A Numerical Fitting Optimization Method for LED Device Lifetime Detection

Abstract

The present invention relates to the field of semiconductor device testing, and particularly to a numerical fitting optimization method for LED device lifetime detection. The method includes: collecting the luminous flux time-series data of the LED device; performing luminous flux change analysis on the luminous flux time-series data to obtain a window adjustment factor and performing dynamic window division through the window adjustment factor; performing distance optimization factor analysis on any two dynamic windows to obtain a distance optimization factor, and optimizing the distance metric between the dynamic windows through the distance optimization factor to obtain an optimized distance; performing nearest neighbor matching through the optimized distance to obtain a nearest neighbor matching result; performing arc crossing number sequence evaluation on the nearest neighbor matching result and the luminous flux time-series data to obtain an arc crossing number sequence; performing adaptive segmentation results and piecewise fitting according to the arc crossing number sequence; performing lifetime detection analysis according to the piecewise fitting results to obtain the lifetime detection result of the LED device, thereby improving the accuracy of LED device lifetime detection.

Description

Technical Field

[0001] The invention relates to the technical field of semiconductor detection, and in particular to a numerical fitting optimization method for life detection of LED devices. Background Art

[0002] With the widespread application of LEDs (light emitting diodes) in lighting, display, and electronic equipment, how to accurately evaluate the life of LEDs has become a key issue of concern to the industry. According to the light decay characteristics of LEDs, the life of LEDs is usually measured by the time when the luminous flux decays to the initial value. As the life evaluation standard of LED. Traditional LED life detection methods are mainly based on long-term operation tests, that is, continuously lighting the LED under a specific environment and recording the change of luminous flux over time. However, this method is time-consuming and difficult to meet the needs of rapid evaluation. Therefore, the light decay model based on numerical fitting has become the mainstream life prediction method. Currently common methods include exponential decay model, Weibull distribution fitting, etc. These methods predict the overall life of LED based on the light decay trend within a certain period of time. Although these methods can provide effective predictions to a certain extent, their accuracy is affected by multiple factors, such as LED packaging characteristics, working environment, drive current, junction temperature, etc.

[0003] Existing LED life detection methods mainly rely on the exponential decay model, which assumes that the LED luminous flux decays at a fixed rate. However, in the actual detection process, the early light decay is often affected by the release of package stress, driving current fluctuations and ambient temperature changes, resulting in the light decay curve not conforming to a single exponential decay trend in the initial stage. At present, the segmented fitting method based on time series analysis has become an important means to improve the accuracy of LED life prediction. However, most of the existing segmentation methods use fixed windows or simple sliding window strategies, which cannot fully adapt to the light decay variation characteristics of different LED devices. In addition, there are problems with the alignment of time scales and local features when matching windows of different lengths, which affects the accuracy of the light decay trend. Therefore, how to construct an adaptive dynamic window division method and optimize the window matching strategy to improve the segmented fitting accuracy of LED luminous flux detection time series data is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a numerical fitting optimization method for LED device life detection to solve the problem of low precision of segmented fitting of LED luminous flux detection timing data.

[0005] An embodiment of the present invention provides a numerical fitting optimization method for LED device life detection, and the numerical fitting optimization method for LED device life detection includes the following steps:

[0006] Perform a luminous flux detection on the LED device to obtain luminous flux time series data; analyze the luminous flux change of the luminous flux time series data to obtain window adjustment factors at different times; perform dynamic window division on the luminous flux time series data through the window adjustment factors to obtain the dynamic window of the luminous flux time series data; analyze the distance optimization factor by performing distance optimization factor analysis on any two of the dynamic windows to obtain the distance optimization factor; optimize the distance metric between any two of the dynamic windows through the distance optimization factor to obtain the optimized distance between any two dynamic windows of the luminous flux time series data; perform nearest neighbor matching on the dynamic windows of the luminous flux time series data with the optimized distance to obtain a nearest neighbor matching result; evaluate the arc crossing number sequence of the nearest neighbor matching result and the luminous flux time series data to obtain an arc crossing number sequence; obtain an adaptive segmentation result of the luminous flux time series data by performing segmented evaluation on the arc crossing number sequence; obtain a segmented fitting result of the luminous flux time series data by performing segmented fitting on the adaptive segmentation result; obtain a lifetime detection result of the LED device by performing lifetime detection analysis on the segmented fitting result. Optimize the distance metric between any two of the dynamic windows through the distance optimization factor to obtain the optimized distance between any two dynamic windows of the luminous flux time series data; perform nearest neighbor matching on the dynamic windows of the luminous flux time series data with the optimized distance to obtain a nearest neighbor matching result; evaluate the arc crossing number sequence of the nearest neighbor matching result and the luminous flux time series data to obtain an arc crossing number sequence; obtain an adaptive segmentation result of the luminous flux time series data by performing segmented evaluation on the arc crossing number sequence; obtain a segmented fitting result of the luminous flux time series data by performing segmented fitting on the adaptive segmentation result; obtain a lifetime detection result of the LED device by performing lifetime detection analysis on the segmented fitting result.

