Method and system for testing stability of flexible optoelectronic display material

Abstract

The application provides a stability test method and system of flexible optoelectronic display material, relates to the technical field of stability evaluation, and the method comprises the following steps: collecting a working condition application data sequence and a measured performance degradation time curve in real time; inputting a degradation confidence prediction network for prediction; obtaining a comprehensive confidence score; when the spatial confidence does not meet the preset requirement, extracting the working condition application data sequence corresponding to the minimum value of the comprehensive confidence score as a target point, generating a plurality of candidate continuous working condition change paths; and optimizing the test process by automatically adjusting the test path and conditions until the preset confidence requirement is met to obtain a stability test evaluation result. The application solves the technical problem that in the prior art, due to the fixed test working condition and the disconnection with the real scene, there is a large amount of information blind test and resource waste in the test process, thereby affecting the stability test efficiency, so that the most comprehensive degradation data is obtained in the least test period, and the stability test efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of stability assessment technology, specifically to stability testing methods and systems for flexible optoelectronic display materials. Background Technology

[0002] Flexible optoelectronic display materials have broad application prospects in foldable and rollable display devices, and their stability directly determines the lifespan and display performance of the devices. Currently, stability testing of these materials typically employs preset fixed-condition sequences or simple combined aging tests, usually with open-loop control. Testers must pre-set all stress application paths, and cannot dynamically adjust subsequent conditions based on the material's actual degradation behavior during testing. Flexible optoelectronic display materials experience multi-field coupled composite stresses during actual service, and failure modes and sensitive stress regions differ significantly between different material systems and under different process conditions. Existing preset testing strategies relying on human experience struggle to adaptively explore the entire operating range, resulting in insufficient coverage of critical failure areas or repeated testing of already well-understood areas. This not only leads to lengthy testing cycles and wasted resources but may also miss actual failure modes, thus affecting the accuracy and reliability of stability assessments.

[0003] In summary, existing technologies suffer from technical problems such as fixed test conditions that are disconnected from real-world scenarios, leading to a large amount of blind testing and wasted resources, which in turn affects the efficiency of stability testing. Summary of the Invention

[0004] The purpose of this application is to provide a stability testing method and system for flexible optoelectronic display materials, in order to solve the technical problem in the prior art that the fixed test conditions are out of touch with real-world scenarios, resulting in a large amount of blind testing and wasted resources, which in turn affects the efficiency of stability testing.

[0005] To achieve the above objectives, this application provides a method and system for testing the stability of flexible optoelectronic display materials.

[0006] Firstly, this application provides a stability testing method for flexible optoelectronic display materials. The stability testing method is implemented using a stability testing system for flexible optoelectronic display materials. The method includes: mounting a sample of the flexible optoelectronic display material onto a multi-field coupling test platform; applying mechanical and environmental conditions to the multi-field coupling test platform according to a preset test scenario; and real-time acquiring of the sample's luminance, resistance, and film adhesion signals during the application process to obtain a real-time application data sequence and a real-time measured performance degradation time-series curve; inputting the real-time application data sequence and the measured performance degradation time-series curve into a degradation confidence prediction network to obtain a degradation confidence prediction result, wherein the degradation confidence prediction network is trained in a self-supervised manner; and applying the degradation confidence... The prediction results are weighted to obtain a comprehensive confidence score, which is then bound to real-time measured performance degradation time-series curves and real-time operating condition application data sequences and added to the operating condition space. When the spatial confidence of the operating condition space does not meet the preset requirements, the operating condition application data sequence corresponding to the minimum comprehensive confidence score in the operating condition space is extracted as the target point. Starting from the real-time operating condition application data sequence, multiple candidate continuous operating condition change paths pointing to the target point are generated. The path confidence evaluation network selects the path with the highest information gain from the multiple candidate continuous operating condition change paths to drive the test platform to execute automatically. This process is repeated until the spatial confidence of the operating condition space meets the preset requirements. Based on the multiple measured performance degradation time-series curves stored in the operating condition space, the stability of the flexible optoelectronic display material is evaluated to obtain the stability test evaluation results.

[0007] Optionally, the degradation confidence prediction results include the luminance prediction confidence interval, resistance prediction confidence interval, and film adhesion prediction confidence interval for multiple future test cycles.

[0008] Optionally, multiple historical data points of samples that have completed full testing are acquired, where each historical data point contains multiple operating condition application data sequences and multiple performance degradation time series curves. Each historical data point is divided into an early stage and a late stage according to the test cycle, where the late stage contains a preset number of consecutive test cycles. Only the multiple operating condition application data sequences and multiple performance degradation time series curves of the early stage are used as network inputs, and the network outputs the luminescence brightness prediction confidence interval, resistance value prediction confidence interval, and film adhesion prediction confidence interval for each test cycle in the late stage. The luminescence brightness prediction confidence interval, resistance value prediction confidence interval, and film adhesion prediction confidence interval of each test cycle output by the network are compared with the actual data of the late stage, and the ratio of the prediction interval coverage to the average interval width is calculated. The negative of this ratio is used as a self-supervised loss function, and the network is trained by minimizing this loss function until convergence, thus obtaining a trained degradation confidence prediction network.

[0009] Optionally, a first fusion weight corresponding to the luminance, a second fusion weight corresponding to the resistance value, and a third fusion weight corresponding to the film adhesion are determined based on the layer structure position of the flexible optoelectronic display material; the degradation confidence prediction result is weighted based on the first fusion weight, the second fusion weight, and the third fusion weight to obtain a comprehensive confidence score.

[0010] Optionally, the half-width of the luminance prediction confidence interval, the half-width of the resistance prediction confidence interval, and the half-width of the film adhesion prediction confidence interval are obtained for each future test cycle; the three half-widths of each test cycle are multiplied by the corresponding first fusion weight, second fusion weight, and third fusion weight, respectively, and then summed to obtain the weighted half-width of that cycle; the average of the multiple weighted half-widths of multiple future test cycles is calculated, and the average value is compared with the preset maximum allowable half-width to obtain a comprehensive confidence score.

[0011] Optionally, the first fusion weight is the same as the second fusion weight, and the first fusion weight and the second fusion weight are respectively greater than the third fusion weight.

[0012] Optionally, starting from the real-time operating condition applied data sequence, a path search algorithm is used to generate multiple candidate continuous operating condition change paths with the endpoint as the target point; wherein, each candidate continuous operating condition change path contains a preset number of test cycles, and the bending radius change in each test cycle is sampled from the first change step set, the stretching percentage change is sampled from the second change step set, the temperature change is sampled from the third change step set, and the humidity change is sampled from the fourth change step set.

