means to ensure the stability of the quality of UV glue products. At present, the detection methods for the curing process of UV glue in the industry mainly rely on single-dimensional data collection, such as only using optical sensors to obtain light response signals at a certain wavelength, or only relying on temperature sensors to monitor the curing environment temperature. This single-dimensional data collection method cannot fully reflect the complex physical and chemical changes in the curing process of UV glue, making it difficult to accurately capture key feature information in the curing process. When processing the collected data, the existing technology often lacks effective noise compensation mechanisms. There are many interference factors in the production site of UV glue, such as equipment operation vibration, electromagnetic radiation, and changes in ambient light, etc. These factors will cause a large amount of noise to be mixed into the collected optical response time series data. If these noises cannot be effectively processed, it will cause large deviations in subsequent data-based spectral attribute index calculations, and thus affect the accuracy of judging the quality fluctuation period of UV glue. In terms of defect recognition, the existing technology usually only focuses on the appearance defects of UV glue after curing, such as surface scratches and bubbles, while ignoring the formation and development of interlayer interface defects during the curing process. The formation of interlayer interface defects is closely related to the curing rate of UV glue, process parameter settings, material properties, and other factors, and these defects are often hidden and difficult to detect through appearance detection. However, interlayer interface defects can seriously affect the mechanical properties, adhesive strength, and aging resistance of UV glue products, leading to premature failure of the products during use. The existing technology lacks effective early warning mechanisms for dynamic performance decay zones in the production process of UV glue. The curing process of UV glue is a dynamic process, and its performance changes accordingly at different curing stages. When process parameters deviate or raw material quality fluctuates during production, it may cause UV glue to enter the dynamic performance decay zone during the curing process, thereby affecting product quality. Due to the lack of effective early warning means, production personnel often cannot timely discover these problems, and can only discover quality defects during product out-of-factory testing, which not only causes a large amount of raw material waste and increases production costs, but also affects the production efficiency and market reputation of the enterprise. SUMMARY

[0003] The present application aims to provide a machine learning-based UV glue detection data performance analysis method and system to solve the problems raised in the background art.

[0004] To achieve the above-mentioned purpose, the present application provides a machine learning-based UV glue detection data performance analysis method, which comprises: Collecting optical response time series data of the curing process from a multi-dimensional optical sensing array of a UV glue production device, and synchronously acquiring real-time working condition parameters of the UV glue production device; In a multi-layer feature fusion mode, performing noise compensation processing on the optical response time series data to obtain spectral attribute indexes in the curing reaction process, and determining a quality fluctuation period when the UV glue is tracked for quality in the continuous curing stage according to the spectral attribute indexes; Acquiring process path information of the UV glue production line, performing defect feature analysis on the process path information to obtain interlayer interface defect data of the UV glue in the curing process, and determining a defect coupling level in the curing process according to the interlayer interface defect data and the optical response time series data; According to the quality fluctuation period and the defect coupling level, a dynamic performance decay area in the UV glue production process is warned.

[0005] Preferably, the multi-dimensional optical sensing array comprises an ultraviolet light intensity distribution sensor, an infrared spectrum sensor, and a visible light imaging sensor.

[0006] Preferably, determining a quality fluctuation period when the UV glue is tracked for quality in the continuous curing stage according to the spectral attribute indexes specifically comprises: Building a spatiotemporal evolution model of the curing reaction kinetics according to the spectral attribute indexes; Outputting a tracking response boundary of dynamic quality tracking from the spatiotemporal evolution model; Determining the quality fluctuation period when the UV glue is tracked for quality in the continuous curing stage from the tracking response boundary.

[0007] Preferably, performing defect feature analysis on the process path information to obtain interlayer interface defect data of the UV glue in the curing process specifically comprises: Building a spatiotemporal association model of interlayer structure behavior and historical defect events based on the process path information; Outputting a defect transmission intensity of the UV glue in the curing process through the spatiotemporal association model; Determining defect distribution characteristics of each process node in the production line network through the defect transmission intensity; Determining interlayer interface defect data of the UV glue in the curing process according to the defect distribution characteristics.

[0008] Preferably, in the multi-layer feature fusion mode, the optical response time series data is subjected to noise compensation processing to obtain the spectral property index in the curing reaction process, which specifically includes: In the multi-layer feature fusion mode, the energy absorption path of the curing stage is determined according to the optical response time series data. The missing waveband response nodes in the optical response time series data are interpolated and compensated to generate a complete optical event sequence. The optical event sequence is subjected to feature selection to obtain the spectral abnormality information in the curing reaction process. The spectral property index in the curing reaction process is determined from the spectral abnormality information.