[0007] Preferably, the step of analyzing the luminous flux change of the luminous flux time series data to obtain window adjustment factors at different times and performing dynamic window division on the time series data through the window adjustment factors to obtain the dynamic window of the luminous flux time series data includes:

[0008] Obtain the set initial window length and the luminous flux time series data; obtain the local average light decay rate weight factor and the local average light decay rate change rate weight factor; for any time series data point in the luminous flux time series data, use the initial window corresponding to the time series data point as the evaluation window of the time series data point, and use the calculation result of multiplying the local average light decay rate weight by the average light decay rate in the evaluation window of the time series data point as the first dynamic window evaluation factor; use the calculation result of multiplying the local average light decay rate change rate weight factor by the average light decay rate change rate in the evaluation window of the time series data point as the second dynamic window evaluation factor; use the calculation result of adding the first dynamic window evaluation factor and the second dynamic window evaluation factor as the third dynamic window evaluation factor of the time series data point; substitute the negative value of the third dynamic window evaluation factor into the exponential function with the natural constant e as the base to correspondingly obtain the window adjustment factor of the time series data point.

[0009] Use the product of the window adjustment factor of the time series data point and the set initial window length as the dynamic window length of the time series data point; obtain the dynamic window length of each time series data point in the luminous flux time series data, and perform dynamic window division on the luminous flux time series data according to the dynamic window length of each time series data point to obtain the dynamic window of the luminous flux time series data.

[0010] Preferably, obtaining a distance optimization factor by performing distance optimization factor analysis on any two of the dynamic windows includes:

[0011] Performing time-scale difference analysis on any two of the dynamic windows to obtain a time-scale matching factor between the dynamic windows of the optical flux time-series data; performing space-scale difference analysis on any two of the dynamic windows to obtain a space-scale matching factor between the dynamic windows of the optical flux time-series data; and performing fusion evaluation on the time-scale matching factor and the space-scale matching factor to obtain a distance optimization factor between any two dynamic windows of the optical flux time-series data.

[0012] Preferably, performing time-scale difference analysis on any two of the dynamic windows to obtain a time-scale matching factor between the dynamic windows of the optical flux time-series data includes:

[0013] Obtaining any two dynamic windows, taking the absolute value of the difference in the dynamic window lengths between the any two dynamic windows as the dynamic window length difference between the any two dynamic windows, substituting the opposite number of the dynamic window length difference between the any two dynamic windows into the exponential function with the natural constant e as the base as the result of the first exponential function, and using the calculation result of subtracting the result of the first exponential function from the constant 1 as the time-scale matching factor between the dynamic windows of the optical flux time-series data.

[0014] Preferably, performing space-scale difference analysis on any two of the dynamic windows to obtain a space-scale matching factor between the dynamic windows of the optical flux time-series data includes:

[0015] The calculation formula for the space-scale matching factor between the dynamic windows of the optical flux time-series data is:

[0016] ;

[0017] Where represents the space-scale matching factor between the th and the th dynamic windows of the optical flux time-series data points; respectively represent the dynamic window lengths of the th and the th dynamic windows of the optical flux time-series data points; represents the smaller dynamic window length among the dynamic window lengths of the th and the th dynamic windows of the optical flux time-series data points; represents the larger dynamic window length among the dynamic window lengths of the th and the th dynamic windows of the optical flux time-series data points; represents the The optical decay rate corresponding to the th data point in the dynamic window of the th optical flux time series data points; The optical decay rate corresponding to the th data point in the dynamic window of the th optical flux time series data points; The optical decay rate corresponding to the

[0018] th data point in the dynamic window with a larger dynamic window length among the dynamic window lengths of the

[0019] th optical flux time series data points and the

[0020] th optical flux time series data points. Preferably, the fusion evaluation by the time scale matching factor and the space scale matching factor to obtain the distance optimization factor between any two dynamic windows of optical flux time series data includes:

[0021] Obtain the time scale matching factor and the space scale matching factor between any two dynamic windows of the optical flux time series data, and use the calculation result of the product of the time scale matching factor and the space scale matching factor as the distance optimization factor between any two dynamic windows of the optical flux time series data.

[0022] Preferably, the optimization of the distance metric between any two dynamic windows by the distance optimization factor to obtain the optimized distance between any two dynamic windows of the optical flux time series data; the optimized distance performs nearest neighbor matching on the dynamic windows of the optical flux time series data to obtain the nearest neighbor matching result, including:

[0023] Obtain the distance optimization factor between any two dynamic windows of the optical flux time series data and the distance optimization factor between any two dynamic windows of the optical flux time series data; use the calculation result of multiplying the distance optimization factor between the dynamic windows of the optical flux time series data by the DTW distance between the dynamic windows of the optical flux time series data as the optimized distance between any two dynamic windows of the optical flux time series data;

[0024] Preferably, the nearest neighbor matching result and the optical flux time series data are evaluated for the arc crossing number sequence to obtain the arc crossing number sequence, including:

[0024] Obtain the nearest neighbor matching results corresponding to the dynamic windows of each data point in the optical flux time series data; for any moment in the optical flux time series data, take the number of nearest neighbor matching relationships corresponding to both sides of this moment as the arc crossing number corresponding to this moment, and obtain an arc crossing number sequence according to the arc crossing number corresponding to each moment in the optical flux time series data.

