LED module online monitoring and life prediction method

Abstract

The application relates to an LED module online monitoring and life prediction method in the technical field of intelligent manufacturing, which comprises the following steps: according to obtained frequency domain impedance curve feature vectors, adopting principal component analysis to extract dominant components, and determining a low-dimensional feature set reflecting the degradation of multiple components; obtaining associated phosphor degradation sub-features from a chip aging type failure mode result obtained through judgment, adopting a clustering analysis method to group the sub-features to determine the phosphor aging degree; according to the determined phosphor aging degree, fusing solder joint crack related impedance phase offset data to obtain a comprehensive multiple component failure index; if the comprehensive multiple component failure index shows a deviation trend, then analyzing the index sequence through a time series prediction model to obtain an estimated value of the remaining life of the LED module; generating health state report data from the obtained estimated value of the remaining life, and judging the overall reliability level to support online monitoring decisions.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method for online monitoring and lifespan prediction of LED modules. Background Technology

[0002] As a core technology in modern lighting and display, the reliability of light-emitting diodes (LEDs) directly impacts product performance and user experience. This is especially true in demanding applications such as smart lighting, automotive lighting, and medical devices, where ensuring the long-term stable operation of LED modules is crucial. However, existing reliability assessment methods have significant shortcomings, failing to meet the real-time and accuracy requirements of practical applications. This makes online monitoring of LED health status a pressing challenge for the industry.

[0003] Traditional methods often rely on accelerated aging tests to simulate extreme environments and assess LED lifespan. This approach is not only time-consuming and costly but also destructive to samples, making it unsuitable for use in field-operated equipment. Some online monitoring technologies attempt to determine LED status by measuring luminous flux or color temperature changes, but these parameters often only show significant changes when the equipment is nearing failure, making early warning difficult. In contrast, monitoring methods based on electrical parameters have gained attention due to their simplicity, but they typically focus on a single parameter, such as forward voltage, lacking in-depth analysis of complex failure mechanisms and making it difficult to accurately distinguish the degradation problems of different components. LED module failures often involve multiple components, such as chips, phosphors, and solder joints, whose degradation processes during operation are intertwined, increasing monitoring complexity. The core technical challenge lies in capturing comprehensive information reflecting the state of all components non-destructively. Impedance characteristics, as an indicator that characterizes the electrical behavior of materials and structures, can reveal minute changes in various parts of the LED. However, the complexity of impedance information makes extracting key features from large amounts of data challenging. The difficulty in feature extraction further leads to the inability to accurately distinguish between different failure modes, such as the difference between chip aging and solder joint cracks, which directly affects the accurate judgment of the health status of the equipment.

[0004] Therefore, how to obtain and analyze impedance information reflecting the state of multiple components during normal LED operation using non-destructive methods, and thus accurately distinguish different failure modes, has become a key issue in online prediction of LED health status and lifetime assessment. For example, in intelligent lighting systems, failure to detect solder joint cracks or phosphor aging in a timely manner may lead to light flickering or color temperature shifts, affecting user experience and system reliability.

[0005] This problem not only requires monitoring technology to capture subtle changes in complex environments, but also to correlate these changes with specific failure mechanisms, thereby providing a reliable basis for maintenance decisions. The lack of effective feature extraction and analysis methods makes it difficult for existing technologies to meet the needs of early warning and accurate diagnosis in practical applications, necessitating innovative solutions to overcome this bottleneck. Summary of the Invention

[0006] This invention provides a method for online monitoring and lifespan prediction of LED modules, mainly including the following steps:

[0007] Impedance characteristic data is acquired from the operating LED module using sensors. This data is then processed to generate a frequency domain impedance curve feature vector. Dominant components are extracted from this feature vector to form a low-dimensional feature set reflecting the degradation of multiple components. A classifier is used to perform pattern recognition on this low-dimensional feature set to determine the existence of specific failure modes. Relevant sub-features are extracted from the failure mode results and analyzed to determine the degree of component aging. The degree of component aging is fused with relevant impedance data to generate a comprehensive multi-component failure index. Based on the trend of this comprehensive multi-component failure index, a predictive model is used to obtain an estimate of the remaining lifespan of the LED module. A health status report is generated based on the remaining lifespan estimate to assess the overall reliability level and support monitoring decisions.

[0008] Furthermore, the step of acquiring impedance characteristic data from the operating LED module via sensors and processing the impedance characteristic data to generate a frequency domain impedance curve feature vector includes: acquiring impedance characteristic data of the operating LED module using a non-destructive sensor to obtain a raw time-domain signal; processing the raw time-domain signal using Fourier transform to generate a frequency domain impedance curve, which includes the impedance amplitude and phase value at each frequency point; extracting a feature vector from the frequency domain impedance curve to determine the curve peak and phase shift, whereby the curve peak is obtained by identifying the point of maximum amplitude and the phase shift is obtained by comparing the actual phase with a preset reference phase difference; calculating the module aging index based on the curve peak and the phase shift, whereby the module aging index is obtained by a weighted average of the peak decay rate and the shift angle; acquiring the brightness decay data and temperature change data of the LED module, and combining them with the module aging index to calculate a comprehensive evaluation value using a linear regression method; and generating a fault warning signal if the comprehensive evaluation value exceeds a preset threshold.

[0009] Furthermore, the step of extracting the dominant component based on the feature vector of the frequency domain impedance curve to form a low-dimensional feature set reflecting the degradation of multiple components includes: acquiring the frequency domain impedance curves of multiple components and collecting the original data points of the frequency domain impedance curves; processing the original data points using Fourier transform to generate feature vectors; extracting the dominant component based on the feature vectors using principal component analysis, determining the variance contribution rate corresponding to the dominant component by calculating the covariance matrix of the feature vectors and performing eigenvalue decomposition; if the variance contribution rate exceeds a preset threshold, retaining the dominant component to form a preliminary low-dimensional feature set; obtaining degradation reflection features from historical data of multi-component degradation, determining the matching degree by calculating the cosine similarity between the preliminary low-dimensional feature set and the degradation reflection features, and obtaining an adjusted low-dimensional feature set; and obtaining verification samples based on the adjusted low-dimensional feature set and combined with the real-time input of component status monitoring to determine the final low-dimensional feature set reflecting the degradation of multiple components.

[0010] Furthermore, the step of performing pattern recognition on the low-dimensional feature set using a classifier to determine whether a specific failure mode exists includes: acquiring a low-dimensional feature set of chip aging data, wherein the low-dimensional feature set contains quantized values ​​of multiple aging features; calculating an extended index of the chip aging mode using the low-dimensional feature set to obtain a preliminary mapping for aging type identification; for the preliminary mapping, using a threshold exceedance detection process as the dominant component to determine the association vector for feature set classification; extracting control samples as verification data based on the association vector; if the verification data meets preset conditions, determining the failure mode in the chip aging mode using the verification data to obtain a failure mode discrimination result; and updating the feature weights of the low-dimensional feature set based on the failure mode discrimination result to optimize the mapping accuracy of aging type identification.