[0009] Preferably, the defect coupling level in the curing process is determined according to the interlayer interface defect data and the optical response time series data, which specifically includes: The spatial constraint condition of the defect in the curing process is determined according to the interlayer interface defect data. The time-domain correlation density of the defect in the curing process is determined through the optical response time series data. The defect coupling level in the curing process is determined according to the spatial constraint condition and the time-domain correlation density.

[0010] Preferably, the process path information of the UV glue production line is obtained, which specifically includes: The material state information between process nodes is collected in real time. The core process path of the curing process is determined according to the material state information. The process path information of the UV glue production line is determined through the core process path.

[0011] Preferably, the dynamic performance attenuation area in the UV glue production process is warned according to the quality fluctuation period and the defect coupling level, which specifically includes: The elastic monitoring threshold in the UV glue production process is determined according to the quality fluctuation period. The attenuation risk exposure index in the dynamic performance attenuation area is determined according to the defect coupling level. The elastic monitoring threshold and the attenuation risk exposure index are matched with the warning rules to generate a dynamically adjustable performance attenuation warning map.

[0012] Preferably, the real-time working condition parameters include UV light source intensity, environmental temperature gradient, and conveyor belt speed.

[0013] Preferably, the present application also includes a machine learning-based UV glue detection data performance analysis system, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of a machine learning-based UV glue detection data performance analysis method as described above.

[0014] Compared with the prior art, the present application has the following advantages: The optical response time series data of the curing process is collected from the multi-dimensional optical sensing array of the UV glue production device, and real-time working condition parameters are synchronously obtained, breaking the limitations of traditional single-dimensional data collection. The multi-dimensional optical sensing array can capture the optical response changes at different wavelengths during the UV glue curing process, while the real-time working condition parameters can reflect the key conditions such as temperature, pressure, and UV irradiation intensity of the curing environment. By combining these two types of data, a more comprehensive and detailed presentation of the physical and chemical changes during the UV glue curing process can be achieved, providing a richer and more reliable data foundation for subsequent performance analysis, which helps to more accurately grasp the key characteristics of the curing process and avoid analysis bias caused by incomplete data information. Noise compensation processing of the optical response time series data in the multi-layer feature fusion mode can effectively reduce the influence of various interference factors in the production site on data quality. The multi-layer feature fusion mode can more accurately identify the noise components in the data by integrating feature information at different levels, and use targeted compensation algorithms to eliminate or suppress noise, thereby obtaining more realistic and accurate spectral attribute indicators. Based on these accurate spectral attribute indicators, the quality fluctuation period of the UV glue in the continuous curing stage can be determined, allowing production personnel to clearly understand the variation of UV glue quality during the curing process and timely grasp the time node of quality fluctuation, facilitating timely adjustment measures when quality fluctuation occurs, and avoiding further expansion of quality problems. Obtaining the process path information of the UV glue production line and analyzing the defect features realizes effective identification of interlayer interface defects in the curing process. The process path information includes various process steps, parameter settings, and material flow in the production process of the UV glue. Through in-depth analysis of these information, the correlation between different process links and the formation of interlayer interface defects can be determined, and then the interlayer interface defect data can be accurately obtained. Combined with the optical response time series data, the defect coupling level in the curing process can be determined, which can clearly judge the mutual influence degree between the interlayer interface defects and other curing process parameters, understand the severity of the defects and the possible influence range on product performance, and thus provide a clear direction for the control of defects in the production process, which helps to reduce the generation of interlayer interface defects from the root cause and improve the internal quality of the UV glue product. According to the quality fluctuation period and the defect coupling level, the dynamic performance attenuation area in the UV glue production process can be warned, so that the real-time monitoring and early intervention of the production process can be realized. Through the analysis of the quality fluctuation period, the possible decline stage of the UV glue quality can be predicted, and the defect coupling level can reflect the risk level of the defect in the current production process. When the system detects that the quality fluctuation period changes abnormally or the defect coupling level reaches the preset threshold, a warning signal can be sent in time to remind the production personnel to pay attention to the relevant production link. The production personnel can quickly investigate the causes of the abnormality according to the warning information, such as adjusting the process parameters, checking the quality of raw materials, and maintaining the production equipment, so that effective measures can be taken before the UV glue enters the dynamic performance attenuation area, the production of unqualified products can be avoided, the waste of raw materials can be reduced, the production cost can be reduced, the stability and continuity of the production process can be ensured, and the production efficiency and product competitiveness of the enterprise can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A working principle diagram of the machine learning-based UV glue detection data performance analysis method is provided. Figure 2 A flowchart for determining the quality fluctuation period is provided. Figure 3 A flowchart for defect feature analysis is provided. Figure 4 A flowchart for noise compensation processing is provided. Figure 5 A flowchart for determining the defect coupling level is provided. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] Please refer to Figure 1 The present application provides a machine learning-based UV glue detection data performance analysis method, which comprises: The optical response time series data of the curing process is collected from the multi-dimensional optical sensing array of the UV glue production device, and the real-time working condition parameters of the UV glue production device are synchronously acquired, including the UV light source intensity, the environmental temperature gradient, and the conveying belt speed. In the multi-layer feature fusion mode, the optical response time series data is processed for noise compensation to obtain the spectral attribute index in the curing reaction process, and accordingly the quality fluctuation period of the UV glue in the continuous curing stage is determined for quality tracking. At the same time, the process path information of the UV glue production line is acquired, and the defect feature analysis is performed to obtain the interlayer interface defect data in the curing process, and the defect coupling level in the curing process is determined in combination with the optical response time series data. Finally, the dynamic performance attenuation zone in the UV glue production process is warned according to the quality fluctuation period and the defect coupling level.