[0025] Preferably, the adaptive segmentation result of the optical flux time series data is obtained by segmentally evaluating the arc crossing number sequence, including:

[0026] Obtain the segmentation degree threshold of the optical flux time series data; take the linear normalization result of the arc crossing number corresponding to each moment of the optical flux time series data in the arc crossing number sequence as the segmentation degree corresponding to each moment; compare the segmentation degree corresponding to each moment with the segmentation degree threshold of the optical flux time series data, and mark the moment as a segmentation point if it is higher than the segmentation degree threshold; obtain all the segmentation points in the optical flux time series data and segment the optical flux time series data according to the segmentation points to obtain the adaptive segmentation result of the optical flux time series data.

[0027] Preferably, the segmented fitting result of the optical flux time series data is obtained by segmentally fitting the adaptive segmentation result; the lifetime detection result of the LED device is obtained by performing lifetime detection analysis on the segmented fitting result, including:

[0028] Obtain the adaptive segmentation result of the optical flux time series data, perform exponential function fitting on each segmentation result using the least squares method to obtain the segmented fitting result of the optical flux time series data; perform cumulative time evaluation on the optical flux of the LED device dropping to the initial optical flux according to the segmented fitting result of the optical flux time series data when, and take this cumulative time as the lifetime detection result of the LED device.

[0029] The beneficial effects of the embodiments of the present invention compared with the prior art are:

[0030] The present invention performs optical flux detection on an LED device to obtain optical flux time series data; performs optical flux change analysis on the optical flux time series data to obtain window adjustment factors at different moments; divides the optical flux time series data into dynamic windows through the window adjustment factors to obtain the dynamic windows of the optical flux time series data; performs distance optimization factor analysis on any two of the dynamic windows to obtain a distance optimization factor; through the distance optimization factor between any two of the dynamic windows The distance metric is optimized to obtain the optimized distance between any two dynamic windows of the luminous flux time series data; the optimized distance is used for nearest neighbor matching of the dynamic windows of the luminous flux time series data to obtain a nearest neighbor matching result; the nearest neighbor matching result and the luminous flux time series data are evaluated for the arc crossing number sequence to obtain an arc crossing number sequence; by performing segmented evaluation on the arc crossing number sequence, an adaptive segmentation result of the luminous flux time series data is obtained; by performing segmented fitting on the adaptive segmentation result, a segmented fitting result of the luminous flux time series data is obtained; by performing lifetime detection analysis on the segmented fitting result, a lifetime detection result of the LED device is obtained. Among them, the luminous flux time series data points are divided into dynamic windows through a window adjustment factor, so that the nearest neighbor matching relationship between the time series data windows can adapt to the light decay stages of different LEDs, improving the segmented accuracy of the luminous flux time series data and solving the problem that a fixed window length cannot adapt to the change of the light decay rate; after the dynamic window division, the distance optimization factor is further evaluated through the time scale matching factor and the space scale matching factor between the dynamic windows to solve the scale problem in the nearest neighbor matching process of different dynamic window lengths, ensuring the matching accuracy of short windows and long windows, thereby improving the adaptive segmentation accuracy of the luminous flux time series data, and thus improving the segmented fitting accuracy of the luminous flux time series data. Brief Description of the Drawings

[0031] The drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0032] Figure 1 It is a flowchart of a numerical fitting optimization method for LED device lifetime detection according to an embodiment of the present invention; Detailed Embodiments

[0033] The embodiments of the present disclosure are described in detail below. The examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present disclosure and should not be construed as a limitation to the present disclosure.

[0034] It should be noted that the terms "first", "second", etc. in the specification of the present disclosure and the above drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that the data used in this toothpaste can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described here can be implemented in an order other than those illustrated or described here. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0035] To illustrate the technical solution of the present invention, specific embodiments are used for illustration below.

[0036] The specific scenario targeted by the present invention is: during the life detection of LED devices, adaptive segmented fitting optimization is performed on the luminous flux time series data of LED devices.

[0037] In addition, the light decay mentioned in the embodiments of the present invention refers to the attenuation of luminous flux.

[0038] See Figure 1 , which is a flowchart of a numerical fitting optimization method for LED device life detection provided in Embodiment 1 of the present invention. As Figure 1 shown, a numerical fitting optimization method for LED device life detection may include:

[0039] Step S101, perform a luminous flux detection on the LED device to obtain luminous flux time series data.

[0040] First, establish a stable test environment to ensure the accuracy and repeatability of the luminous flux data. Usually, the combination of an integrating sphere and a spectrometer is used for measurement. Among them, the integrating sphere is used to uniformly collect the light emitted by the LED, and the spectrometer is used to accurately measure the luminous flux and record its change over time. During the detection process, the LED device is continuously lit under a constant current drive, and the test system collects the luminous flux data of the LED at a set time interval. In the embodiments of the present invention, the set time interval is 1 hour, so as to obtain the luminous flux time series data.