[0013] Optionally, a gated recurrent unit network is constructed; multiple executed test paths are extracted from historical test data, and the average change in the comprehensive confidence of the data sequence of applied working conditions stored in the working condition space before and after the execution of each test path is calculated. The average change value is used as the label of the path to obtain the test dataset; using the test dataset, the gated recurrent unit network is trained with regression loss until convergence to obtain the path confidence evaluation network.

[0014] Optionally, the input to the gated loop unit network is a sequence of bending radius, stretching percentage, temperature, and humidity of a candidate continuous operating condition change path, and the output is the average change in overall confidence expected after executing the path.

[0015] Secondly, this application also provides a stability testing system for flexible optoelectronic display materials, used to perform the stability testing method for flexible optoelectronic display materials as described in the first aspect. The stability testing system for flexible optoelectronic display materials includes: a data acquisition module for mounting a sample of the flexible optoelectronic display material onto a multi-field coupling test platform, wherein the multi-field coupling test platform applies mechanical and environmental conditions according to preset test scenario requirements, and during the application process, acquires the luminous intensity, resistance value, and film adhesion signals of the sample in real time to obtain a real-time application data sequence and a real-time measured performance degradation time-series curve; a degradation confidence prediction module for inputting the real-time application data sequence and the measured performance degradation time-series curve into a degradation confidence prediction network to obtain a degradation confidence prediction result, wherein the degradation confidence prediction network is trained in a self-supervised manner; and an uncertainty quantification module for quantifying the degradation confidence. The prediction results are weighted to obtain a comprehensive confidence score, which is then bound to the real-time measured performance degradation time-series curve and the real-time operating condition application data sequence and added to the operating condition space. The operating condition change path generation module is used to extract the operating condition application data sequence corresponding to the minimum comprehensive confidence score in the operating condition space as the target point when the spatial confidence of the operating condition space does not meet the preset requirements. Starting from the real-time operating condition application data sequence, multiple candidate continuous operating condition change paths are generated pointing to the target point. The stability evaluation module is used to select the path with the highest information gain from the multiple candidate continuous operating condition change paths through a path confidence evaluation network, driving the test platform to execute automatically. This process continues until the spatial confidence of the operating condition space meets the preset requirements. Based on the multiple measured performance degradation time-series curves stored in the operating condition space, the stability of the flexible optoelectronic display material is evaluated to obtain the stability test evaluation results.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: Mechanical and environmental condition tests are conducted on flexible optoelectronic display materials using a multi-field coupling test platform, performance data is collected in real time, and the stability of the materials is evaluated using a degradation confidence prediction network. By automatically adjusting the test path and conditions, the test process is optimized until the preset confidence requirements are met, ultimately obtaining accurate stability evaluation results. The most comprehensive degradation data is obtained with the fewest test cycles, significantly improving the efficiency and accuracy of stability testing and evaluation.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the stability testing method for the flexible optoelectronic display material of this application.

[0020] Figure 2 This is a schematic diagram of the stability testing system for the flexible optoelectronic display material of this application.

[0021] Figure labeling: Data acquisition module 11, degradation confidence prediction module 12, uncertainty quantification module 13, operating condition change path generation module 14, stability assessment module 15. Detailed Implementation

[0022] This application provides a stability testing method and system for flexible optoelectronic display materials, solving the technical problem in existing technologies where fixed test conditions, disconnected from real-world scenarios, lead to significant blind testing and resource waste, thus affecting stability testing efficiency. The method utilizes a multi-field coupling test platform to conduct mechanical and environmental condition tests on flexible optoelectronic display materials, acquiring performance data in real time, and employing a degradation confidence prediction network to assess material stability. By automatically adjusting test paths and conditions and optimizing the testing process until pre-set confidence requirements are met, accurate stability assessment results are obtained. This approach acquires the most comprehensive degradation data with the fewest test cycles, significantly improving stability testing efficiency and assessment accuracy.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a stability testing method for flexible optoelectronic display materials. The method is applied to a stability testing system for flexible optoelectronic display materials and specifically includes the following steps: A sample of flexible optoelectronic display material is mounted on a multi-field coupling test platform. The multi-field coupling test platform applies mechanical and environmental conditions according to the preset test scenario requirements. During the application process, the luminous intensity, resistance value and film adhesion signal of the sample are collected in real time to obtain the real-time application data sequence and the real-time measured performance degradation time series curve.

[0025] Specifically, the flexible optoelectronic display material sample to be tested is loaded onto a dedicated fixture on the testing platform. The fixture design must ensure that the sample's test area is securely fixed without slippage, while also not affecting its free deformation. After installation, the necessary measuring leads and optical probes are connected, and all sensors are ensured to be in normal operating positions. Subsequently, the platform's self-test and calibration procedures are run to ensure the zero point of the mechanical motion axis, the temperature and humidity control accuracy of the environmental chamber, and the accuracy of the reading references of all measuring instruments.

[0026] After installation, operators set preset test scenarios in the test platform's control software, simultaneously applying mechanical and environmental conditions. Mechanical conditions are specific parameters applied by the platform's mechanical components that cause the sample to deform or bear mechanical forces, including bending and tensile conditions. Bending conditions typically involve bending radius, bending angle, bending speed, and cycle frequency; tensile conditions involve tensile strain rate, maximum strain amplitude, and number of cycles. Environmental conditions are non-mechanical external conditions surrounding the sample created by the platform's environmental simulation unit, including temperature, humidity, light, and gas atmosphere. Temperature conditions refer to the ambient air temperature; humidity conditions refer to the relative humidity of the ambient air, simulating the material's use environment under different climatic conditions.

[0027] The multi-field coupling test platform begins operation according to program instructions. The environmental chamber starts heating and humidifying at a preset rate, while the robotic arm begins performing bending cycles. In each control cycle, the control core synchronously acquires the current actual bending radius, angle, and chamber temperature and humidity values, and packages and records them as a real-time operating condition application data point. Simultaneously, the measurement system is triggered: the luminance meter captures the brightness value of the illuminated area of ​​the sample; the high-resistivity meter measures the resistance between electrodes; and the acoustic emission sensor, used to indirectly evaluate adhesion, records the intensity of the stress wave signal generated when the bending action is triggered, and packages and records it as a real-time performance data point. The operating condition data point stream, sorted by time, is generated into a real-time operating condition application data sequence, and the performance data point stream is generated into a real-time measured performance degradation time series curve, indexed by timestamps, and stored in the database.