[0011] Furthermore, the step of extracting relevant sub-features from the failure mode results and analyzing the sub-features to determine the aging degree of the component includes: obtaining phosphor degradation sub-features from the chip aging type failure mode results, wherein the sub-features reflect the characteristic changes of the phosphor during the aging process; grouping the phosphor degradation sub-features using a clustering method, and obtaining a preliminary group set by iteratively calculating the center points; calculating the distribution density based on the preliminary group set, and if the distribution density exceeds a preset threshold, determining a high degradation sub-feature group, otherwise determining a low degradation sub-feature group; extracting spectral attenuation parameters from the high degradation sub-feature group or the low degradation sub-feature group; obtaining the coefficient of variation of the spectral attenuation parameters to determine the intermediate phosphor aging index; and combining the intermediate phosphor aging index with the temperature influence factor, obtaining a comprehensive aging score by weighted summation, and judging the aging degree of the phosphor.

[0012] Furthermore, the step of fusing the aging degree of the components with relevant impedance data to generate a comprehensive multi-component failure index includes: acquiring the fluorescence aging degree of the components; extracting corresponding solder crack data from a preset aging database to obtain solder crack values; performing weighted average fusion using a data fusion method based on the solder crack values ​​and impedance offset to obtain fused aging characteristic values; determining the component aging rate based on the aging characteristic values; if the component aging rate exceeds a preset threshold, adjusting the sampling frequency in the phase acquisition method according to the crack propagation rate to obtain an adjusted sampling frequency; acquiring real-time monitoring data of multiple components for the adjusted sampling frequency to obtain a real-time data set; generating a comprehensive evaluation score based on the real-time data set and the component aging rate; and fusing the data in the real-time data set to generate a comprehensive multi-component failure index.

[0013] Furthermore, the step of obtaining the estimated remaining lifespan of the LED module through prediction model analysis based on the trend of the comprehensive multi-component failure indicators includes: collecting multi-component failure indicators through sensors to obtain an offset trend sequence; analyzing the offset trend sequence using a time series prediction model to obtain a preliminary value of the remaining lifespan, wherein the time series prediction model is constructed by a preset autoregressive order and a moving average order; inputting the offset trend sequence into the time series prediction model to determine the preliminary value of the remaining lifespan; if the preliminary value of the remaining lifespan is lower than a preset threshold, adjusting the preliminary value of the remaining lifespan in conjunction with the ambient temperature factor; multiplying the preliminary value of the remaining lifespan by a preset temperature correction coefficient to determine the adjusted estimated value of the remaining lifespan; and generating the final estimated value of the remaining lifespan of the LED module based on the adjusted estimated value of the remaining lifespan.

[0014] Furthermore, the step of generating a health status report based on the remaining lifetime estimate and determining the overall reliability level to support monitoring decisions includes: obtaining an offset trend sequence from the remaining lifetime estimate, the offset trend sequence containing multiple threshold comparison analysis points; comparing each point value in the offset trend sequence with a preset threshold, and marking any point value exceeding the preset threshold as an anomaly; summarizing the proportion of anomalies in the offset trend sequence to generate a health status report; determining a reliability level judgment based on the health status report; correcting the reliability level judgment in conjunction with ambient temperature, multiplying the reliability level judgment by a preset temperature coefficient to obtain an overall evaluation index; and weighting and adjusting the preliminary evaluation value based on the overall evaluation index to form a final evaluation result supporting online monitoring decisions. The technical solution provided by this embodiment of the invention may include the following beneficial effects:

[0015] This invention discloses an online monitoring and lifespan prediction method for LED modules. Addressing the reliability degradation caused by the degradation of multiple components in operating LED modules, it acquires impedance characteristic data using non-destructive sensors, extracts the feature vector of the frequency domain impedance curve using impedance spectroscopy analysis, and refines a low-dimensional feature set using principal component analysis to accurately capture the degradation characteristics of the chip and phosphor. When the dominant component exceeds a threshold, a support vector machine classifier identifies the chip aging failure mode, and cluster analysis is used to determine the degree of phosphor aging. Impedance phase shift data related to solder joint cracks are fused to construct a comprehensive multi-component failure index. If the index shows a shift trend, a time series prediction model is used to analyze the remaining lifespan and generate a health status report to support online monitoring decision-making. This invention, through multi-dimensional feature fusion and intelligent analysis, achieves accurate characterization and lifespan prediction of LED module degradation mechanisms, improving system reliability and maintenance efficiency. Attached Figure Description

[0016] Figure 1 This is a flowchart of an online monitoring and lifespan prediction method for LED modules according to the present invention.

[0017] Figure 2 This is a schematic diagram of an online monitoring and lifespan prediction method for LED modules according to the present invention.

[0018] Figure 3 This is another schematic diagram of an online monitoring and lifespan prediction method for LED modules according to the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0020] like Figure 1-3 This embodiment of an online monitoring and lifespan prediction method for LED modules may specifically include:

[0021] S101. Impedance characteristic data are obtained from the running LED module through a non-destructive sensor, and the original signal is processed by impedance spectrum analysis to obtain the characteristic vector of the frequency domain impedance curve.

[0022] Impedance characteristic data is acquired from the operating LED module using non-destructive sensors to obtain the raw time-domain signal. Fourier transform is applied to the raw time-domain signal to obtain a frequency-domain impedance curve, which includes the impedance amplitude and phase value at each frequency point. Feature vectors are extracted from the frequency-domain impedance curve to obtain the curve peak and phase shift. The curve peak is obtained by identifying the point of maximum amplitude, and the phase shift is obtained by comparing the actual phase with a preset reference phase difference. Module aging indicators are calculated based on the curve peak and the phase shift, obtained as a weighted average of the peak decay rate and the shift angle. Brightness decay data and temperature change data of the LED module are acquired, and combined with the module aging indicators, a linear regression method is used to calculate a comprehensive evaluation value. If the comprehensive evaluation value exceeds a preset threshold, a fault warning signal is generated.

[0023] Specifically, in one implementation, impedance characteristic data is obtained from an operating LED module using a non-destructive sensor. First, it is necessary to select a suitable sensor type, such as a current and voltage probe or an impedance analyzer. These sensors can be attached to the circuit board of the LED module without affecting its normal light-emitting function.

[0024] Specifically, non-destructive sensors operate based on electromagnetic induction or capacitive coupling, avoiding direct intervention in the circuit, thereby ensuring the continuous operation of LED modules in lighting applications.

[0025] For example, in a maintenance scenario for an LED display, sensors are installed at the input of the module to collect voltage and current signals in real time, forming raw impedance data. This method is applicable to various LED modules, such as indoor lighting fixtures or outdoor advertising screens, demonstrating the versatility of data acquisition. In this way, stable impedance characteristic data can be obtained, providing a foundation for subsequent analysis. Furthermore, the acquired impedance characteristic data includes complex impedance values, which reflect the internal electrical behavior of the LED module, such as combinations of resistance, inductance, and capacitance.

[0026] It should be noted that the advantage of non-destructive sensors is that they can operate in high-temperature or high-humidity environments without causing physical damage to the module.

[0027] In one possible implementation, the sensor transmits data to the processing unit wirelessly, avoiding interference from wired connections. This data acquisition process emphasizes real-time performance, such as acquiring multiple sample points per second to ensure the capture of dynamic changes.

[0028] Preferably, in the scenario where LED modules are used in automotive headlights, sensors are integrated within the lamp housing to monitor impedance fluctuations during driving, thereby supporting fault warnings. This embodiment covers applications in the field of mobile lighting and enhances the flexibility of the technical solution. Processing the raw signal using impedance spectroscopy is a core step, which involves converting the time-domain signal to the frequency domain to extract the impedance curve.

[0029] Specifically, impedance spectroscopy analysis works by applying excitation signals of different frequencies and measuring the response of the LED module to construct an impedance-frequency curve. This analysis helps identify internal defects in the module, such as loose solder joints or chip aging.