[0041] Step S102, through the analysis of the luminous flux change of the luminous flux time series data, obtain the window adjustment factor at different times, and perform dynamic window division on the luminous flux time series data through the window adjustment factor to obtain the dynamic window of the luminous flux detection time series data.

[0042] After obtaining the luminous flux time series data, the luminous flux time series data can be segmented for segmented fitting analysis. In the adaptive segmentation process of the LED luminous flux detection time series data, the time series data can be segmented by the Fast Semantic Segmentation of Time Series (FLOSS) algorithm. However, in the actual scenario, the luminous flux attenuation process of the LED device is not a simple exponential decline, but is affected by various factors. These factors may include that the luminous flux changes slowly in the early stage, affected by the release of packaging stress and the increase of initial temperature, and the attenuation trend is not obvious; in the transition stage, the light decay rate gradually increases, and the luminous flux attenuation gradually approaches the stable model; in the long-term stable attenuation stage, the LED undergoes stable attenuation, conforming to the exponential attenuation model; in the mutation stage, the luminous flux may experience local mutations at specific time points due to the influence of high temperature, overload, or current fluctuations. Since these change stages are not completely fixed and each LED may have different light decay characteristics, using a fixed time series data window for segmenting the LED luminous flux detection data may lead to problems such as the loss of key points of the light decay trend, the failure of matching between short-term and long-term windows, and inaccurate segmentation caused by mis-matching between the luminous flux time series data windows.

[0043] Therefore, in the adaptive segmentation process of the LED luminous flux detection data, the core problem to be solved first is how to determine an appropriate window size to adapt to the change of the LED light decay rate, rather than using a fixed window division. First, by analyzing the change of the luminous flux in the luminous flux time series data, the window adjustment factor at different times is obtained, and the time series data is dynamically windowed by the window adjustment factor to obtain the dynamic window of the luminous flux time series, including:

[0044] Obtain the set initial window length and the luminous flux time series data; obtain the local average light decay rate weight factor and the local average light decay rate change rate weight factor; for any time series data point in the luminous flux time series data, take the initial window corresponding to this time series data point as the evaluation window of this time series data point, and take the calculation result of multiplying the local average light decay rate weight by the local average light decay rate in the evaluation window of this time series data point as the first dynamic window evaluation factor; take the calculation result of multiplying the local average light decay rate change rate weight factor by the average light decay rate change rate in the evaluation window of this time series data point as the second dynamic window evaluation factor; take the calculation result of adding the first dynamic window evaluation factor and the second dynamic window evaluation factor as the third dynamic window evaluation factor of this time series data point; substitute the opposite number of the third dynamic window evaluation factor into the exponential function with the natural constant e as the base to obtain the window adjustment factor corresponding to this time series data point;

[0045] Multiply the window adjustment factor of the time-series data points by the set initial window length as the dynamic window length of the time-series data points; obtain the dynamic window length of each time-series data point in the luminous flux time-series data, and perform dynamic window division on the luminous flux time-series data according to the dynamic window length of each time-series data point to obtain the dynamic window of the luminous flux time-series data.

[0046] In one embodiment, assume that the local average light decay rate in the dynamic window of the -th time-series data point is , and the change rate of the local average light decay rate in the dynamic window of the -th time-series data point is . Then the calculation expression of the window adjustment factor of the -th time-series data point is:

[0047] ;

[0048] Where, represents the window adjustment factor of the -th time-series data point; represents the local average light decay rate in the dynamic window of the -th time-series data point; represents the change rate of the local average light decay rate in the dynamic window of the -th time-series data point; represents the local average light decay rate weight factor; represents the change rate of the local average light decay rate weight factor; represents the natural constant.

[0049] It should be noted that the trend of LED luminous flux attenuation is not uniform attenuation, but shows different change patterns over time. In the early stage of the luminous flux time-series data, the luminous flux attenuation is slow and the local change is small, which is suitable for segmenting with a larger window to reduce the influence of short-term fluctuations. In the exponential attenuation stage of the luminous flux time-series data, the light attenuation rate gradually accelerates. If a large window is still used, the key turning points of the light attenuation trend will be ignored. Therefore, the window should be reduced to more precisely capture the light attenuation change. In the case of mutations in the luminous flux time-series data, when the LED is affected by external factors such as temperature and current, the luminous flux may have local mutations. At this time, the window should be reduced to improve the detection sensitivity and ensure that the mutation information is not averaged. Therefore, it is necessary to dynamically adjust the window through the local window light attenuation rate and the change rate of the light attenuation rate of the luminous flux time-series data points. When the local average light attenuation rate of the luminous flux time-series data points is large, it indicates that the light attenuation rate is fast, and the window should be reduced to improve the time resolution. When the local average light attenuation rate of the luminous flux time-series data points is small, it indicates that the light attenuation is slow, and the window should be increased to reduce unnecessary short-term fluctuation interference. When the change rate of the local average light attenuation rate of the luminous flux time-series data points is large, it indicates that the light attenuation rate is changing rapidly, and the window should be reduced to accurately capture the change trend. When the change rate of the local average light attenuation rate of the luminous flux time-series data points is small, it indicates that the light attenuation rate is relatively stable, and the window can be appropriately increased to smooth the data. In the embodiment of the present invention, the local average light attenuation rate weight factor and the local average light attenuation rate change rate weight factor are both set to , so as to equally refer to the local average light attenuation rate and the local light attenuation rate change rate. The local average light attenuation rate weight factor and the local average light attenuation rate change rate weight factor can be adjusted according to the actual scenario, and there is no requirement.