[0028] For example, taking the everyday use scenario of foldable mobile phone screens as an example, the mechanical operating parameters are set as follows: the bending radius decreases linearly from 5mm to 1mm, decreasing by 0.5mm every 10 cycles; the stretching percentage is fixed at 0.5% to simulate the slight stretching during screen bonding; the environmental operating parameters are set as follows: the temperature increases from room temperature (25℃) to 65℃ at a rate of 2℃ every 10 minutes, and then remains constant; the relative humidity increases from 40% to 85% at a rate of 5% every 10 minutes, and then remains constant. Each test cycle is set to 60 seconds. Within each cycle, a bending action is performed first, with a bending angle of 180° and a rate of 30° / s, followed by a stretching action at a rate of 0.1% / s, while maintaining the set temperature and humidity. The control platform synchronously starts each actuator according to the above preset sequence. The servo motor drives the bending arm to bend the sample around the given radius, the linear motor drives the stretching fixture to apply axial stretching, and the heater and humidifier regulate the temperature and humidity within the sealed cavity. During the application process, the real-time acquisition module simultaneously records three signals at a sampling rate of 10 times per second: the luminance value acquired by the photodetector, the resistance value measured by the four-probe method, and the film adhesion monitored by the micro-force sensor. At the end of each test cycle, the average bending radius, average stretching percentage, average temperature, and average humidity within that cycle are stored as a set of working condition data points in chronological order in the working condition application data sequence; simultaneously, the luminance, resistance value, and film adhesion at the end of that cycle are stored as a set of performance data points in chronological order in the measured performance degradation time series curve. The test stops when all 100 preset test cycles are completed or the material performance degrades to 50% of its initial value, thus obtaining a complete real-time working condition application data sequence and a real-time measured performance degradation time series curve.

[0029] The real-time operating condition applied data sequence and the measured performance degradation time series curve are input into the degradation confidence prediction network to obtain the degradation confidence prediction result, wherein the degradation confidence prediction network is trained in a self-supervised manner.

[0030] Furthermore, this application also includes the following steps: the degradation confidence prediction results include the luminance prediction confidence interval, the resistance value prediction confidence interval, and the film adhesion prediction confidence interval for multiple future test cycles.

[0031] Specifically, all historical data from the start of this test to the current moment is retrieved from the test platform's data buffer, including real-time load application data sequences and measured performance degradation time-series curves. These two data sets are preprocessed to match the network's input requirements, ensuring that the load data and performance data correspond precisely in timestamps. If the sampling frequencies differ, they are unified to the same standard time grid using interpolation or other methods. The original load parameters and performance parameters are converted into the feature format used during network training, such as through normalization or the calculation of derived features. The processed data is then organized into a fixed-length time-series sequence. If the current test duration has reached or exceeded the input length defined during network training, the data from the most recent M cycles is used as input; if there are fewer than M cycles, the sequence is padded with the initial state values.

[0032] The preprocessed, uniformly formatted time-series data sequence is input into a pre-trained degradation confidence prediction network. The network's internal parameters were fixed during the self-supervised training phase. The network performs forward computation on the input sequence. Its encoder extracts the implicit material degradation state and stress pattern features from the input sequence, and its decoder or output layer performs probabilistic inference based on these features. The direct output of the forward computation is K sets of data for the three parameters—luminance, resistance, and film adhesion—for the next K test periods. Each set of data includes the lower and upper quantiles of the predicted probability distribution of that parameter in the corresponding future period, collectively forming a confidence interval. These output confidence intervals are the degradation confidence prediction results, i.e., the predicted confidence intervals for luminance, resistance, and film adhesion for multiple future test periods. Operators can temporarily store these results in memory for subsequent calculation of the overall confidence score. It is important to note that the confidence interval coverage probability of the degraded confidence prediction network output is implicitly learned during training through a self-supervised loss function. In actual use, the interval value of the degraded confidence prediction network output can be directly taken without additional calibration.

[0033] By inputting real-time collected operating conditions and performance sequences into a trained degradation confidence prediction network, a quantitative estimate of the performance degradation range for multiple future test cycles can be obtained immediately at the current test moment. The width of the confidence interval directly reflects the degree of certainty of the degradation confidence prediction network regarding the prediction result: the narrower the interval, the more certain the degradation confidence prediction network is about the future changes in performance; the wider the interval, the higher the uncertainty, requiring further experimental testing to reduce the uncertainty.

[0034] Furthermore, this application also includes the following steps: acquiring multiple historical data points of samples that have completed full testing, wherein each historical data point contains multiple operating condition application data sequences and multiple performance degradation time series curves; dividing each historical data point into an early stage and a late stage according to the test cycle, wherein the late stage contains a preset number of consecutive test cycles, using only the multiple operating condition application data sequences and multiple performance degradation time series curves of the early stage as network input, and outputting the luminescence brightness prediction confidence interval, resistance value prediction confidence interval, and film adhesion prediction confidence interval for each test cycle in the late stage; comparing the luminescence brightness prediction confidence interval, resistance value prediction confidence interval, and film adhesion prediction confidence interval of each test cycle output by the network with the actual data of the late stage, calculating the ratio of the prediction interval coverage to the average interval width, using the negative of this ratio as a self-supervised loss function, training the network by minimizing this loss function until convergence, and obtaining the trained degradation confidence prediction network.

[0035] Specifically, all data from multiple samples that have completed full testing are extracted from the historical test database. Each sample's data includes a complete record of the entire testing cycle. Each record contains four operating parameters for each cycle: bending radius, stretching percentage, temperature, and humidity, as well as three performance indicators: luminance, resistivity, and film adhesion. Data alignment and cleaning are performed on the multiple historical data from the samples that have completed full testing to ensure strict alignment between the operating condition sequence and the timestamps of each performance curve, and to handle any missing or outlier values.

[0036] Each historical data point is divided into an early stage and a late stage based on the test period. For each historical data point of length N stages, starting from the first stage, M consecutive stages are taken as the early stage, and the remaining K stages are taken as the late stage, where M, K, and N are all positive integers, and M+K=N. Multiple operating condition application data sequences and multiple performance degradation time series curves from the early stage are used as inputs to a degradation confidence prediction network. The degradation confidence prediction network outputs the predicted confidence intervals for luminance, resistance, and film adhesion for each test stage in the late stage. A degradation confidence prediction network to be trained is constructed using a temporal neural network structure, such as a gated recurrent unit network. Its input layer receives the operating condition application data sequences and measured performance degradation time series curves from all stages in the early stage. Specifically, the input format is as follows: for each of the M stages in the early stage, a 7-dimensional vector is input, containing four operating condition parameters and three performance indicators. For each cycle in the later stage (a total of K cycles), the network output layer outputs the predicted confidence intervals for three performance indicators: luminance, resistance, and film adhesion. Each confidence interval is represented by a lower limit and an upper limit value, so the total output dimension of the network is K multiplied by 3 multiplied by 2 = 6K values.