[0030] In one embodiment, the original signal is first filtered to remove noise, for example, by using a low-pass filter to preserve the effective frequency band. Then, the signal is converted from the time domain to the frequency domain using a Fast Fourier Transform (FFT) to obtain a complex impedance spectrum. Further, curve fitting techniques are applied to the spectrum to extract features such as the radius or phase angle of the Nyquist plot. These features form the eigenvectors of the frequency domain impedance curve, used for subsequent diagnostics. This detailed process ensures the accuracy of the analysis and is particularly useful in the mass production testing of LED modules, enabling non-contact quality control.

[0031] For example, when processing raw signals, impedance spectral analysis methods involve equivalent circuit modeling, such as equating an LED module to a resistor-capacitor parallel circuit. Key points of the impedance curve, such as the peak values ​​of the real and imaginary parts, are obtained by fitting the model parameters using the least squares method. This modeling process clarifies how eigenvectors are derived from the raw data: first, the impedance magnitude and phase at each frequency point are calculated, and then vectorized into a multidimensional array, such as an impedance value vector containing 10 frequency points. In one implementation, for high-brightness LED modules used in stage lighting, the analysis focuses on the mid-frequency curve to detect impedance changes caused by thermal stress. This approach not only explains the internal logic of the analysis engine but also demonstrates its application in entertainment lighting, resulting in improved module reliability.

[0032] Preferably, the generation of the eigenvector of the frequency domain impedance curve involves a feature selection step, such as reducing dimensionality through principal component analysis to ensure that the vector contains the most representative information.

[0033] Specifically, principal component analysis involves performing singular value decomposition on the impedance curve data matrix, retaining the first few principal components to form a simplified vector. This vector can characterize the overall impedance characteristics of the LED module and is used to compare the states of different modules.

[0034] In one possible implementation, the vector length is set to 20 dimensions, covering the impedance response from low to high frequencies.

[0035] It should be noted that this feature extraction enhances data processability, enabling rapid fault location in online monitoring systems for LED modules without requiring system shutdown and disassembly. Furthermore, in another embodiment, for flexible LED modules used in wearable devices, the impedance spectroscopy analysis method is adjusted to low-power excitation to adapt to battery-powered environments. The specific process includes acquiring impedance data under bending conditions, then applying a window function to smooth the signal and avoid edge effects. The resulting frequency domain curve feature vector is used to evaluate the module's mechanical stability, such as detecting impedance drift caused by bending. This embodiment demonstrates the versatility of the technical solution in the portable lighting field, ensuring an expanded protection range through objective description of the analysis process. In one implementation, combining the entire data acquisition and analysis process, data is first acquired through sensors and then immediately input into the impedance spectroscopy analysis module.

[0036] For example, the module internally implements pipelined processing for signal preprocessing, frequency domain transformation, and feature extraction. This integration approach is suitable for industrial LED module production lines, ensuring that the impedance characteristics of each module meet standards. Furthermore, the output of the feature vectors can be connected to a database for long-term trend analysis, thereby enabling predictive maintenance in lighting systems.

[0037] Specifically, for the implementation of high-power LED modules in street lighting applications, sensors are installed inside the light pole to collect operational data. Impedance spectroscopy analysis focuses on high-frequency curves to identify impedance increases caused by corrosion. The resulting feature vectors are grouped using clustering algorithms to distinguish between normal and abnormal states. This detailed explanation highlights the business value of the analysis process, enabling the reduction of failure rates in urban lighting management.

[0038] Preferably, in scenarios involving multi-module arrays, such as LED video walls, the impedance spectroscopy analysis method is extended to parallel processing, where each module independently generates a feature vector, which is then fused into a unified vector. This approach explains how to handle large-scale data while ensuring the efficiency of the analysis.

[0039] In one embodiment, vector fusion employs a weighted average, with weights based on module location. This implementation enhances the flexibility of the solution. Finally, in a comprehensive embodiment, the entire process, from sensor acquisition to feature vector output, forms a closed-loop system for LED module lifetime assessment. In this way, the technical solution can be applied in various scenarios within the lighting field, achieving the goal of non-destructive monitoring.

[0040] S102. Based on the obtained frequency domain impedance curve eigenvectors, principal component analysis is used to extract the dominant components and determine the low-dimensional feature set reflecting the degradation of multiple components.

[0041] The frequency domain impedance curves of multiple components are acquired, and the original data points of the frequency domain impedance curves are collected. Fourier transform is used to process the original data points to obtain the feature vectors of the frequency domain impedance curves. Based on the feature vectors, principal component analysis is used to extract the dominant components from the feature vectors. By calculating the covariance matrix of the feature vectors and performing eigenvalue decomposition, the variance contribution rate corresponding to the dominant component is determined. If the variance contribution rate exceeds a preset threshold, the dominant component is retained, forming a preliminary low-dimensional feature set. For the preliminary low-dimensional feature set, degradation reflection features are obtained from historical data on multi-component degradation. These degradation reflection features are quantitative indicators based on historical degradation patterns. By calculating the cosine similarity between the preliminary low-dimensional feature set and the degradation reflection features, the matching degree between the preliminary low-dimensional feature set and the degradation reflection features is determined, resulting in an adjusted low-dimensional feature set. Based on the adjusted low-dimensional feature set and combined with real-time input from component status monitoring, verification samples are obtained. These verification samples are control data sampled from the real-time input, thus determining the low-dimensional feature set reflecting multi-component degradation.

[0042] Specifically, in one implementation, principal component analysis is used to extract the dominant components based on the obtained frequency domain impedance curve eigenvectors, thereby determining a low-dimensional feature set reflecting the degradation of multiple components.

[0043] Specifically, the frequency domain impedance curve is first processed to generate an eigenvector. This vector includes impedance amplitude and phase values ​​at multiple frequency points, which are derived from the time-domain signal through Fourier transform to capture changes in the electrical characteristics of the component. Further, principal component analysis (PCA) is a statistical method used to reduce high-dimensional data to a lower-dimensional space while retaining the main variation information of the original data. In this embodiment, the eigenvector of the frequency domain impedance curve is used as the input data matrix, where each row represents the eigenvector of a sample, and each column corresponds to the impedance characteristic at a frequency point. By calculating the covariance matrix and solving for its eigenvalues ​​and eigenvectors, the top few eigenvectors with larger eigenvalues ​​are selected as the dominant components. These dominant components reflect the main patterns in the data; for example, in a cable system, the dominant components may correspond to impedance changes caused by insulation degradation or conductor corrosion, thus forming a low-dimensional feature set.

[0044] For example, in a degradation monitoring scenario for power transmission cables, the eigenvector of the frequency domain impedance curve might contain impedance values ​​at 100 frequency points. When using principal component analysis, the data is first standardized to eliminate dimensional differences between frequency points. Then, the covariance matrix is ​​calculated to obtain the eigenvalue spectrum, and the top five dominant components with a cumulative variance contribution rate of over 85% are selected. These components are combined into a low-dimensional feature set for subsequent degradation assessment. This method effectively reduces noise interference and highlights the joint degradation characteristics of multiple components, such as joints and insulation.

[0045] Preferably, in another embodiment, the amount of principal components can be adjusted to suit different monitoring needs.