[0018] Embodiment 1: refer to Figure 2 , which relates to the whole process of collecting data from the multi-dimensional optical sensing array of the UV glue production device and analyzing and processing. The multi-dimensional optical sensing array is composed of an ultraviolet light intensity distribution sensor, an infrared spectrum sensor, and a visible light imaging sensor. The ultraviolet light intensity distribution sensor is arranged in an array around the UV light source to monitor the ultraviolet radiation intensity at different spatial positions at a high sampling frequency, and record the uniformity of the light intensity distribution and the fluctuation over time. The infrared spectrum sensor uses Fourier transform infrared spectrum technology to real-time collect the infrared absorption spectrum of the curing area, monitor the change of molecular bond vibration frequency, and pay special attention to the change of absorption peaks of carbon-carbon double bond, epoxy group and other characteristic functional groups. The visible light imaging sensor is equipped with a high-resolution industrial camera and appropriate optical filters to capture surface images of the curing area at a fixed frame rate, and record morphological changes such as glue flow, bubble formation and surface shrinkage. These sensors realize synchronous data acquisition through a unified clock signal, ensuring that all optical response data have consistent time stamps.

[0019] In the data acquisition process, the real-time working condition parameters of the UV glue production device are synchronously acquired. The UV light source intensity is monitored in real time by a luminosity probe installed in the light box to record the fluctuation of the light source output power. The environmental temperature gradient is measured by thermocouple sensors arranged in multiple partitions of the curing furnace to record the temperature distribution changes along the conveying belt direction and the vertical direction. The conveying belt speed is monitored in real time by an encoder to record the slight changes in transmission rate. All these working condition parameters are transmitted to the central control system through the industrial field bus to establish a time synchronization relationship with the optical sensing data. The multi-dimensional optical response time series data collected is first preprocessed. The ultraviolet light intensity data is normalized to eliminate the baseline differences between sensors. The infrared spectrum data is background subtracted and atmosphere compensated to eliminate the interference absorption of environmental carbon dioxide and water vapor. The visible light image sequence is brightness corrected and contrast enhanced to improve feature visibility. The preprocessed multi-source optical data is aligned and fused in the time dimension to form a multi-dimensional optical response time series data set.

[0020] Based on the fused optical data, spectral property indicators are extracted, including time-varying sequences of ultraviolet absorption rate, infrared characteristic peak intensity ratio, and visible light image texture features. A spatiotemporal evolution model of the curing reaction kinetics is constructed using these spectral property indicators. The model uses a deep learning method with a hybrid architecture of convolutional neural networks and recurrent neural networks. The convolutional layers process spatially distributed optical data to extract local region feature patterns, while the recurrent layers process time series data to capture the dynamic evolution law of the curing reaction. The model input is multi-time multi-dimensional optical data, and the output is the spatial distribution prediction of the curing degree at future time points.