[0037] The network outputs a predicted confidence interval for K periods, which is then compared with the corresponding actual performance data. For each performance metric, in each period, it is determined whether the actual value falls within the predicted interval. The total number of times the actual value falls within the interval across all K periods and the three metrics is counted, and this number is divided by the total number of comparisons to obtain the predicted interval coverage. Simultaneously, for each performance metric, the predicted interval width is calculated in each period. This is done by subtracting the lower bound from the upper bound and then averaging the results over all K periods and the three metrics to obtain the average interval width. The ratio of coverage to average interval width is calculated. Since the desired coverage is as high as possible and the width as small as possible, a larger ratio indicates better prediction quality. The negative of this ratio is taken as the self-supervised loss function, i.e., loss function = -(coverage / average interval width). Note that when the average interval width is very small, the ratio may be very large; to prevent numerical instability, a very small positive number is added to the denominator. The goal of network training is to minimize the loss, i.e., to push the loss towards a more negative direction, which is equivalent to maximizing the coverage / average interval width. Using all N historical data points, calculate the loss value for each data point as described above, and average the loss values ​​for the entire batch. Update the network parameters using the backpropagation algorithm, repeating this process multiple times until the loss function value no longer significantly decreases in several consecutive iterations or reaches the preset maximum number of iterations. At this point, the network converges, and the trained degraded confidence prediction network is obtained. In other words, the network is trained multiple times using samples generated from a large amount of historical data. After each certain number of training iterations, the performance of the current network is evaluated on a separate validation set of historical data, calculating its coverage and average width. When the loss on the validation set no longer decreases in several consecutive iterations, or even begins to increase, training is stopped, and a snapshot of the network parameters at this point is saved; this is the trained degraded confidence prediction network. Throughout the entire training process, the network never directly sees the performance data of the later stages as input, but relies solely on information from the earlier stages to predict the confidence interval of the later stages, learning by comparing it with actual later stage data. Therefore, this is a self-supervised training method.

[0038] By training a degradation confidence prediction network using a self-supervised approach, the network learns to predict future confidence intervals based on past performance degradation trends without relying on manual annotation, utilizing historical test data. The trained network can then output performance prediction confidence intervals for multiple future periods in subsequent real-time tests, based solely on currently measured operating conditions and performance data, thus quantifying the uncertainty of the prediction.

[0039] The degradation confidence prediction results are weighted to obtain a comprehensive confidence score, and the comprehensive confidence score is bound to the real-time measured performance degradation time series curve and the real-time operating condition application data sequence and added to the operating condition space.

[0040] Furthermore, this application also includes the following steps: determining a first fusion weight corresponding to the luminance, a second fusion weight corresponding to the resistance value, and a third fusion weight corresponding to the film adhesion based on the layer structure position of the flexible optoelectronic display material; and weighting the degradation confidence prediction result based on the first fusion weight, the second fusion weight, and the third fusion weight to obtain a comprehensive confidence score.

[0041] Furthermore, this application also includes the following steps: obtaining the half-width of the luminance prediction confidence interval, the half-width of the resistance prediction confidence interval, and the half-width of the film adhesion prediction confidence interval for each future test cycle; multiplying the three half-widths of each test cycle by the corresponding first fusion weight, second fusion weight, and third fusion weight respectively, and then summing them to obtain the weighted half-width of that cycle; calculating the average of the multiple weighted half-widths for multiple future test cycles, and comparing the average with the preset maximum allowable half-width to obtain a comprehensive confidence score.

[0042] Furthermore, this application also includes the following steps: the first fusion weight is the same as the second fusion weight, and the first fusion weight and the second fusion weight are respectively greater than the third fusion weight.

[0043] Specifically, based on the specific layer structure of the flexible optoelectronic display material under test, the values ​​of the three fusion weights are determined. Since the first fusion weight corresponding to luminance is the same as the second fusion weight corresponding to resistance, and both are greater than the third fusion weight corresponding to film adhesion, the operator can set the first fusion weight to 0.4, the second fusion weight to 0.4, and the third fusion weight to 0.2, with a sum of 1. The specific values ​​of the weights can be fine-tuned according to the actual sensitivity of the material's layer structure, but the first two must remain equal and greater than the third. The first fusion weight is the weighting coefficient assigned to the luminance index, denoted by w1; the second fusion weight is the weighting coefficient assigned to the resistance index, denoted by w2; and the third fusion weight is the weighting coefficient assigned to the film adhesion index, denoted by w3.

[0044] Obtain the prediction results for multiple future test periods from the degradation confidence prediction network. For each period, extract the half-widths of the luminance prediction confidence interval, the resistance prediction confidence interval, and the film adhesion prediction confidence interval. Calculate the weighted sum of the three half-widths for each period: multiply the luminance half-width by the first fusion weight, add the resistance half-width multiplied by the second fusion weight, and add the adhesion half-width multiplied by the third fusion weight to obtain the weighted half-width for that period. For a prediction confidence interval, its half-width is equal to (upper limit of the interval minus the lower limit of the interval) divided by 2.

[0045] Calculate the arithmetic mean of all weighted half-widths over the next K periods. Sum the weighted half-widths for each period and divide by K to obtain the average weighted half-width. Compare the calculated average weighted half-width with the preset maximum allowable half-width to obtain a comprehensive confidence score. The comprehensive confidence score is set to 1 when the average weighted half-width is 0; and 0 when the average weighted half-width is equal to or exceeds the preset maximum allowable half-width. Intermediate values ​​can be calculated using linear interpolation or other monotonically decreasing functions, such as comprehensive confidence score = 1 - (average weighted half-width / preset maximum allowable half-width), with the result limited to between 0 and 1. When the average weighted half-width exceeds the maximum allowable half-width, the score may be zero or negative, indicating extremely low cognitive certainty. The comprehensive confidence score reflects the overall reliability of the network's predictions of future performance at the current operating point. A higher score indicates that the operating point is well understood, while a lower score indicates significant uncertainty at that operating point, requiring further experimental testing. For example, the prediction confidence intervals for cycles 46 to 50 are as follows: for cycle 46, the brightness interval is [850, 880], the resistance interval is [152, 160], and the adhesion interval is [2.10, 2.18]; for cycle 47, the brightness interval is [820, 855], the resistance interval is [160, 170], and the adhesion interval is [2.00, 2.18]; for cycle 48, the brightness interval is [785, 825], the resistance interval is [170, 182], and the adhesion interval is [1088, 2.00]; for cycle 49, the brightness interval is [745, 792], the resistance interval is [182, 196], and the adhesion interval is [1.75, 1.89]; and for cycle 50, the brightness interval is [702, 756], the resistance interval is [196, 214], and the adhesion interval is [1.60, 1.76]. Calculate the half-width for each cycle: Cycle 46: Brightness half-width = 15, Resistance half-width = 4, Adhesion half-width = 0.04; Cycle 47: Brightness half-width = 17.5, Resistance half-width = 5, Adhesion half-width = 0.05; Cycle 48: Brightness half-width = 20, Resistance half-width = 6, Adhesion half-width = 0.06; Cycle 49: Brightness half-width = 23.5, Resistance half-width = 7, Adhesion half-width = 0.07; Cycle 50: Brightness half-width = 27, Resistance half-width = 9, Adhesion half-width = 0.08. Calculate the weighted half-width for each cycle, with a brightness weight of 0.4, a resistance weight of 0.4, and an adhesion weight of 0.2. The weighted half-widths are 7.608 for cycle 46, 9.01 for cycle 47, 10.412 for cycle 48, 12.214 for cycle 49, and 14.416 for cycle 50. The average of the weighted half-widths is 10.732. Based on the testing accuracy requirements, the maximum allowable half-width is set to 20.0.Using linear normalization, the overall confidence score = 1 - (average weighted half-width / preset maximum allowable half-width) = 0.4634. This score is lower than 0.5, indicating that the prediction uncertainty at the current operating point is high and further field measurements are needed.