[0046] Specifically, if the monitored object is a high-voltage cable system, a threshold can be set based on experience, such as a variance contribution rate of 90%, to extract the top 3 to 7 dominant components. This low-dimensional feature set is obtained by projecting the original vector onto these components. For example, the original vector v is used to obtain the low-dimensional vector using the formula v*U, where U is the dominant eigenvector matrix. The feature set obtained in this way reflects the comprehensive pattern of component degradation, such as impedance drift caused by temperature changes.

[0047] It should be noted that the principal component analysis process includes data centralization and singular value decomposition to ensure computational stability. In cable degradation monitoring, this analysis helps identify hidden degradation patterns. For example, when multiple components age simultaneously, the principal components capture the overall shift in the impedance curve, thus providing more reliable diagnostic information. In this way, the system can process large amounts of frequency domain data, achieving efficient feature extraction.

[0048] In one possible implementation, for medium-voltage cable scenarios, the eigenvectors of the frequency domain impedance curve can be obtained from a wideband scan, covering the range of 10Hz to 1MHz. After principal component analysis extracts the dominant components, the size of the low-dimensional feature set is typically reduced to 10% of the original dimension, which facilitates integration into monitoring equipment.

[0049] For example, the feature set can be used as input to a classifier to distinguish between normal, mildly degraded, and severely degraded states. Furthermore, the process of determining this low-dimensional feature set emphasizes the reflection of degradation across multiple components; for instance, in a cable system, the dominant component may correspond to variations in the resonant frequencies of different components. By analyzing the impedance curves of multiple samples, the system can construct a degradation model, improving prediction accuracy.

[0050] For example, in implementation, the effectiveness of the dominant component can be verified by combining historical data.

[0051] Specifically, cable samples with known degradation levels were collected, and principal component analysis was applied to observe whether the low-dimensional feature set could linearly separate different degradation levels. This validation ensured the robustness of the feature set, resulting in stable monitoring performance in practical deployments.

[0052] Understandably, this method is applicable to various scenarios within the same field, such as degradation monitoring of underground cables or overhead lines, and is not limited to a specific type. By flexibly selecting the number of dominant components, the technical solution demonstrates versatility.

[0053] S103. If the dominant component in the low-dimensional feature set exceeds the preset threshold, then the feature set is pattern-recognized by a support vector machine classifier to determine whether there is a chip aging failure mode.

[0054] A low-dimensional feature set of chip aging data is obtained, comprising quantized values ​​of multiple aging features. Using this low-dimensional feature set, an extended index of the chip aging mode is calculated to obtain a preliminary mapping for aging type identification. For this preliminary mapping, a threshold exceedance detection process is used to determine the dominant component, which involves comparing the dominant component with a preset threshold to identify the association vector for feature set classification. Based on the association vector, control samples are extracted as validation data, consisting of sample features obtained from the association vector. If the validation data meets preset conditions, the failure mode in the chip aging mode is determined using the validation data, resulting in a failure mode discrimination result. Based on the failure mode discrimination result, the feature weights of the low-dimensional feature set are updated to optimize the mapping accuracy of aging type identification.

[0055] Specifically, in one implementation, when the dominant component in the low-dimensional feature set exceeds a preset threshold, the system performs pattern recognition on the feature set using a support vector machine classifier to determine whether there is an aging-type failure mode of electronic components in the cable system.

[0056] Specifically, the first step is to check if the value of the dominant component exceeds a threshold set based on historical degradation data, such as 1.2 times the cumulative variance contribution rate in cable monitoring, to trigger further analysis. This threshold determination ensures that only significantly changed feature sets are classified, thus focusing on potential degradation signals. Furthermore, a support vector machine (SVM) classifier is a supervised learning method used to construct a hyperplane in a high-dimensional space to classify the input data into different categories. In this implementation, a low-dimensional feature set is used as the input vector, and the classifier distinguishes between normal states and aging failure modes by maximizing the margin.

[0057] For example, in the scenario of monitoring the degradation of underground cables, the feature set may reflect the joint changes of the insulation layer and conductor components, while the classifier is trained on labeled samples to identify failure modes such as thermal aging or mechanical fatigue.

[0058] It should be noted that the working principle of support vector machines involves mapping feature vectors to a higher-dimensional space and using kernel functions such as radial basis functions to handle nonlinear separation. This can effectively capture the aging characteristics of electronic components such as sensor chips in cable systems, which may experience impedance drift due to long-term exposure to humid environments.

[0059] Preferably, in another implementation, the training process of the support vector machine classifier includes collecting multiple sets of cable sample data containing known aging types, such as oxidation corrosion or insulation breakdown. First, the low-dimensional feature sets of these samples are labeled, and then the classification model is optimized by minimizing structural risk.

[0060] Specifically, the classifier solves for support vectors, which define the decision boundary. For example, in the monitoring of overhead cable lines, if the dominant component exceeds a threshold, after the system inputs a feature set, the classifier outputs probability scores to determine whether there is a chip aging failure mode, such as poor contact of components on a circuit board due to vibration. This method ensures the model's generalization ability through cross-validation and can handle degradation data under different cable lengths in actual deployments.

[0061] For example, in the implementation of medium-voltage cable systems, when the dominant component exceeds a threshold, a support vector machine classifier performs pattern recognition on the feature set. This process includes feature scaling to normalize the input and then calculating the decision function value. If the result points to the aging category, the system labels it as a failure mode. This identification helps to detect potential problems in the cable network early and provides a basis for maintenance.

[0062] Understandably, this technical solution is applicable to multiple scenarios within the same field, such as monitoring the degradation of submarine cables, where the support vector machine classifier adjusts its parameters to adapt to aging assessments under salt corrosion environments. In this way, the system achieves accurate assessment of degradation across multiple components.

[0063] In one possible implementation, when monitoring high-voltage cables, a support vector machine classifier, combined with a low-dimensional feature set, determines the chip aging type and failure mode, and a soft interval can be introduced to tolerate noisy data.

[0064] Specifically, the classifier balances accuracy and robustness through a penalty function. For example, when processing long-distance cable data, it ensures accurate threshold response to the dominant component, thereby outputting reliable failure judgment results. This flexibility enhances the versatility of the technical solution in cable monitoring.

[0065] S104. From the chip aging type failure mode results obtained by judgment, obtain the associated phosphor degradation sub-features, and use cluster analysis method to group the sub-features to determine the phosphor aging degree.

[0066] Phosphor degradation sub-features are obtained from the chip aging failure mode results. These sub-features reflect the changes in phosphor properties during the aging process. K-means clustering is used to group these phosphor degradation sub-features. K-means clustering takes the sub-feature vectors as input and iteratively calculates the centroids to obtain a preliminary group set. The distribution density of the phosphor degradation sub-features is calculated based on the preliminary group set, obtained by averaging the spatial distances of the sub-features within each group. If the distribution density exceeds a preset threshold, a high-degradation sub-feature group is identified; otherwise, a low-degradation sub-feature group is identified. Spectral attenuation parameters are extracted from the high-degradation or low-degradation sub-feature groups, obtained from the slope of the spectral curves of the sub-feature groups. The coefficient of variation of the spectral attenuation parameters is obtained, and the coefficient of variation is used to determine the phosphor aging intermediate index by dividing the standard deviation of the parameter by the mean. A temperature influence factor is integrated into the phosphor aging intermediate index, obtained from a preset temperature correlation coefficient in the chip aging type. A comprehensive aging score is obtained by weighted summation, and this comprehensive aging score is used to determine the degree of phosphor aging.