[0021] The tracking response boundary for dynamic quality tracking is extracted from the trained spatiotemporal evolution model, which is determined by analyzing the model's internal attention weight distribution, identifying the spatial regions and time nodes most sensitive to curing quality changes. The tracking response boundary is represented as a set of characteristic curves, describing the trajectory of key parameters during the curing reaction process. Based on the tracking response boundary, the quality fluctuation period of the UV glue during continuous curing is further determined. Time-frequency analysis is performed on the boundary curves, and frequency spectrum analysis is used to identify periodic components. The fluctuation period represents the regular fluctuations of the curing quality parameters within a certain time range, reflecting systematic fluctuation factors in the production process. These periodic information is used for subsequent quality stability evaluation and early warning judgment.

[0022] Throughout the implementation process, the collection and processing of multi-dimensional optical sensing data, the synchronous monitoring of real-time working conditions, the construction and application of spatiotemporal evolution models, the extraction of tracking response boundaries, and the determination of quality fluctuation periods form a complete technical chain. In this way, comprehensive monitoring and in-depth analysis of the UV glue curing process are achieved, providing necessary data support and analysis foundation for subsequent quality early warning. All data processing and analysis algorithms are based on machine learning methods, which can adaptively learn the curing behavior characteristics under different production conditions without relying on pre-set fixed thresholds or rules, and have strong applicability and flexibility.

[0023] Example 2: Referring to Figure 3 , involves the acquisition and defect feature analysis of UV glue production line process path information. Process path information is derived from the manufacturing execution system and contains complete production process data, specifically recording the equipment number, process parameter setting value, material batch code, timestamp sequence, and environmental control record at each process node. These data are stored in structured and unstructured forms and need to be extracted and integrated through data interfaces.

[0024] Based on the acquired process path information, a spatio-temporal correlation model of interlayer structure behavior and historical defect events is established. This model uses a graph neural network architecture, modeling each process node as a node in the graph. The node attributes include device type, process parameters, processing time, and historical defect records related to the node. The connection edges between nodes represent material flow paths, and the edge weights are calculated based on actual transmission frequency and historical defect correlation strength. The spatio-temporal dimension is realized by adding time series features to each node, recording the production state changes of the node at different time periods.

[0025] During model training, historical production data is used as the training set, including normal production records and known defect event records. The graph neural network uses a message passing mechanism to aggregate the feature information of adjacent nodes for each node, learning the pattern of defect propagation in the process network. The trained model can output the defect propagation intensity of UV glue during the curing process, which is a quantitative indicator reflecting the degree of defect generation or propagation tendency of a specific process node under current production conditions. By calculating the defect propagation intensity, the defect distribution characteristics of each process node in the production line network are further analyzed. This process involves statistical analysis and spatial mapping of the intensity values of all network nodes, identifying key nodes and node clusters with high defect propagation intensity. At the same time, the propagation path of defect intensity in the process network is analyzed to identify potential defect sources or defect amplifiers.

[0026] According to the defect distribution characteristics, interfacial defect data of UV glue during the curing process is determined. These data are represented in the form of a multi-dimensional vector, containing defect type encoding, defect quantity statistics, defect spatial position coordinates, and defect severity rating. During the generation of interfacial defect data, the actual process parameters and real-time monitoring data of the current production batch are considered to dynamically adjust the defect prediction results.

[0027] After each production run, new production data and quality test results are fed back into the system for updating the historical defect database and optimizing the spatio-temporal correlation model. This iterative learning mechanism enables the system to adapt to changes in production processes, continuously improving the accuracy of defect prediction. In specific operations, the acquisition of process path information requires deep integration with existing production management systems to ensure real-time access to the latest production state data. The defect feature analysis process needs to run on a computing cluster to handle large-scale process network data. The generated interfacial defect data will be transmitted to the quality monitoring system to provide input for subsequent quality analysis and early warning decision-making.

[0028] The embodiment is characterized by connecting discrete process nodes into a network, capturing the propagation law of defects in the process network by using a graph neural network, thereby realizing accurate prediction of interlayer interface defects. This method not only considers the state of a single process node, but more importantly, analyzes the mutual influence between nodes and the defect propagation path, providing an effective means for comprehensively mastering the defect generation mechanism in the solidification process.

[0029] Embodiment 3: see Figure 4 , focusing on noise compensation processing of optical response time series data in a multi-layer feature fusion mode, and finally generating spectral attribute indicators in the solidification reaction process. The process starts with the raw data stream collected by the multi-dimensional optical sensing array, including time series signals of three spectral channels of ultraviolet, infrared and visible light. The ultraviolet channel records the radiation intensity in the wavelength range of 320-400 nm, the infrared channel covers the mid-infrared spectral region of 4000-400 cm⁻¹, and the visible light channel collects image sequences in the 400-700 nm band. These raw data carry device noise, environmental interference and transmission errors, which need to be systematically processed.