[0046] The overall confidence score is bound to the real-time measured performance degradation time-series curve and the real-time operating condition application data sequence, and stored in the operating condition space. The identification information of the current operating condition point is determined, i.e., the actual applied operating condition parameter values ​​at the end of the current test cycle, including bending radius, stretching percentage, temperature, and humidity. Since the operating condition parameters may change over time during the test, the operating condition parameter values ​​at the end of the current cycle represent the coordinates of a specific operating condition point. The real-time measured performance degradation time-series curve up to the current cycle is extracted, including all measured values ​​of luminance, resistance, and film adhesion from cycle 1 to the current cycle. Simultaneously, the real-time operating condition application data sequence is extracted, including all applied values ​​of bending radius, stretching percentage, temperature, and humidity from cycle 1 to the current cycle. The overall confidence score, the real-time measured performance degradation time-series curve, and the real-time operating condition application data sequence are bound together, and a record item is created for the current operating condition point in the data structure of the operating condition space, containing the operating condition point coordinates, the overall confidence score, the performance degradation time-series curve, and the operating condition application data sequence. If the operating point has already been recorded in the operating condition space, it is necessary to determine whether to update it. Typically, since the testing is conducted incrementally, the combination of operating parameters for each cycle may be unique. However, if duplicates occur, the latest calculated composite confidence score can be used to overwrite the old value.

[0047] By assigning differentiated fusion weights to different performance indicators based on the location of the material layer structure, and calculating the average weighted half-width over multiple future periods, the multi-dimensional prediction uncertainty is compressed into a scalarized comprehensive confidence score, reflecting the overall reliability of the network's prediction of future performance degradation under the current operating conditions. In the weight settings, luminance and resistance are equally important and higher than film adhesion, consistent with the actual physical law that photoelectric performance is the core failure mode in flexible display devices. A lower score indicates higher uncertainty at that operating point, which should be prioritized for active exploration; a higher score indicates that the operating point has been fully understood and does not require repeated testing.

[0048] When the spatial confidence of the working condition space does not meet the preset requirements, the working condition application data sequence corresponding to the minimum comprehensive confidence score in the working condition space is extracted as the target point, and multiple candidate continuous working condition change paths pointing to the target point are generated, starting from the real-time working condition application data sequence.

[0049] Furthermore, this application also includes the following steps: starting from the real-time operating condition application data sequence, a path search algorithm is used to generate multiple candidate continuous operating condition change paths with the endpoint as the target point; wherein, each candidate continuous operating condition change path contains a preset number of test cycles, and the bending radius change in each test cycle is sampled from the first change step set, the stretching percentage change is sampled from the second change step set, the temperature change is sampled from the third change step set, and the humidity change is sampled from the fourth change step set.

[0050] Specifically, statistical analysis is performed on all currently stored comprehensive confidence scores in the operating condition space. Each record in the operating condition space is traversed, and the comprehensive confidence score corresponding to each operating condition point is read. The arithmetic mean of these scores is calculated. The number of points with scores below a preset low score threshold is counted, and this number is divided by the total number of points in the operating condition space to obtain the percentage of low-scoring points.

[0051] The spatial confidence level is determined based on two criteria: an average overall confidence score greater than a first preset threshold and a percentage of low-scoring points less than a second preset threshold. If both conditions are met, the spatial confidence level of the working space is deemed to meet the preset requirements, and the active exploration loop can be terminated, proceeding to the final stability assessment step. If either condition is not met—such as an excessively low average score, too many low-scoring points, or neither—the spatial confidence level is deemed insufficient, and active exploration must continue.

[0052] When a record is determined to be non-compliant with preset requirements, the record with the lowest overall confidence score is identified from the operating condition space, and the operating condition application data sequence stored in that record is extracted. Since each operating condition point corresponds to a set of operating condition parameters, the last cycle in the sequence is typically used as the coordinates of the target point. The target point consists of a set of target operating condition parameters, including the target bending radius, target stretching percentage, target temperature, and target humidity.

[0053] Multiple candidate continuous operating condition change paths are generated, pointing from the current real-time operating point to the target point. The starting point is the actual applied operating condition parameter value at the end of the current test cycle. The operator sets each candidate path to contain a preset number of test cycles, denoted as L. Each test cycle in the path corresponds to a set of operating condition parameters, starting from the starting point and gradually changing through L cycles to finally reach the target point. To generate multiple different candidate paths, a path search algorithm is used to explore feasible trajectories from the starting point to the target point within the operating condition space. During the search process, constraints must be imposed: the operating condition parameters of all intermediate points and the operating condition parameters of the endpoint must be within the safe operating boundaries of the multi-field coupling test platform, such as a bending radius of not less than 1.0 mm and not more than 10.0 mm, a stretching percentage between 0% and 2%, a temperature between 10℃ and 85℃, and a humidity between 20%RH and 95%RH.

[0054] For each candidate path, the changes in each operating parameter within each test cycle relative to the previous cycle are determined by sampling from a predefined set of change step sizes. The change in bending radius is sampled from the first set of change step sizes, the change in stretching percentage from the second set, the change in temperature from the third set, and the change in humidity from the fourth set. Each set of step sizes typically contains multiple discrete values; for example, the first set could be -0.5mm, -0.3mm, -0.1mm, 0mm, +0.1mm, +0.3mm, and +0.5mm. During sampling, all possible combinations can be generated using a certain enumeration strategy, resulting in multiple different paths. The path search algorithm ensures that after accumulating all sampled changes, the endpoint falls precisely near the target point, generating a set of candidate continuous operating condition change paths. Each path consists of a sequence of operating parameters from L test cycles. The first set of step sizes is a predefined set of allowed values ​​for the change in bending radius, containing several allowable values, typically including negative, zero, and positive values. The size of the step size reflects the minimum controllable accuracy of the test platform actuator and the material's tolerance to changes in bending rate. The second set of step sizes is a predefined set of allowed values ​​for the change in tensile percentage. The third set of step sizes is a predefined set of allowed values ​​for the change in temperature. The fourth set of step sizes is a predefined set of allowed values ​​for the change in humidity.