[0067] Specifically, in one implementation, the associated phosphor degradation sub-features are first obtained from the determined chip aging type failure mode results. These failure mode results typically originate from accelerated aging tests or real-time monitoring data of the chip, such as failure types obtained through temperature cycling or humidity exposure experiments, like thermal stress failure or optical decay failure. Phosphor degradation sub-features refer to specific degradation indicators exhibited by the phosphor material during aging, such as brightness decay rate, color temperature shift, or spectral peak shift. These sub-features are closely related to the failure modes; for example, in thermal stress failure mode, the phosphor may exhibit brightness decay due to a loose crystal structure. The acquisition process can be implemented through database queries or feature matching algorithms, taking the failure mode as input and outputting a set of related sub-features, thus providing a data foundation for subsequent analysis. This approach ensures the targeted nature of feature extraction and is widely used in quality control scenarios in chip manufacturing. Furthermore, the obtained phosphor degradation sub-features need to be preprocessed to improve data quality.

[0068] For example, normalizing the sub-feature data can convert indicators with different dimensions, such as brightness decay rate and color temperature shift value, into a unified scale, thus avoiding bias in cluster analysis.

[0069] It should be noted that the association of phosphor degradation sub-features can be obtained based on an association model trained on historical data. This model learns the correspondence between failure modes and sub-features through machine learning methods, such as using a decision tree algorithm to map thermal failure to brightness decay sub-features. This preprocessing step helps improve clustering accuracy, reduces noise interference, and ensures reliable analysis results in practical LED chip aging monitoring. Clustering analysis methods are used to process the grouped sub-features to determine the degree of phosphor aging. Here, clustering analysis methods refer to unsupervised learning techniques, such as the K-means algorithm, which group sub-features based on similarity.

[0070] For example, the distance between sub-features is first calculated, such as using Euclidean distance to measure the difference between sub-feature vectors. Then, the cluster centers are iteratively optimized until convergence. After grouping, each cluster represents a degradation mode; for example, one cluster may correspond to slight degradation, and another to severe degradation. The degree of aging is quantified by calculating the average value within the cluster or the deviation of the cluster center from a standard value. For example, if the brightness decay rate of the cluster center exceeds a threshold, it is judged as moderate aging. This method achieves automated grouping in chip aging assessment, which is more efficient than traditional manual judgment.

[0071] In one possible implementation, cluster analysis can be combined with hierarchical clustering methods to further refine the groupings. Hierarchical clustering starts with a single sub-feature and gradually merges similar clusters to form a tree-like structure, which facilitates the observation of the gradual process of degradation.

[0072] For example, color temperature shifters and brightness attenuators of phosphors can be merged, and when the merging reaches a certain level, the cluster compactness index can be calculated to determine the optimal number of groups. This extension ensures the flexibility of the analysis and can be applied to testing different chip batches within the same field.

[0073] Preferably, after determining the degree of phosphor aging, a report can be generated, including the aging level and recommended maintenance measures.

[0074] For example, mild aging requires continued monitoring, while moderate aging suggests replacing the phosphor layer. This output plays a role in quality management on the chip production line, improving the accuracy of equipment lifespan prediction.

[0075] Specifically, the clustering analysis process needs to be detailed to ensure clarity. First, prepare a sub-feature dataset, such as collecting values ​​for brightness decay rate, color temperature shift, and spectral peak shift from multiple chip samples, forming a multi-dimensional vector. Then, initialize cluster centers, for example, by randomly selecting K vectors as starting points. Next, assign each sub-feature to the nearest center by minimizing the intra-cluster variance, and update the center positions, repeating this process until the centers stabilize. After grouping, the degree of aging can be assessed using the statistical properties of the clusters, such as calculating the average degradation index of the sub-features within the cluster. An index between 0 and 0.3 indicates mild aging, between 0.3 and 0.7 indicates moderate aging, and above 0.7 indicates severe aging. This detailed process provides a traceable analytical path in chip aging diagnostics, supporting subsequent optimization.

[0076] Understandably, this method is applicable to multiple scenarios within the same field, such as aging tests of white LED chips, where phosphor degradation directly affects luminous efficiency. Through clustering, not only is the degree of aging determined, but also potential degradation mode correlations are revealed; for example, thermal failure is often associated with brightness decay clusters, thus guiding material improvements.

[0077] For example, in one embodiment, after performing aging tests on a batch of LED chips and obtaining the thermal stress failure mode results, correlated sub-features such as a brightness decay rate of 15% and a color temperature shift of 2% are extracted. K-means clustering is used to divide these sub-features into three groups: the first group represents initial degradation, the second group represents moderate degradation, and the third group represents severe degradation. Ultimately, the overall phosphor aging level is determined to be moderate. This embodiment demonstrates the practicality of the method. Furthermore, to enhance robustness, noise filtering can be introduced before clustering, for example, using principal component analysis to reduce the dimensionality of the sub-feature space and retain the main variation components. This optional feature improves grouping accuracy in complex aging environments.

[0078] In one embodiment, the clustering model is dynamically updated by incorporating real-time monitoring data, such as recalculating cluster centers every cycle, to adapt to gradual degradation during chip use. This approach enables continuous evaluation, resulting in more precise aging management in the field of lighting equipment maintenance.

[0079] S105. Based on the determined phosphor aging degree, the phase shift data of the impedance related to solder joint cracks are integrated to obtain a comprehensive multi-component failure index.

[0080] The fluorescence aging degree of the component is obtained, and the corresponding solder joint crack data is extracted from a preset aging database to obtain the solder joint crack value. Based on the solder joint crack value and impedance offset, a weighted average fusion method is used to obtain the fused aging characteristic value. The component aging rate is determined based on the aging characteristic value. If the component aging rate exceeds a preset threshold, the sampling frequency in the phase acquisition method is adjusted according to the crack propagation rate to obtain the adjusted sampling frequency. Real-time monitoring data of multiple components is collected at the adjusted sampling frequency to obtain a real-time data set. A comprehensive evaluation score is generated based on the real-time data set and the component aging rate. Based on the comprehensive evaluation score, the data in the real-time data set is fused to generate a comprehensive multi-component failure index.

[0081] Specifically, in one implementation, the degree of phosphor aging is first determined by analyzing changes in the optical properties of the phosphor in electronic devices such as LED lighting systems.

[0082] Specifically, phosphor aging typically manifests as a decrease in luminous efficiency and a shift in color temperature. This can be addressed by irradiating the sample with a spectrometer and recording the peak wavelength and intensity ratio of the emission spectrum. The degree of aging can then be calculated by comparing the initial state with the current state.

[0083] For example, the percentage of peak intensity decay can be used as a quantification criterion. This method is applicable to various LED applications, such as indoor lighting or displays, ensuring the accuracy of aging assessments. In practical business operations, this determination process helps identify potential failure risks early and improves equipment maintenance efficiency. Furthermore, impedance phase shift data related to solder joint cracks is obtained; this data comes from electrical impedance spectroscopy analysis. In electronic components, solder joint cracks can cause abnormal impedance phase shifts. This is specifically achieved by applying an AC signal to an impedance analyzer and measuring the impedance value and phase angle of the solder joint at different frequencies. Impedance phase shift refers to the deviation between the actual phase and the ideal phase.