[0030] The multi-layer feature fusion mode is realized by a cascaded deep autoencoder architecture, which includes three layers of encoders and three layers of decoders, each layer consisting of a convolution module and a long short-term memory unit. The first layer of encoder processes ultraviolet spectral data, and the convolution kernel slides along the wavelength dimension to extract local features, and the long short-term memory unit captures time dependence. The second layer of encoder processes infrared spectrum, and different sizes of convolution kernel combination are used to identify feature absorption peaks. The third layer of encoder processes visible light images, and extracts texture features through convolutional neural network. The outputs of each layer of encoder are spliced in the feature space to form a fusion feature tensor. The decoder reconstructs the multi-source signal through deconvolution and upsampling operation, and the training target is to minimize the difference between the reconstructed signal and the original signal.

[0031] According to the fused optical response time series data, the energy absorption path in the solidification stage is determined. The path is established by analyzing the absorption depth change mode of the ultraviolet spectrum, and specifically represented as the trajectory of the evolution of the absorption peak intensity of a specific wavelength over time. The identification of the energy absorption path uses the dynamic time warping algorithm to align the spectral change sequence of the current production batch with the standard solidification template and calculate the path offset. The band region with an offset greater than the threshold is marked as an abnormal absorption area, which reflects the abnormal energy transfer of the solidification reaction. The missing band response nodes in the optical response time series data are interpolated and compensated. When the sensor has a transient failure or data transmission packet loss, a gap in the spectral data is formed. A time series prediction model based on long short-term memory network is used to fill in the gap, and its calculation process is represented as:

[0032] wherein: denotes the interpolated prediction value at time t, denotes a long short-term memory network function, denotes a long short-term memory network function, denotes the observation values at the first k time points, denotes the working condition parameter vector at time t. The model is pre-trained on the complete data set to learn the spatio-temporal evolution rule of the spectral sequence. The interpolation process is divided into two steps: first, the local spectral profile is reconstructed using the data of adjacent nodes, and then the profile shape is adjusted according to the energy absorption path.

[0033] After generating the complete optical event sequence, feature selection is performed to extract key information. A recursive feature elimination algorithm is used, with the solidification reaction completion degree as the target variable. The algorithm first trains a support vector machine model using all spectral bands, and then iteratively removes the band with the smallest weight until the prediction accuracy of the remaining band subset on the validation set begins to decline. The final remaining band constitutes the feature spectral set, containing 15-20 most discriminative characteristic wavelengths.

[0034] Based on the selected feature spectral set, spectral abnormality information during the solidification reaction process is calculated. This information is obtained by monitoring the dynamic behavior of the feature bands: for each characteristic wavelength, the coefficient of variation (ratio of standard deviation to mean) of its intensity sequence is calculated, and the time points where the coefficient of variation exceeds twice the standard deviation of the historical baseline are recorded. The cooperative change relationship between characteristic wavelengths is also analyzed, and when asynchronous phenomena occur in wavelength pairs that should change synchronously, they are marked as cooperative abnormal events. The spectral abnormality information is finally quantified into three-dimensional indicators: abnormal event frequency, abnormal duration proportion, and abnormal wavelength impact weight.

[0035] The process of determining the spectral attribute indicators from the spectral abnormality information uses a fuzzy inference system. The system input is the aforementioned three abnormality dimension data, and the output is three spectral attribute indicators: absorption rate change gradient, spectral shift, and peak stability coefficient. The absorption rate change gradient describes the rate of change of the absorption intensity of the characteristic wavelength per unit time, reflecting the solidification reaction rate. The spectral shift measures the degree of drift of the characteristic wavelength position relative to the standard position, indicating molecular structure changes. The peak stability coefficient represents the degree of preservation of the characteristic peak shape, obtained by calculating the correlation coefficient matrix of consecutive spectra. These three indicators constitute the spectral fingerprint describing the state of the solidification reaction.

[0036] The raw data is collected and stored in a time series database. The noise compensation module is deployed on a GPU computing node, and feature selection and anomaly detection run on a memory computing cluster. The processed spectral attribute indicators are updated at intervals of every 5 seconds and transmitted to the quality analysis system through a message queue. The system has a built-in self-calibration mechanism that uses standard sample data to recalibrate the feature wavelength selection model and the fuzzy inference rule base every week to maintain the accuracy of the analysis results. This implementation deeply integrates multi-source optical information and working condition parameters to achieve a fine characterization of the solidification reaction process.