[0055] For each cycle, a change value is randomly selected from the set of change step sizes, but it must be ensured that the deviation between the cumulative change value and the target total change value after L cycles is within the allowable range. During the sampling process, the system must always check whether the operating parameters of each intermediate cycle are within the safe operating boundaries of the multi-field coupling test platform. If the change value obtained by sampling in a certain cycle causes the bending radius of that cycle to be less than the minimum value allowed by the platform or greater than the maximum value, then that change value cannot be used, and resampling or abandoning that path is required. Similarly, the tensile percentage, temperature, and humidity must also be within their respective safety boundaries.

[0056] By calculating the average comprehensive confidence score and the proportion of low-scoring points in the working space, the overall average level and the proportion of local low-quality areas are combined, avoiding the problem that relying solely on the average might mask local defects. When the spatial confidence score does not meet the requirements, the lowest-scoring point is automatically extracted as the target point for active exploration, ensuring that test resources are prioritized for allocation to the most uncertain areas. A path search algorithm is adopted, combined with safe operating boundaries and discrete variable step size sets, to generate multiple candidate paths, ensuring both path feasibility and path diversity.

[0057] The path confidence evaluation network selects the path with the highest information gain from multiple candidate continuous operating condition change paths to drive the test platform to execute automatically. This process is repeated until the spatial confidence of the operating condition space meets the preset requirements. Based on multiple measured performance degradation time series curves stored in the operating condition space, the stability of the flexible optoelectronic display material is evaluated to obtain the stability test evaluation results.

[0058] Furthermore, this application also includes the following steps: constructing a gated recurrent unit network; extracting multiple executed test paths from historical test data, and calculating the average change in the comprehensive confidence of the stored work condition application data sequence in the work condition space before and after the execution of each test path, using the average change value as the label of the path to obtain a test dataset; using the test dataset, training the gated recurrent unit network with regression loss until convergence to obtain the path confidence evaluation network.

[0059] Furthermore, this application also includes the following steps: the input of the gated loop unit network is a sequence of bending radius, stretching percentage, temperature and humidity of a candidate continuous operating condition change path, and the output is the average change in comprehensive confidence expected to be brought about after executing the path.

[0060] Specifically, a gated recurrent unit network model is constructed. The input layer is designed to receive four parallel time series: a bending radius sequence, a stretching percentage sequence, a temperature sequence, and a humidity sequence. The length of each sequence is equal to the preset number of test periods L. The network internally consists of one or more gated recurrent unit layers for extracting temporal features, which are then connected to fully connected layers to map to a scalar output. The activation function of the output layer can be linear, as the output value can be positive or negative without range limitations.

[0061] After the gated recurrent unit network model is built, the operator needs to prepare training data. Multiple complete test paths that have already been executed are extracted from the historical test database. Each historical path contains a sequence of operating parameters for L test cycles from the start to the end, as well as records of the comprehensive confidence scores of all stored points in the operating space before and after executing the path. For each historical path, the operator calculates the average comprehensive confidence score before executing the path (i.e., the average score of all points in the operating space before the path begins) and the average comprehensive confidence score after executing the path (i.e., the average score of all points after the path is completed and the operating space is updated). The latter is then subtracted from the former to obtain the average change in comprehensive confidence for that path. This change value is the label for that path. The four operating parameter sequences for each path are used as input features, and the corresponding average change value is used as the output label to form a training sample. The set of all samples constitutes the test dataset. The gated recurrent unit network is trained using this test dataset. During training, the bend radius sequence, stretching percentage sequence, temperature sequence, and humidity sequence for each path are used as network inputs. The network outputs a predicted value, which is the average change in the overall confidence score. The predicted value is compared with the true label value, and a regression loss, such as mean squared error loss, is calculated. The network parameters are updated using the backpropagation algorithm, and the training is repeated for multiple epochs until the loss function value no longer decreases significantly or the preset maximum number of iterations is reached. At this point, the network converges. The converged network is the trained path confidence evaluation network.

[0062] The trained network can be used in real-time testing. After generating multiple candidate paths for continuous operating condition changes, the four parameter sequences of each path are input into the network. The network outputs the expected average change in overall confidence for each path, which is the predicted value of information gain. The path with the largest output value, i.e., the path expected to bring the greatest confidence improvement, is selected to drive the testing platform to execute automatically.

[0063] After generating multiple candidate continuous operating condition change paths and completing the path confidence evaluation network to predict the information gain of each path, the path with the largest output value is selected from all candidate paths, i.e., the path that is expected to bring the highest average change in overall confidence. The operator extracts the operating condition parameter sequence from the selected path and generates control commands sequentially according to the order of each test cycle, sending them to the multi-field coupled test platform.

[0064] Upon receiving the command, the multi-field coupling test platform first adjusts the bending arm and tension fixture to the specified bending radius and tension percentage within each test cycle. Simultaneously, it controls the heater and humidifier to maintain the set temperature and humidity within the chamber. This set of operating conditions is then maintained for a complete stress application cycle, such as 60 seconds, while simultaneously acquiring the sample's luminescence intensity, resistivity, and film adhesion signals. After one cycle, the platform automatically proceeds to the next cycle's operating parameter settings until all L test cycles for that path have been completed.

[0065] After the path execution is complete, new real-time operating condition application data sequences and real-time measured performance degradation time-series curves are obtained. These are used to update the input of the degradation confidence prediction network, recalculate the prediction confidence intervals and overall confidence scores for multiple future test cycles, and bind the new operating condition points, their confidence scores, and performance curves to the operating condition space. The average overall confidence score and the proportion of low-scoring points among all stored points in the operating condition space are recalculated to determine whether the spatial confidence meets the preset requirements. If the requirements are met, the active exploration loop terminates; if the requirements are not met, the operating condition point corresponding to the minimum new overall confidence score is extracted as the target point. A new candidate path is generated starting from the current real-time operating condition point. The path confidence evaluation network selects the path with the highest information gain and executes it, and this process is repeated iteratively.