[0084] For example, when a crack appears, the low-frequency phase may shift negatively by several degrees. This shift data can be obtained by averaging multiple measurements, making it suitable for circuit boards with dense solder joints, such as power modules or drive circuits, thus providing a reliable basis for crack detection. A detailed explanation of this process includes: first, preparing the test sample and ensuring the solder joints are clean; then setting the analyzer parameters, such as a frequency range from 1Hz to 1MHz; next, recording the phase value at each frequency point and calculating the shift, for example, shift = actual phase - reference phase. In this way, the severity of the crack can be quantified, providing data support for subsequent fusion. This technology plays a crucial role in electronic device failure analysis because it directly relates to the mechanical integrity of components, and fusion can improve the accuracy of the overall assessment.

[0085] Preferably, the above data is fused to obtain a comprehensive multi-component failure index. This fusion process is the core step and involves weighted averaging or machine learning methods to integrate multi-source information.

[0086] Specifically, the phosphor aging level is first converted into a numerical score, such as a normalized value from 0 to 1, where 0 represents no aging and 1 represents complete failure. Simultaneously, the impedance phase offset data is also normalized into a similar score, with a high-risk score assigned when the offset exceeds a threshold. Then, a comprehensive index is calculated using a fusion formula, for example, index = weight 1 × aging score + weight 2 × offset score, where the weights are adjusted according to the importance of the components; for example, the phosphor weight is 0.6 and the offset weight is 0.4. This method ensures data complementarity. In LED equipment failure monitoring, the process includes data preprocessing to remove noise, followed by fusion calculation, ultimately outputting a failure index value from 0 to 100. This detailed process helps identify multi-component interactive failures; for example, aging phosphor may exacerbate solder joint stress, leading to accelerated cracking, thus enabling preventative maintenance and reducing equipment downtime.

[0087] In one possible implementation, the fusion process can be further refined for specific LED lighting scenarios. First, multiple sets of data samples are collected, such as extracting aging and offset data from lighting fixtures with different usage durations. Then, clustering algorithms are applied to group similar failure modes, such as grouping minor aging and small offsets together. Next, a comprehensive index for each group is calculated to ensure that the index reflects the overall failure trend. This implementation demonstrates the versatility of the technology and is suitable for quality control in mass production environments.

[0088] It should be noted that the above fusion method can be adapted to different environments by adjusting parameters. For example, under high temperature conditions, the weight of the offset data can be increased to reflect the additional impact of thermal stress on the weld joint, thereby enhancing the robustness of the index.

[0089] In one embodiment, the output of integrated multi-component failure indicators can be used to generate a report, for example, by setting an indicator threshold of 50, triggering an alarm when the threshold is exceeded. This approach provides intuitive decision support in electronic equipment maintenance. Furthermore, the logical flow of the entire process, from data acquisition to fusion, ensures the coherence of each step; for example, aging data, as input, directly affects the fusion weight allocation.

[0090] Understandably, this technical solution can be extended to other electronic components, such as sensor modules, within the same field, maintaining consistency across business areas.

[0091] Specifically, the fused indicators enable a quantitative assessment of the interaction effects of multiple components, improving the accuracy of failure prediction in practical applications.

[0092] S106. If the failure indicators of multiple components show a deviation trend, the indicator sequence is analyzed by time series prediction model to obtain the estimated value of the remaining life of the LED module.

[0093] Failure indicators of multiple components are collected by sensors to obtain an offset trend sequence. An ARIMA model is used to analyze the offset trend sequence to obtain a preliminary value of the remaining lifetime. The ARIMA model is constructed based on a preset autoregressive order and a moving average order. The offset trend sequence is input into the ARIMA model to determine the preliminary value of the remaining lifetime. It is then determined whether the preliminary value of the remaining lifetime is lower than a preset threshold. If the preliminary value of the remaining lifetime is lower than the preset threshold, the preliminary value is adjusted based on the ambient temperature factor. The adjusted estimated remaining lifetime is determined by multiplying the preliminary value of the remaining lifetime by a preset temperature correction coefficient. The adjusted estimated remaining lifetime is then obtained to determine the estimated remaining lifetime of the LED module. (The process is repeated twice in the original text.) The adjusted remaining lifespan estimate is calculated by multiplying a preset temperature correction factor by the preliminary remaining lifespan value. Based on the adjusted remaining lifespan estimate, the remaining lifespan estimate of the LED module is determined. After collecting failure indicators from multiple components to generate an offset trend sequence, the sequence is analyzed using an ARIMA model to generate a preliminary remaining lifespan value. The ARIMA model is constructed using preset autoregressive and moving average orders. The offset trend sequence is input into the ARIMA model to obtain the preliminary remaining lifespan value. It is determined whether the preliminary remaining lifespan value is lower than a preset threshold. If it is lower than the preset threshold, the preliminary value is adjusted based on the ambient temperature factor. The adjusted remaining lifespan estimate is determined by multiplying the preset temperature correction factor by the preliminary remaining lifespan value. Based on the adjusted estimate, the remaining lifespan estimate of the LED module is generated.

[0094] Specifically, in one implementation, estimating the remaining lifetime of an LED module first requires monitoring a comprehensive set of failure indicators for multiple components. These indicators typically include brightness decay of the LED chip, voltage fluctuations in the driver circuit, and temperature changes in the heat dissipation components. By collecting this data in real time, a multi-dimensional sequence of indicators is formed.

[0095] For example, in LED lighting systems, failure indicators can be obtained from sensors, and the overall failure level can be calculated comprehensively. If these indicators show an offset trend, i.e., the sequence values ​​deviate from the normal range or exhibit a continuous upward or downward pattern, further analysis is triggered. This offset trend is determined based on statistical methods, such as calculating the deviation between the moving average and the standard deviation. When the deviation exceeds a preset threshold, the trend is confirmed. This ensures early detection of potential faults and promotes the maintenance and optimization of LED modules. Furthermore, time series prediction models are used to analyze the aforementioned indicator sequences. This model can take the form of an ARIMA model or an LSTM neural network, aiming to capture the time dependence in the sequence.

[0096] Specifically, the ARIMA model uses differencing to station the sequence, then estimates autoregressive and moving average parameters to predict future values. In LED module applications, the indicator sequence is first preprocessed, such as removing noise and filling missing values, and then input into the model for training. Model training is based on historical data, such as failure indicator records from the past few months, learning the pattern of the sequence to generate a prediction curve. In this way, the model can estimate the remaining time from the current state of the LED module to failure, providing quantitative values ​​such as the number of hours remaining. This helps in planning replacement cycles in real-world lighting scenarios and avoiding sudden failures. Preferably...

[0097] In one possible implementation, the deviation trend of the comprehensive multi-component failure index can be identified by threshold comparison or anomaly detection algorithm.

[0098] For example, the isolated forest algorithm is used to isolate outliers, and when indicators of multiple components show deviations simultaneously, the overall trend is confirmed. This method is particularly suitable for LED display modules because displays involve more pixel-level components, and deviations can lead to image distortion. Next, when analyzing the time series forecasting model, seasonal factors, such as the impact of ambient temperature on LED performance, are considered. By decomposing the series into trend, seasonal, and residual components, the model can more accurately predict the remaining lifetime.

[0099] It should be noted that this analysis process does not rely on complex numerical calculations, but focuses on data pattern recognition to ensure applicability under different LED module configurations.

[0100] For example, in the implementation scenario of LED street light modules, the failure metric sequence might include luminous flux attenuation and current instability data. If the sequence shows an offsetting trend, the model can simulate the evolution of the metrics over the next few weeks, estimating the remaining lifetime to be several hundred hours. This allows maintenance personnel to intervene early, improving system reliability. In another embodiment, for indoor LED lighting modules, the model can integrate humidity factors to predict accelerated degradation in high-humidity environments. Through these scenarios, the technical solution demonstrates its versatility in the lighting field.