[0037] Example 4: Refer to Figure 5 , which involves determining defect coupling levels during the curing process based on interlayer interface defect data and optical response temporal data, and obtaining process path information of the UV glue production line. The implementation process starts with the processing of interlayer interface defect data, which is derived from the output of Example 2, containing defect type codes, spatial location coordinates, and severity ratings. The spatial constraints of the defects are determined by analyzing the spatial distribution characteristics of the defect clusters. Spatial clustering algorithms are used to identify the aggregation areas of the defects, and the statistical distribution of the density radius and defect point spacing of each aggregation area is calculated. The spatial constraints are specifically manifested as the geometric characteristics of the defect distribution, including the maximum diameter of the aggregation area, the average nearest neighbor distance of the defect points, and the directional preference of the defect distribution. These features collectively describe the distribution pattern and constraint range of the defects in three-dimensional space.

[0038] The time-domain correlation density of defects during the curing process is determined by optical response temporal data, which includes the temporal signals of ultraviolet, infrared, and visible light channels collected at a frequency of 10 times per second. The calculation of time-domain correlation density is based on the time synchronization analysis of defect events and optical response events. The correspondence between defect occurrence time and optical signal change time is established, and the co-occurrence probability of optical signal anomalies and defect detection within a specific time window is calculated. The time-domain correlation density quantifies the time correlation between optical signal changes and defect formation, represented as a value between 0 and 1, with a higher value indicating a stronger indication of optical signals on defects.

[0039] The defect coupling level is determined based on the spatial constraint condition and the time-domain correlation density, which uses a rule-based rating system to quantify the spatial constraint strength index and convert the time-domain correlation density into a time-domain correlation degree index. The defect coupling level is divided into five levels, from level one (lowest coupling) to level five (highest coupling). The rating rules consider indicators in both spatial and temporal dimensions, and when the spatial constraint strength is high and the time-domain correlation degree is high, it is determined as a high-level defect coupling.

[0040] The process path information of the UV glue production line is obtained by real-time collection of material state information between process nodes. Multiple material state monitoring points are arranged on the production line, and each monitoring point collects parameters such as viscosity, temperature, and flow rate. The distance between monitoring points is set according to the complexity of the process, and the monitoring density is usually increased before and after the key process nodes. Material state information is collected at a frequency of 1 time per second and transmitted to the central database through industrial Internet of Things.

[0041] Table 1: Process node material state monitoring parameters.

[0042] Monitoring point number Node position Viscosity range (cP) Temperature range (°C) Flow rate range (cm / s) Sampling frequency (Hz) MP-001 Feed inlet 1200-1500 25-28 15-18 1 MP-002 Premix zone 950-1200 28-32 12-15 1 MP-003 Main reaction zone 800-950 35-40 10-12 2 MP-004 Curing inlet 700-850 40-45 8-10 2 MP-005 Curing zone 600-750 45-50 5-8 5 MP-006 Outlet zone 550-700 25-30 3-5 1 The core process path of the curing process is determined according to the material state information. The clustering analysis method is used to identify the characteristic mode of the material state change. Process nodes with similar state change rules are divided into the same process stage. The core process path is determined by analyzing the state transition probability between each process stage. The path with the most frequent state transition and the most continuous material parameter change is selected as the core process path. This path reflects the main processing stages and state evolution trajectory of the material in the curing process. The process path information of the UV glue production line is determined through the core process path. The process path information is represented in the form of a directed graph. The nodes represent process equipment or processing units, and the edges represent material transfer paths. Each edge is assigned a weight value, which is calculated based on the throughput and quality stability of the material on that path. The process path information also includes metadata such as the processing parameter range, allowed residence time, and environmental control requirements of each node. These information together constitute a complete process path description, providing a basic framework for defect analysis and quality tracking.

[0043] The entire implementation process relies on real-time data acquisition and distributed computing platform. The material state monitoring data is preliminarily processed by the edge computing node, and the feature parameters are extracted and transmitted to the central server. The calculation of defect coupling level adopts the stream processing architecture, which receives the interfacial defect data and optical response data in real time, and outputs the dynamically updated defect coupling level. The generation of process path information adopts batch processing mode, which analyzes and updates the data every hour to ensure that the path information reflects the latest production state. The system establishes a data quality check mechanism to automatically mark and exclude abnormal monitoring data, ensuring the reliability of the analysis results. During the implementation process, attention is paid to the time synchronization of each data source. All monitoring equipment uses a unified network time protocol clock source to ensure the time consistency of cross-platform data.