[0066] Once the spatial confidence level meets the preset requirements, the active exploration phase ends, and the stability assessment phase begins. Operators retrieve all stored operating points and their corresponding measured performance degradation time-series curves from the operating space, covering the complete performance change process from the start to the end of the test, and including degradation data under different combinations of operating conditions. For example, for each performance indicator—brightness, resistance, and adhesion—an envelope curve of performance changes over time is plotted for all operating points, determining the minimum number of cycles required for performance degradation to reach the failure threshold, such as 80% of initial brightness or a 50% increase in resistance, as a conservative lifetime estimate for the material. Stability test evaluation results are output; for example, the flexible OLED sample, within a bending radius of 1.0 mm to 5.0 mm, a temperature range of 25°C to 45°C, and a humidity range of 40%RH to 80%RH, experiences at least 500 cycles of brightness degradation to 80% of the initial value, meeting the requirements for foldable screen applications.

[0067] By automatically selecting the path with the highest information gain through path confidence evaluation and driving platform execution, this step achieves fully automated and intelligent closed-loop control of the testing process. Each cycle performs targeted testing on the most uncertain area, shortening the number of test cycles required for full exploration. Once the confidence level of the operating space meets the preset requirements, a stability assessment is performed based on all measured performance degradation time-series curves stored in the space. This ensures that the assessment results are based on the most comprehensive and informative data, improving the accuracy and reliability of the assessment.

[0068] In summary, the stability testing method for flexible optoelectronic display materials provided in this application has the following technical advantages: Mechanical and environmental condition tests are conducted on the flexible optoelectronic display materials using a multi-field coupling test platform, performance data is collected in real time, and the material stability is evaluated using a degradation confidence prediction network. By automatically adjusting the test path and conditions, the test process is optimized until the preset confidence requirements are met, ultimately obtaining accurate stability evaluation results. This method acquires the most comprehensive degradation data with the fewest test cycles, significantly improving the efficiency and accuracy of stability testing and evaluation.

[0069] Example 2: Based on the same inventive concept as the stability testing method for flexible optoelectronic display materials in Example 1, this application also provides a stability testing system for flexible optoelectronic display materials. Please refer to the appendix. Figure 2The stability testing system for the flexible optoelectronic display material includes: a data acquisition module 11, used to mount the sample of the flexible optoelectronic display material onto a multi-field coupling test platform. The multi-field coupling test platform applies mechanical and environmental conditions according to preset test scenario requirements. During the application process, the luminance, resistance value, and film adhesion signals of the sample are acquired in real time to obtain a real-time application data sequence and a real-time measured performance degradation time series curve; a degradation confidence prediction module 12, used to input the real-time application data sequence and the measured performance degradation time series curve into a degradation confidence prediction network to obtain a degradation confidence prediction result, wherein the degradation confidence prediction network is trained in a self-supervised manner; and an uncertainty quantification module 13, used to weight the degradation confidence prediction result to obtain a comprehensive confidence score, and to convert the comprehensive confidence score into a weighted score. The scores are bound to the real-time measured performance degradation time-series curves and the real-time operating condition application data sequences, and added to the operating condition space. The operating condition change path generation module 14 is used to extract the operating condition application data sequence corresponding to the minimum comprehensive confidence score in the operating condition space as the target point when the spatial confidence of the operating condition space does not meet the preset requirements, and generate multiple candidate continuous operating condition change paths pointing to the target point with the real-time operating condition application data sequence as the starting point. The stability evaluation module 15 is used to select the path with the highest information gain from multiple candidate continuous operating condition change paths through the path confidence evaluation network to drive the test platform to execute automatically, and so on, until the spatial confidence of the operating condition space meets the preset requirements. Based on the multiple measured performance degradation time-series curves stored in the operating condition space, the stability of the flexible optoelectronic display material is evaluated to obtain the stability test evaluation results.

[0070] Furthermore, the degradation confidence prediction module 12 in the stability testing system for the flexible optoelectronic display material is also used to: include the degradation confidence prediction results as the luminance prediction confidence interval, the resistance value prediction confidence interval, and the film adhesion prediction confidence interval for multiple future test cycles.

[0071] Furthermore, the degradation confidence prediction module 12 in the stability testing system for the flexible optoelectronic display material is also used to: acquire multiple historical data of samples that have completed a full test, wherein each historical data contains multiple operating condition application data sequences and multiple performance degradation time series curves; divide each historical data into an early stage and a late stage according to the test cycle, wherein the late stage contains a preset number of consecutive test cycles, and only uses the multiple operating condition application data sequences and multiple performance degradation time series curves of the early stage as network input, and outputs the luminance prediction confidence interval, resistance prediction confidence interval, and film adhesion prediction confidence interval for each test cycle in the late stage; compare the luminance prediction confidence interval, resistance prediction confidence interval, and film adhesion prediction confidence interval of each test cycle output by the network with the actual data of the late stage, calculate the ratio of the prediction interval coverage to the average interval width, use the negative of the ratio as a self-supervised loss function, train the network by minimizing the loss function until convergence, and obtain the trained degradation confidence prediction network.

[0072] Furthermore, the uncertainty quantification module 13 in the stability testing system for the flexible optoelectronic display material is also used to: determine the first fusion weight corresponding to the luminous brightness, the second fusion weight corresponding to the resistance value, and the third fusion weight corresponding to the film adhesion based on the layer structure position of the flexible optoelectronic display material; and to weight the degradation confidence prediction result based on the first fusion weight, the second fusion weight, and the third fusion weight to obtain a comprehensive confidence score.

[0073] Furthermore, the uncertainty quantification module 13 in the stability testing system for the flexible optoelectronic display material is also used to: obtain the half-width of the luminance prediction confidence interval, the half-width of the resistance prediction confidence interval, and the half-width of the film adhesion prediction confidence interval for each future test cycle; multiply the three half-widths of each test cycle by the corresponding first fusion weight, second fusion weight, and third fusion weight respectively, and then sum them to obtain the weighted half-width of that cycle; calculate the average value of multiple weighted half-widths for multiple future test cycles, compare the average value with the preset maximum allowable half-width, and obtain a comprehensive confidence score.

[0074] Furthermore, the uncertainty quantification module 13 in the stability testing system for the flexible optoelectronic display material is also used to: make the first fusion weight the same as the second fusion weight, and make the first fusion weight and the second fusion weight greater than the third fusion weight respectively.

[0075] Furthermore, the working condition change path generation module 14 in the stability testing system for the flexible optoelectronic display material is also used to: generate multiple candidate continuous working condition change paths with the endpoint as the target point, starting from the real-time working condition application data sequence; wherein, each candidate continuous working condition change path contains a preset number of test cycles, and the bending radius change in each test cycle is sampled from the first change step set, the stretching percentage change is sampled from the second change step set, the temperature change is sampled from the third change step set, and the humidity change is sampled from the fourth change step set.

[0076] Furthermore, the stability evaluation module 15 in the stability testing system for the flexible optoelectronic display material is also used for: constructing a gated recurrent unit network; extracting multiple executed test paths from historical test data, and calculating the average change in the comprehensive confidence of the data sequence of applied conditions stored in the working condition space before and after the execution of each test path, using the average change value as the label of the path to obtain a test dataset; using the test dataset, training the gated recurrent unit network with regression loss until convergence to obtain the path confidence evaluation network.