[0101] Specifically, the steps to obtain an estimate of the remaining lifetime of an LED module include post-processing of the model output.

[0102] For example, the remaining time interval is calculated by comparing the end value of the predicted sequence with a failure threshold. This estimate can be used to generate reports and support the decision-making process. In practical applications, this method enables continuous monitoring of module status, improving overall efficiency.

[0103] In one embodiment, the selection of the time series forecasting model can be adjusted according to the data scale; a simple ARIMA model is used for small-scale data, while LSTM is used for large datasets, to adapt to different LED module production environments. This enhances the flexibility of the solution. Furthermore, through the above analysis, the remaining lifetime estimate helps optimize inventory management, for example, in LED display system factories, by adjusting spare parts procurement based on the forecast value.

[0104] Understandably, this implementation covers the entire process from trend detection to prediction, ensuring the reliable operation of LED modules in the lighting field.

[0105] S107. From the obtained remaining lifetime estimate, generate health status report data to determine the overall reliability level to support online monitoring decisions.

[0106] From remaining lifetime estimation: Based on the offset remaining trend sequence, the lifetime estimation offset trend sequence includes multiple offset trend sequence threshold comparison analysis and anomaly point ratio summary to generate a health status report. The health status report reliability level judgment is combined with the multiplication correction point of the environmental temperature coefficient to obtain the overall value. The preset threshold is used to compare the value of each point in the offset trend sequence. If the value of each point exceeds the preset threshold, it is marked as an anomaly point summary index. The proportion of anomalies in the offset trend sequence is used to obtain the health status report. The reliability level judgment is determined by the health status report. The reliability level judgment is corrected by combining the environmental temperature. The correction is obtained by multiplying the reliability level judgment by the preset temperature coefficient to obtain the overall evaluation index. The preliminary overall evaluation value is adjusted according to the overall evaluation index. The weight adjustment is formed by adjusting the weight value of the preliminary index integrated into the overall evaluation to form online monitoring and decision-making based on the preliminary value. The online monitoring decision is made based on the adjusted preliminary value. < /

[0107] Specifically, in one implementation, health status report data is generated from the obtained estimated remaining lifespan of the LED modules. This process first involves integrating the estimates with other monitoring parameters, such as current operating hours and environmental factors, to form a comprehensive dataset.

[0108] Specifically, the report data includes remaining lifespan figures, historical trend charts, and potential risk warnings. This integration ensures the report reflects the overall health of the LED modules. In LED lighting systems, such as indoor luminaire modules, this dataset can be used to generate structured report documents, allowing maintenance personnel to quickly review them. Furthermore, the overall reliability level is assessed based on preset threshold standards.

[0109] For example, the remaining lifetime estimate is compared with the standard lifetime. If the estimate exceeds 80% of the total lifetime, it is considered a high reliability level; if it is between 50% and 80%, it is considered a medium level; and if it is below 50%, it is considered a low level. This judgment is achieved through simple comparison without complex calculations, but differences in module type must be considered.

[0110] In one possible implementation, for LED display modules, pixel failure rate can be introduced as an auxiliary indicator to further refine the reliability assessment and ensure the accuracy of the judgment.

[0111] Preferably, after the report is generated, online monitoring and decision-making are supported. This decision-making process is driven by the report data.

[0112] For example, when the reliability level is low, the system automatically triggers an alarm and recommends that the module be replaced immediately.

[0113] Specifically, in the application of LED street light modules, reports can be linked to an online platform, allowing for remote viewing of remaining life estimates and planning of maintenance schedules based on reliability levels. This support helps reduce unexpected failures and promotes continuous operation of the lighting system.

[0114] For example, in another embodiment, health status report data can be extended to a visualization format, such as a bar chart showing the relationship between remaining lifespan and reliability level.

[0115] It should be noted that this visualization is based on the quantitative output of estimated values, helping users intuitively understand the module status. In indoor LED lighting module scenarios, the report can also include an analysis of the impact of humidity. This factor is integrated when assessing reliability; if humidity causes accelerated degradation, the reliability level is lowered, thereby guiding decisions to adjust environmental control measures.

[0116] Understandably, this method maintains flexibility under different LED module configurations.

[0117] For example, when assessing the reliability level of high-power LED modules, thermal stress indicators can be prioritized and combined with remaining lifetime estimates to generate more comprehensive report data. This flexibility ensures the universal applicability of the technical solution in the lighting field.

[0118] Specifically, the steps for generating health status report data include data aggregation and formatting. First, the remaining lifetime estimate is aggregated with real-time monitoring data, and then a template is applied to generate the report. Further, after determining the reliability level, the decision support module can output recommended actions, such as "It is recommended to purchase spare parts when the remaining lifetime is below a threshold."

[0119] In one embodiment, for LED display systems, this process can be automated, with online updates via a software interface to ensure timely decision-making. This implementation improves the efficiency of module management.

[0120] In one possible implementation, the overall reliability level can be determined using a hierarchical model, for example, by defining three levels and mapping the range of estimated values.

[0121] For example, in LED street light modules, if the estimated reliability exceeds 500 hours, high reliability allows for continued online monitoring without intervention; conversely, low reliability triggers decisions such as dispatching a maintenance team. This tiered model sets thresholds based on business experience, ensuring the judgment process is objective and repeatable. Furthermore, through the aforementioned report generation and judgment, a complete process supporting online monitoring decisions is realized.

[0122] For example, in large-scale lighting projects, reported data can be integrated into a central control system, allowing for reliability assessments of batch modules and thus optimizing resource allocation.

[0123] It should be noted that this support is not limited to a single module, but extends to system-level decision-making to promote reliable operation in the lighting field.

[0124] This invention provides an online monitoring and lifespan prediction system for LED modules, mainly comprising: a signal acquisition and processing module, used to acquire impedance characteristic data from operating LED modules through non-destructive sensors, and process the raw signal using impedance spectroscopy analysis to obtain a frequency domain impedance curve feature vector; a feature extraction module, used to extract dominant components based on the obtained frequency domain impedance curve feature vector using principal component analysis to determine a low-dimensional feature set reflecting the degradation of multiple components; and a pattern recognition module, used to perform pattern recognition on the feature set using a support vector machine classifier if the dominant components in the low-dimensional feature set exceed a preset threshold, to determine whether there is a chip aging type failure mode; and an aging analysis module. The system comprises the following modules: an analysis module, which obtains associated phosphor degradation sub-features from the determined chip aging type failure mode results and uses clustering analysis to group the sub-features to determine the phosphor aging degree; a fusion evaluation module, which fuses solder joint crack-related impedance phase shift data based on the determined phosphor aging degree to obtain a comprehensive multi-component failure index; a lifetime prediction module, which analyzes the index sequence using a time series prediction model to obtain an estimated remaining lifetime value for the LED module if the comprehensive multi-component failure index shows a shift trend; and a report generation module, which generates a health status report from the obtained remaining lifetime estimate to determine the overall reliability level and support online monitoring decisions. The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of one or more embodiments of this specification.