[0044] Example 5: Involves early warning of dynamic performance attenuation zone in UV glue production process according to quality fluctuation period and defect coupling level. The implementation process starts with the input data of the quality fluctuation period, which is derived from the output results of Example 1, representing the periodic fluctuation characteristics of the curing quality parameters over time. The elastic monitoring threshold is determined according to the quality fluctuation period. The calculation of this threshold is based on the amplitude envelope analysis of the periodic signal. The specific process is as follows: extract the main frequency components in the quality fluctuation period, calculate the amplitude variation curve of each frequency component on the time axis; smooth the curve to generate the amplitude envelope; the moving average of the envelope is taken as the basic threshold, and the standard deviation of the envelope fluctuation range is added as a dynamic adjustment item. The elastic monitoring threshold is dynamically updated with the production batch. When the production process parameters change, the threshold recalculation process is automatically started.

[0045] The defect coupling level data is derived from the output results of Example 4, including coupling degree ratings from level one to level five. According to the defect coupling level, an attenuation risk exposure index in the dynamic performance attenuation zone is determined, and the index is calculated in combination with a historical failure database. A mapping relationship between the defect coupling level and the historical failure severity is established, and the failure probability distribution corresponding to different coupling levels is learned by a machine learning model. The attenuation risk exposure index is quantified on a percentage basis, and a time attenuation factor is introduced in the index calculation, so that recent defect events have a higher weight. At the same time, the spatial aggregation effect of defects is considered, and the multiple defects occurring in the same physical region are calculated by superimposing the index.

[0046] The elastic monitoring threshold and the attenuation risk exposure index are matched with the early warning rules, and the early warning rule library includes multiple judgment logics, each of which defines the combination conditions of the threshold and the index and the corresponding early warning level. The core rule is: when the attenuation risk exposure index exceeds 120% of the elastic monitoring threshold for three minutes, level one early warning is triggered; when the index exceeds 150% of the threshold and there are spatially aggregated defects, level two early warning is triggered; when the index breaks through 200% of the threshold and is accompanied by a sharp decline in quality parameters, level three early warning is triggered. The rule matching engine uses a real-time stream processing architecture, which scans the rules for all monitoring points in the production line every second.

[0047] A dynamically adjustable performance attenuation early warning map is generated, which is constructed based on the digital twin model of the production line and divides the physical production area into grid units. Each unit calculates the early warning state according to the monitoring point data it contains, and the state is divided into three levels: normal, attention, and early warning. The map rendering uses a heat map form, with normal areas showing green, attention areas showing yellow, and early warning areas showing orange to red according to the early warning level. The map update frequency is synchronized with data collection, and the visualization state is refreshed every second. The early warning map supports a dynamic adjustment mechanism: when the operator confirms a false alarm, the system automatically reduces the sensitivity of the region; when a new monitoring point is added, the map automatically expands the coverage area. Historical early warning data forms a spatiotemporal distribution cloud map, which is used to identify recurring attenuation hotspots.

[0048] The early warning information transmission adopts a hierarchical push strategy, with level one early warning displayed on the workshop board with a yellow warning; level two early warning triggers a sound and light alarm on the post terminal; and level three early warning is pushed to the mobile management terminal and starts the automatic deceleration program simultaneously. All early warning events generate complete event reports containing time stamps, location coordinates, attenuation parameters, associated defects, and other elements, which are stored in the quality analysis database. The system sets up an early warning feedback loop, requiring operators to label the disposal of each early warning, and the labeling information is used to optimize the early warning rules.

[0049] The system is deployed in an industrial cloud platform, including a data acquisition layer, a real-time calculation layer and a visualization layer. The data acquisition layer obtains real-time data of equipment through OPC-UA protocol; the real-time calculation layer uses an in-memory database to process flow data; and the visualization layer realizes rendering of a three-dimensional early warning atlas through WebGL technology. The system sets a simulation operation mode, and can test the effectiveness of early warning rules without affecting production. The system performs self-checking every week to verify the data channel integrity and the response speed of the rule engine. The implementation mode fuses the periodic quality fluctuation characteristics and the defect coupling state to build a dynamic early warning network covering the whole production line.