[0050] In one embodiment, assuming that the set initial window length is , then the calculation expression for the dynamic window length of the th time-series data point is:

[0051] ;

[0052] Among them, represents the dynamic window length of the th time-series data point; represents the window adjustment factor of the th time-series data point; represents the initial window length.

[0053] It should be noted that in the embodiment of the present invention, the initial window length is set to , and the initial window length can be adjusted according to the actual scenario, and there is no requirement.

[0054] Step S103: By performing distance optimization factor analysis on any two dynamic windows, a distance optimization factor is obtained. The DTW distance metric between any two dynamic windows is optimized by the distance optimization factor to obtain an optimized distance.

[0055] After obtaining the dynamic window partitioning result of the LED luminous flux detection timing data, the window nearest neighbor matching process can be performed through the dynamic windows of the LED luminous flux detection timing data, thereby completing the matrix profile matching of the timing data. In this process, it is necessary to determine the nearest neighbor window of each timing data window among all the timing data windows and form a matching relationship with the nearest neighbor window, so as to perform adaptive segmentation of the LED luminous flux detection timing data through the matching relationship. During the matching process, it is necessary to measure the distance between dynamic length windows. However, during the process of measuring the distance between two dynamic length windows, there are differences in window lengths between the dynamic windows. At the same time, in two dynamic windows, the short window may be affected by local fluctuations due to a small time span, while the long window may ignore local details due to a large time span, resulting in matching errors. To ensure the accuracy of the nearest neighbor matching of the luminous flux timing data dynamic windows, it is necessary to perform time scale difference analysis on any two of the dynamic windows to obtain the time scale matching factor between the luminous flux timing data dynamic windows and perform spatial scale difference analysis on any two of the dynamic windows to obtain the spatial scale matching factor between the luminous flux timing data dynamic windows. Finally, the time scale matching factor and the spatial scale matching factor are fused and evaluated to obtain the distance optimization factor between any two luminous flux timing data dynamic windows, so as to optimize the distance evaluation between the dynamic windows through the distance optimization factor.

[0056] First, by performing time scale difference analysis on any two of the dynamic windows, the time scale matching factor between the luminous flux timing data dynamic windows is obtained, including:

[0057] Obtain any two dynamic windows. Take the absolute value of the difference in dynamic window lengths between the two dynamic windows as the dynamic window length difference between the two dynamic windows. Substitute the opposite number of the dynamic window length difference between the two dynamic windows into the exponential function with the natural constant e as the base as the result of the first exponential function, and use the result of subtracting the result of the first exponential function from the constant 1 as the time scale matching factor between the luminous flux timing data dynamic windows.

[0058] In one embodiment, the th and

[0059] ;

[0060] Among them, represents the th time scale matching factor between the dynamic window of the th optical flux time series data points; represents the window length of the dynamic window of the th optical flux time series data points; represents the

[0061] After obtaining the time scale matching factor between the dynamic windows of the optical flux time series data, the spatial scale matching factor between the dynamic windows of the optical flux time series data can be obtained by performing spatial scale difference analysis on any two of the said dynamic windows.

[0062] In one embodiment, the calculation formula for the spatial scale matching factor between the th and the

[0063] th

[0064] optical flux time series data points is: represents the th and the respectively represent the th window lengths of the dynamic windows of the and the th optical flux time series data points; represents the th window length of the dynamic window with the smaller window length among the dynamic window lengths of the and the th optical flux time series data points; represents the th window length of the dynamic window with the larger window length among the dynamic window lengths of the represents the th The optical decay rate corresponding to the th data point in the dynamic window with a larger dynamic window length among the dynamic window lengths of

[0065] It should be noted that to solve this problem, we introduce a time-scale matching factor and a space-scale matching factor to optimize the calculation of the distance between windows. The time-scale matching factor is used to measure whether there is a significant difference in the time span of two windows, and the similarity calculation between windows is adjusted by normalization, so that windows of different lengths can be matched on the same scale. The space-scale matching factor is used to evaluate whether the optical decay trends inside the windows are consistent, ensuring that the matching of short windows in long windows will not lead to incorrect matching due to different local trends.

[0066] After obtaining the time-scale matching factor and the space-scale matching factor, it is necessary to perform a fusion evaluation through the time-scale matching factor and the space-scale matching factor to obtain a distance optimization factor between any two dynamic windows of optical flux time series data, including:

[0067] Obtain the time-scale matching factor and the space-scale matching factor between any two of the dynamic windows of the optical flux time series data, and use the calculation result of the product of the time-scale matching factor and the space-scale matching factor as the distance optimization factor between any two of the dynamic windows of the optical flux time series data.