Claims

1. A method for online monitoring and lifespan prediction of LED modules, characterized in that, include: Impedance characteristic data is acquired from the operating LED module using a sensor, and the impedance characteristic data is processed to generate a frequency domain impedance curve feature vector. The dominant component is extracted from the characteristic vector of the frequency domain impedance curve to form a low-dimensional feature set that reflects the degradation of multiple components. The low-dimensional feature set is subjected to pattern recognition by a classifier to determine whether a specific failure mode exists. Relevant sub-features are extracted from the failure mode results, and the sub-features are analyzed to determine the degree of component aging. By integrating the aging degree of the components with relevant impedance data, a comprehensive multi-component failure index is generated. Based on the trend of the comprehensive multi-component failure indicators, the estimated remaining lifespan of the LED module is obtained through predictive model analysis; A health status report is generated based on the estimated remaining lifetime to determine the overall reliability level and support monitoring decisions.

2. The method for online monitoring and lifespan prediction of LED modules as described in claim 1, characterized in that, The process of acquiring impedance characteristic data from a running LED module via a sensor and processing the impedance characteristic data to generate a frequency domain impedance curve feature vector includes: The impedance characteristic data of the LED module in operation is collected using non-destructive sensors to obtain the raw time-domain signal; The original time-domain signal is processed using Fourier transform to generate a frequency-domain impedance curve, which includes the impedance amplitude and phase value at each frequency point. Based on the frequency domain impedance curve, feature vectors are extracted to determine the curve peak and phase shift. The curve peak is obtained by identifying the point of maximum amplitude, and the phase shift is obtained by comparing the actual phase with the preset reference phase difference. The module aging index is calculated based on the peak value of the curve and the phase offset. The module aging index is obtained by weighted average of peak attenuation rate and offset angle. Obtain the brightness decay data and temperature change data of the LED module, and calculate the comprehensive evaluation value using a linear regression method in conjunction with the module aging index; If the comprehensive evaluation value exceeds the preset threshold, a fault warning signal is generated.

3. The method for online monitoring and lifespan prediction of LED modules as described in claim 1, characterized in that, The step of extracting the dominant component based on the characteristic vector of the frequency domain impedance curve to form a low-dimensional feature set reflecting the degradation of multiple components includes: Obtain the frequency domain impedance curves of multiple components, and collect the original data points of the frequency domain impedance curves; The original data points are processed using Fourier transform to generate feature vectors; Based on the eigenvectors, principal component analysis is used to extract the dominant components. By calculating the covariance matrix of the eigenvectors and performing eigenvalue decomposition, the variance contribution rate corresponding to the dominant components is determined. If the variance contribution rate exceeds a preset threshold, the dominant component is retained to form a preliminary low-dimensional feature set; Degradation reflection features are obtained from historical data on multi-component degradation. The matching degree is determined by calculating the cosine similarity between the preliminary low-dimensional feature set and the degradation reflection features, and an adjusted low-dimensional feature set is obtained. Based on the adjusted low-dimensional feature set, combined with the real-time input of component status monitoring, verification samples are obtained to determine the final low-dimensional feature set reflecting the degradation of multiple components.

4. The method for online monitoring and lifespan prediction of LED modules as described in claim 1, characterized in that, The step of performing pattern recognition on the low-dimensional feature set using a classifier to determine whether a specific failure mode exists includes: Obtain a low-dimensional feature set of chip aging data, wherein the low-dimensional feature set contains quantized values ​​of multiple aging features; The extended index of chip aging mode is calculated using the low-dimensional feature set to obtain a preliminary mapping for aging type identification. For the initial mapping, a threshold exceeding detection is used to process the dominant component and determine the association vector for feature set classification; Based on the correlation vector, control samples are extracted as validation data; If the verification data meets the preset conditions, the failure mode in the chip aging mode is determined by the verification data, and the failure mode discrimination result is obtained. Based on the failure mode discrimination results, the feature weights of the low-dimensional feature set are updated to optimize the mapping accuracy of aging type identification.

5. The method for online monitoring and lifespan prediction of LED modules as described in claim 1, characterized in that, The step of extracting relevant sub-features from the failure mode results and analyzing the sub-features to determine the degree of component aging includes: Phosphor degradation sub-features are obtained from the chip aging type failure mode results, and the sub-features reflect the property changes of phosphor during the aging process; The phosphor degradation features are grouped using a clustering method, and a preliminary group set is obtained by iteratively calculating the centroids. The distribution density is calculated based on the preliminary group set. If the distribution density exceeds a preset threshold, a high-degenerate sub-feature group is determined; otherwise, a low-degenerate sub-feature group is determined. Spectral attenuation parameters are extracted using the high-degradation sub-feature group or the low-degradation sub-feature group; Obtain the coefficient of variation of the spectral attenuation parameter to determine the intermediate index of phosphor aging; The phosphor aging intermediate index is combined with the temperature influence factor, and a comprehensive aging score is obtained by weighted summation to determine the degree of phosphor aging.

6. The method for online monitoring and lifespan prediction of an LED module as described in claim 1, characterized in that, The aging degree of the components and related impedance data are integrated to generate a comprehensive multi-component failure index, including: Obtain the fluorescence aging degree of the component, extract the corresponding solder joint crack data from the preset aging database, and obtain the solder joint crack value; Based on the weld crack value and impedance offset, a weighted average fusion method is used to obtain the fused aging characteristic value. The aging rate of the component is determined based on the aging characteristic values. If the aging rate of the component exceeds a preset threshold, the sampling frequency in the phase acquisition method is adjusted according to the crack propagation rate to obtain the adjusted sampling frequency. Based on the adjusted sampling frequency, real-time monitoring data from multiple components are collected to obtain a real-time data set; Based on the real-time data set and component aging rate, a comprehensive evaluation score is generated, and the data in the real-time data set is integrated to generate a comprehensive multi-component failure index.

7. The method for online monitoring and lifespan prediction of LED modules as described in claim 1, characterized in that, The step of obtaining an estimated remaining lifespan of the LED module through predictive model analysis based on the trends of the comprehensive multi-component failure indices includes: Failure indicators of multiple components are collected by sensors to obtain offset trend sequences; The offset trend sequence is analyzed using a time series forecasting model to obtain a preliminary value of remaining life. The time series forecasting model is constructed by pre-setting the autoregressive order and the moving average order. Input the offset trend sequence into the time series prediction model to determine the preliminary value of remaining lifetime; If the initial value of the remaining lifetime is lower than a preset threshold, the initial value of the remaining lifetime is adjusted in combination with the ambient temperature factor. The adjusted estimated value of the remaining lifetime is determined by multiplying the initial value of the remaining lifetime by a preset temperature correction factor. Based on the adjusted remaining lifetime estimate, the final remaining lifetime estimate of the LED module is generated.

8. The method for online monitoring and lifespan prediction of an LED module as described in claim 1, characterized in that, The step of generating a health status report based on the remaining lifetime estimate and determining the overall reliability level to support monitoring decisions includes: An offset trend sequence is obtained from the remaining lifetime estimate, the offset trend sequence containing multiple threshold comparison analysis points; Each point value in the offset trend sequence is compared with a preset threshold. If the point value exceeds the preset threshold, it is marked as an anomaly. The proportions of the anomalies in the offset trend sequence are summarized to generate a health status report; The reliability level is determined by the health status report; The reliability level judgment is corrected by taking into account the ambient temperature. The overall evaluation index is obtained by multiplying the reliability level judgment by a preset temperature coefficient. The preliminary evaluation values ​​are weighted and adjusted based on the overall evaluation indicators to form the final evaluation result that supports online monitoring decisions.

Images