[0050] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, alternatives and variations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing performance of UV glue detection data based on machine learning, characterized in that, The performance analysis method comprises the following steps: Collecting optical response time series data of the curing process from a multi-dimensional optical sensing array of the UV glue production device, and synchronously acquiring real-time working condition parameters of the UV glue production device; In a multi-layer feature fusion mode, performing noise compensation processing on the optical response time series data to obtain spectral attribute indexes in the curing reaction process, and determining a quality fluctuation period when the UV glue is tracked in quality in the continuous curing stage according to the spectral attribute indexes; Acquiring process path information of the UV glue production line, performing defect feature analysis on the process path information to obtain interlayer interface defect data of the UV glue in the curing process, and determining a defect coupling level in the curing process according to the interlayer interface defect data and the optical response time series data; According to the quality fluctuation period and the defect coupling level, a dynamic performance attenuation area in the UV glue production process is warned.

2. The method of claim 1, wherein the method is based on machine learning. The multi-dimensional optical sensing array comprises an ultraviolet light intensity distribution sensor, an infrared spectrum sensor and a visible light imaging sensor.

3. The method for performance analysis of UV adhesive detection data based on machine learning as described in claim 1, characterized in that, According to the spectral attribute indexes, the quality fluctuation period when the UV glue is tracked in quality in the continuous curing stage is determined, specifically comprising: According to the spectral attribute indexes, a time-space evolution model of the curing reaction kinetics is constructed; From the time-space evolution model, a tracking response boundary of dynamic quality tracking is outputted; From the tracking response boundary, the quality fluctuation period when the UV glue is tracked in quality in the continuous curing stage is determined.

4. The method of claim 1, wherein the method is based on machine learning. The defect feature analysis on the process path information to obtain the interlayer interface defect data of the UV glue in the curing process specifically comprises: Based on the process path information, a time-space association model of interlayer structure behavior and historical defect events is established; Through the time-space association model, a defect transmission intensity of the UV glue in the curing process is outputted; Through the defect transmission intensity, defect distribution characteristics of each process node in the production line network are determined; According to the defect distribution characteristics, the interlayer interface defect data of the UV glue in the curing process is determined.

5. The machine learning-based UV glue inspection data performance analysis method of claim 1, wherein, In the multi-layer feature fusion mode, the noise compensation processing on the optical response time series data to obtain the spectral attribute indexes in the curing reaction process specifically comprises: In the multi-layer feature fusion mode, according to the optical response time series data, an energy absorption path in the curing stage is determined; The missing waveband response nodes in the optical response time series data are interpolated and compensated to generate a complete optical event sequence; The optical event sequence is subjected to feature selection to obtain spectral abnormality information in the curing reaction process; From the spectral abnormality information, the spectral attribute indexes in the curing reaction process are determined.

6. The machine learning-based UV glue inspection data performance analysis method of claim 1, wherein, According to the interlayer interface defect data and the optical response time series data, the defect coupling level in the curing process is determined, specifically comprising: According to the interlayer interface defect data, a spatial constraint condition of the defect in the curing process is determined; Through the optical response time series data, a time-domain correlation density of the defect in the curing process is determined; According to the spatial constraint condition and the time-domain correlation density, the defect coupling level in the curing process is determined.

7. The machine learning-based UV glue inspection data performance analysis method of claim 1, wherein, Acquiring the process path information of the UV glue production line specifically comprises: Real-time acquisition of material state information between process nodes; determine a core process path of a curing process according to the material state information; determine process path information of a UV glue production line through the core process path.

8. The machine learning-based UV glue inspection data performance analysis method of claim 1, wherein, The pre-warning of the dynamic performance attenuation area in the UV glue production process according to the quality fluctuation period and the defect coupling level specifically includes: determining an elastic monitoring threshold in the UV glue production process according to the quality fluctuation period; determining an attenuation risk exposure index in the dynamic performance attenuation area according to the defect coupling level; matching the elastic monitoring threshold and the attenuation risk exposure index according to a pre-warning rule to generate a dynamically adjustable performance attenuation pre-warning atlas.

9. The machine learning-based UV glue detection data performance analysis method of claim 1, wherein, The real-time working condition parameters include UV light source intensity, environmental temperature gradient, and conveyor belt speed. 10.A machine learning based UV glue detection data performance analysis system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, The processor, when executing the computer program, implements the steps of the UV glue detection data performance analysis method based on machine learning in any one of claims 1 to 9.

Images