[0068] In one embodiment, the calculation expression of the distance optimization factor between the th and the th dynamic windows of the optical flux time series data is:

[0069] ;

[0070] Wherein, represents the distance optimization factor between the th and the th dynamic windows of the optical flux time series data; represents the time-scale matching factor between the th and the th dynamic windows of the optical flux time series data; represents the space-scale matching factor between the th and the th dynamic windows of the optical flux time series data.

[0071] After obtaining the distance optimization factor, optimize the distance metric between any two of the dynamic windows through the distance optimization factor to obtain the optimized distance between any two of the dynamic windows of the optical flux time series data, including:

[0072] ​Obtain the distance optimization factor between any two of the dynamic windows of the luminous flux time series data; use the calculation result of multiplying the distance optimization factor between the dynamic windows of the luminous flux time series data by the DTW distance between the dynamic windows of the luminous flux time series data as the optimized distance between any two of the dynamic windows of the luminous flux time series data.

[0073] In one embodiment, assume that the th and the th dynamic windows of the luminous flux time series data points, the distance is , then the calculation expression for the optimized distance between the th and the th dynamic windows of the luminous flux time series data points is:

[0074] ;

[0075] where, represents the optimized distance between the th and the th dynamic windows of the luminous flux time series data points; represents the distance optimization factor between the th and the th dynamic windows of the luminous flux time series data points; represents the th and the th dynamic windows of the luminous flux time series data points distance.

[0076] It should be noted that by using the distance optimization factor to weight and optimize the DTW distance between the long and short windows, the matching calculation can not only consider the consistency of the global trend but also ensure the alignment of the local light decay patterns, thereby improving the accuracy of the segmentation of the luminous flux time series data and ensuring that the true characteristics of the LED light decay process can be accurately captured.

[0077] Step S104, perform nearest neighbor matching on the dynamic windows of the luminous flux time series data through the optimized distance, and evaluate the arc crossing number sequence according to the nearest neighbor matching result to obtain the arc crossing number sequence; perform adaptive segmentation on the luminous flux time series data according to the arc crossing number sequence to obtain the adaptive segmentation result.

[0078] After obtaining the optimized distance between any two dynamic windows of the luminous flux time series data, nearest neighbor matching can be performed on the dynamic windows of the luminous flux time series data through the optimized distance to obtain the nearest neighbor matching result, including:

[0079] For each data point in the luminous flux time series data, obtain the optimized distance between the dynamic window corresponding to this data point and the dynamic window corresponding to any other data point, and use the dynamic window corresponding to any other data point with the smallest optimized distance from the dynamic window corresponding to this data point as the near-neighbor matching result for this data point.

[0080] After obtaining the near-neighbor matching result, further, perform an arc crossing number sequence evaluation on the near-neighbor matching result and the luminous flux time series data to obtain an arc crossing number sequence, including:

[0081] Obtain the near-neighbor matching result corresponding to the dynamic window of each data point in the luminous flux time series data; for any moment in the luminous flux time series data, use the number of near-neighbor matching relationships corresponding to both sides of this moment as the arc crossing number corresponding to this moment, and obtain an arc crossing number sequence according to the arc crossing number corresponding to each moment in the luminous flux time series data.

[0082] After obtaining the arc crossing number sequence, perform a segmented evaluation on the arc crossing number sequence to obtain an adaptive segmentation result of the luminous flux time series data, including:

[0083] Obtain the segmentation degree threshold of the luminous flux time series data; use the linear normalization result of the arc crossing number corresponding to each moment of the luminous flux time series data in the arc crossing number sequence as the segmentation degree corresponding to each moment; compare the segmentation degree corresponding to each moment with the segmentation degree threshold of the luminous flux time series data, and if it is higher than the segmentation degree threshold, mark this moment as a segmentation point; obtain all the segmentation points in the luminous flux time series data and segment the luminous flux time series data according to the segmentation points to obtain an adaptive segmentation result of the luminous flux time series data.

[0084] It should be noted that in the embodiments of the present invention, the segmentation degree threshold of the luminous flux time series data is set to , which can be adjusted according to the actual scenario and is not required.

[0085] Step S105, perform segmented fitting on the adaptive segmentation result to obtain a segmented fitting result of the luminous flux detection time series data and perform lifetime detection analysis through the segmented fitting result to obtain a lifetime detection result of the LED device.

[0086] After obtaining the adaptive segmentation result of the luminous flux time series data, perform exponential function fitting on each segmentation result using the least squares method to obtain the segmented fitting result of the luminous flux time series data, and perform cumulative time evaluation on the LED device when the luminous flux drops to the initial luminous flux according to the segmented fitting result of the luminous flux time series data and use this cumulative time evaluation as the lifetime detection result of the LED device.

[0087] In summary, in the embodiment of the present invention, the luminous flux of the LED device is detected to obtain the luminous flux time series data; by analyzing the change of the luminous flux in the luminous flux time series data, the window adjustment factor at different times is obtained; the luminous flux time series data is dynamically windowed by the window adjustment factor to obtain the dynamic window of the luminous flux time series data; by analyzing the distance optimization factor for any two of the dynamic windows, the distance optimization factor is obtained; the distance metric between any two of the dynamic windows is optimized by the distance optimization factor to obtain the optimized distance between any two dynamic windows of the luminous flux time series data; the optimized distance performs a nearest neighbor match on the dynamic windows of the luminous flux time series data to obtain a nearest neighbor match result; the nearest neighbor match result and the luminous flux time series data are evaluated for the arc crossing number sequence to obtain the arc crossing number sequence; by segmenting and evaluating the arc crossing number sequence, the adaptive segmentation result of the luminous flux time series data is obtained; by segment fitting the adaptive segmentation result, the segment fitting result of the luminous flux time series data is obtained; by performing a lifetime detection analysis on the segment fitting result, the lifetime detection result of the LED device is obtained. Among them, the luminous flux time series data points are dynamically windowed by the window adjustment factor, so that the nearest neighbor matching relationship between the time series data windows can adapt to the light decay stages of different LEDs, improving the segmentation accuracy of the luminous flux time series data and solving the problem that the fixed window length cannot adapt to the change of the light decay rate; after the dynamic windowing, the distance optimization factor is further evaluated by the time scale matching factor and the space scale matching factor between the dynamic windows to solve the scale problem in the nearest neighbor matching process of different dynamic window lengths, ensuring the matching accuracy of short windows and long windows, thereby improving the adaptive segmentation accuracy of the luminous flux time series data and thus improving the segment fitting accuracy of the luminous flux time series data. The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

[0088] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A numerical fitting optimization method for LED device life detection, characterized in that: include: Performing luminous flux detection on the LED device to obtain luminous flux time series data; performing luminous flux change analysis on the luminous flux time series data to obtain window adjustment factors at different times; Dynamically window the luminous flux time series data using the window adjustment factor to obtain the dynamic window of the luminous flux time series data; perform distance optimization factor analysis on any two of the dynamic windows to obtain the distance optimization factor; optimize the DTW distance metric between any two of the dynamic windows using the distance optimization factor to obtain the optimized distance between any two of the dynamic windows of the luminous flux time series data; The optimization distance performs nearest neighbor matching on the dynamic window of the luminous flux time series data to obtain the nearest neighbor matching result; the nearest neighbor matching result and the luminous flux time series data are evaluated on the arc span number sequence to obtain the arc span number sequence; the arc span number sequence is evaluated in segments to obtain the adaptive segmentation result of the luminous flux time series data; the adaptive segmentation result is fitted in segments to obtain the segmentation fitting result of the luminous flux time series data; the segmentation fitting result is analyzed for life detection to obtain the life detection result of the LED device.

2. A numerical fitting optimization method for LED device life detection according to claim 1, characterized in that: According to the analysis of the luminous flux change on the luminous flux time series data, the window adjustment factors at different times are obtained; The time series data is dynamically divided into windows by the window adjustment factor to obtain the dynamic window of the luminous flux time series data. The specific steps include: Acquire the set initial window length and luminous flux time series data; acquire the local average light decay rate weight factor and the local average light decay rate change rate weight factor; for any time series data point in the luminous flux time series data, use the initial window corresponding to the time series data point as the evaluation window of the time series data point, and use the local average light decay rate weight multiplied by the average light decay rate in the evaluation window of the time series data point as the first dynamic window evaluation factor; use the local average light decay rate change rate weight factor multiplied by the average light decay rate change rate in the evaluation window of the time series data point as the second dynamic window evaluation factor; use the first dynamic window evaluation factor and the second dynamic window evaluation factor as the third dynamic window evaluation factor of the time series data point; substitute the opposite of the third dynamic window evaluation factor into the exponential function with the natural constant e as the base, and obtain the window adjustment factor of the time series data point; The product of the window adjustment factor of the time series data point and the set initial window length is used as the dynamic window length of the time series data point; the dynamic window length of each time series data point in the luminous flux time series data point is obtained, and the dynamic window of the luminous flux time series data is divided according to the dynamic window length of each time series data point to obtain the dynamic window of the luminous flux time series data.

3. A numerical fitting optimization method for LED device life detection according to claim 1, characterized in that: According to the distance optimization factor analysis performed on any two dynamic windows to obtain the distance optimization factor, the specific steps include: By performing a time scale difference analysis on any two of the dynamic windows, the time scale matching factor between the dynamic windows of the luminous flux time series data is obtained; by performing a spatial scale difference analysis on any two of the dynamic windows, the spatial scale matching factor between the dynamic windows of the luminous flux time series data is obtained; by performing a fusion evaluation of the time scale matching factor and the spatial scale matching factor, the distance optimization factor between any two dynamic windows of the luminous flux time series data is obtained.

4. A numerical fitting optimization method for LED device life detection according to claim 3, characterized in that: According to the method of performing time scale difference analysis on any two dynamic windows to obtain the time scale matching factor between the dynamic windows of the luminous flux time series data, the specific steps include: Obtain any two dynamic windows, take the absolute value of the dynamic window length difference between the any two dynamic windows as the dynamic window length difference between the any two dynamic windows, substitute the opposite of the dynamic window length difference between the any two dynamic windows into an exponential function with the natural constant e as the base as the first exponential function result, and use the constant 1 minus the calculation result of the first exponential function as the time scale matching factor between the dynamic windows of the luminous flux timing data.

5. A numerical fitting optimization method for LED device life detection according to claim 3, characterized in that: According to the method of performing spatial scale difference analysis on any two dynamic windows to obtain the spatial scale matching factor between the dynamic windows of the luminous flux time series data, the specific steps include: The calculation formula of the spatial scale matching factor between the dynamic windows of the luminous flux time series data is: where ξ i,j represents the spatial scale matching factor between the dynamic windows of the ith and jth luminous flux time series data points; ω i ,ω j Respectively represent the dynamic window lengths of the i-th and j-th luminous flux time series data points; Indicates the smaller dynamic window length of the dynamic window lengths of the i-th and j-th light flux time series data points; represents the larger dynamic window length of the dynamic window lengths of the i-th and j-th luminous flux time series data points; R i,m Represents the light decay rate corresponding to the mth data point in the dynamic window of the i-th light flux time series data point; R j,m Represents the light decay rate corresponding to the mth data point in the dynamic window of the jth light flux time series data point; It represents the light decay rate corresponding to the mth data point in the dynamic window with the larger dynamic window length among the dynamic window lengths of the i-th and j-th light flux time series data points.

6. A numerical fitting optimization method for LED device life detection according to claim 3, characterized in that: According to the fusion evaluation of the time scale matching factor and the space scale matching factor, the distance optimization factor between any two dynamic windows of the light flux time series data is obtained, and the specific steps include: The time scale matching factor and the space scale matching factor between any two dynamic windows of the luminous flux time series data are obtained, and the calculation result of the product of the time scale matching factor and the space scale matching factor is used as the distance optimization factor between any two dynamic windows of the luminous flux time series data.

7. A numerical fitting optimization method for LED device life detection according to claim 1, characterized in that: Optimizing the DTW distance metric between any two of the dynamic windows by the distance optimization factor to obtain an optimized distance between any two of the dynamic windows of the light flux time series data; The optimization distance performs nearest neighbor matching on the dynamic window of the luminous flux time series data to obtain the nearest neighbor matching result, and the specific steps include: Obtaining a distance optimization factor between any two of the luminous flux time series data dynamic windows and any two of the luminous flux time series data dynamic windows; The calculation result of multiplying the distance optimization factor between the dynamic windows of the light flux time series data by the DTW distance between the dynamic windows of the light flux time series data is used as the optimized distance between any two dynamic windows of the light flux time series data; For each data point in the luminous flux time series data, the optimized distance between the dynamic window corresponding to the data point and the dynamic window corresponding to any other data point is obtained, and the dynamic window corresponding to any other data point with the smallest optimized distance between the dynamic window corresponding to the data point is taken as the nearest neighbor matching result of the data point.

8. A numerical fitting optimization method for LED device life detection according to claim 1, characterized in that: The arc spanning number sequence is evaluated according to the nearest neighbor matching result and the luminous flux time series data to obtain the arc spanning number sequence, and the specific steps include: Obtain the nearest neighbor matching result corresponding to the dynamic window of each data point in the luminous flux time series data; for any moment in the luminous flux time series data, take the number of corresponding nearest neighbor matching relationships on both sides of the moment as the arc span number corresponding to the moment, and obtain the arc span number sequence according to the arc span number corresponding to each moment in the luminous flux time series data.

9. A numerical fitting optimization method for LED device life detection according to claim 1, characterized in that: According to the method of performing segmented evaluation on the arc span number sequence to obtain the adaptive segmented result of the luminous flux time series data, the specific steps include: Obtain a segmentation degree threshold of the luminous flux time series data; use the linear normalization result of the arc span number corresponding to each moment of the luminous flux time series data in the arc span number sequence as the segmentation degree corresponding to each moment; compare the segmentation degree corresponding to each moment with the segmentation degree threshold of the luminous flux time series data, and mark the moment as a segmentation point if it is higher than the segmentation degree threshold; obtain all the segmentation points in the luminous flux time series data and segment the luminous flux time series data according to the segmentation points to obtain an adaptive segmentation result of the luminous flux time series data.

10. A numerical fitting optimization method for LED device life detection according to claim 1, characterized in that: The step of performing segmented fitting on the adaptive segmented results to obtain segmented fitting results of the luminous flux time series data; and performing life detection analysis on the segmented fitting results to obtain life detection results of the LED device include: Obtain an adaptive segmentation result of the luminous flux timing data, perform exponential function fitting on each segmentation result using the least squares method, and obtain a segmentation fitting result of the luminous flux timing data; evaluate the cumulative time when the luminous flux of the LED device drops to 70% of the initial luminous flux based on the segmentation fitting result of the luminous flux timing data, and use the cumulative time as the L70 life detection result of the LED device.

Images