[0077] Furthermore, the stability evaluation module 15 in the stability testing system for the flexible optoelectronic display material is also used for: the gating loop unit network input is a bending radius sequence, stretching percentage sequence, temperature sequence, and humidity sequence of a candidate continuous operating condition change path, and the output is the average change value of the comprehensive confidence level expected to be brought about after executing the path.

[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The stability testing method and specific examples of the flexible optoelectronic display material in the aforementioned embodiment one are also applicable to the stability testing system of the flexible optoelectronic display material in this embodiment. Through the foregoing detailed description of the stability testing method of the flexible optoelectronic display material, those skilled in the art can clearly understand the stability testing system of the flexible optoelectronic display material in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0080] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for testing the stability of flexible optoelectronic display materials, characterized in that, include: A sample of flexible optoelectronic display material is mounted on a multi-field coupling test platform. The multi-field coupling test platform applies mechanical and environmental conditions according to the preset test scenario requirements. During the application process, the luminous brightness, resistance value and film adhesion signal of the sample are collected in real time to obtain the real-time application data sequence and the real-time measured performance degradation time series curve. The real-time operating condition applied data sequence and the measured performance degradation time series curve are input into the degradation confidence prediction network to obtain the degradation confidence prediction result, wherein the degradation confidence prediction network is trained in a self-supervised manner. The degradation confidence prediction results are weighted to obtain a comprehensive confidence score, and the comprehensive confidence score is bound to the real-time measured performance degradation time series curve and the real-time operating condition application data sequence and added to the operating condition space; When the spatial confidence of the working condition space does not meet the preset requirements, the working condition application data sequence corresponding to the minimum comprehensive confidence score in the working condition space is extracted as the target point, and multiple candidate continuous working condition change paths pointing to the target point are generated with the real-time working condition application data sequence as the starting point. The path confidence evaluation network selects the path with the highest information gain from multiple candidate continuous operating condition change paths to drive the test platform to execute automatically. This process is repeated until the spatial confidence of the operating condition space meets the preset requirements. Based on multiple measured performance degradation time series curves stored in the operating condition space, the stability of the flexible optoelectronic display material is evaluated to obtain the stability test evaluation results.

2. The stability testing method for flexible optoelectronic display materials as described in claim 1, characterized in that, The degradation confidence prediction results include the luminance prediction confidence interval, resistance prediction confidence interval, and film adhesion prediction confidence interval for multiple future test cycles.

3. The stability testing method for flexible optoelectronic display materials as described in claim 2, characterized in that, include: The first fusion weight corresponding to the luminance, the second fusion weight corresponding to the resistance value, and the third fusion weight corresponding to the film adhesion are determined based on the layer structure position of the flexible optoelectronic display material. The degradation confidence prediction result is weighted based on the first fusion weight, the second fusion weight, and the third fusion weight to obtain a comprehensive confidence score.

4. The stability testing method for flexible optoelectronic display materials as described in claim 3, characterized in that, The degradation confidence prediction result is weighted based on the first fusion weight, the second fusion weight, and the third fusion weight to obtain a comprehensive confidence score, including: Obtain the half-width of the confidence interval for predicted luminance, the half-width of the confidence interval for predicted resistance, and the half-width of the confidence interval for predicted film adhesion for each future test cycle. The weighted half-width of each test period is obtained by multiplying the three half-widths of each test period by the corresponding first fusion weight, second fusion weight, and third fusion weight, and then summing them. Calculate the average of multiple weighted half-widths over multiple future test periods, compare the average with the preset maximum allowable half-width, and obtain the overall confidence score.

5. The stability testing method for flexible optoelectronic display materials as described in claim 4, characterized in that, The first fusion weight is the same as the second fusion weight, and both the first fusion weight and the second fusion weight are greater than the third fusion weight.

6. The stability testing method for flexible optoelectronic display materials as described in claim 1, characterized in that, The degradation confidence prediction network is trained in a self-supervised manner, including: Acquire multiple historical data points for samples that have completed full testing. Each historical data point contains multiple operating condition application data sequences and multiple performance degradation time series curves. Each piece of historical data is divided into an early stage and a late stage according to the test cycle. The late stage contains a preset number of consecutive test cycles. Only the data sequences of multiple working conditions applied in the early stage and multiple performance degradation time series curves are used as network inputs. The network outputs the luminous brightness prediction confidence interval, the resistance value prediction confidence interval, and the film adhesion prediction confidence interval for each test cycle in the late stage. The luminance prediction confidence interval, resistance prediction confidence interval, and film adhesion prediction confidence interval of each test cycle output by the network are compared with the actual data of the later stage. The ratio of the prediction interval coverage to the average interval width is calculated. The negative of this ratio is used as the self-supervised loss function. The network is trained by minimizing this loss function until convergence, and the trained degradation confidence prediction network is obtained.

7. The stability testing method for flexible optoelectronic display materials as described in claim 1, characterized in that, The load condition application data sequence corresponding to the minimum comprehensive confidence score in the load condition space is extracted as the target point. Starting from the real-time load condition application data sequence, multiple candidate continuous load condition change paths pointing to the target point are generated, including: Starting with the real-time operating condition application data sequence, a path search algorithm is used to generate multiple candidate continuous operating condition change paths with the endpoint as the target point; Each candidate continuous operating condition change path includes a preset number of test cycles. Within each test cycle, the change in bending radius is sampled from the first change step set, the change in stretching percentage is sampled from the second change step set, the change in temperature is sampled from the third change step set, and the change in humidity is sampled from the fourth change step set.

8. The stability testing method for flexible optoelectronic display materials as described in claim 1, characterized in that, Also includes: Construct a gated cyclic unit network; Extract multiple executed test paths from historical test data, and calculate the average change in the comprehensive confidence of the data sequence of applied working conditions stored in the working condition space before and after the execution of each test path. Use the average change value as the label of the path to obtain the test dataset. Using the test dataset, a gated recurrent unit network is trained with regression loss until convergence, thus obtaining the path confidence evaluation network.

9. The stability testing method for flexible optoelectronic display materials as described in claim 8, characterized in that, The input to the gated loop unit network is a sequence of bending radius, stretching percentage, temperature, and humidity of a candidate continuous operating condition change path, and the output is the average change in overall confidence expected after executing the path.

10. A stability testing system for flexible optoelectronic display materials, characterized in that, The step of implementing the stability testing method for the flexible optoelectronic display material according to any one of claims 1 to 9, wherein the stability testing system for the flexible optoelectronic display material comprises: The data acquisition module is used to mount the sample of flexible optoelectronic display material on a multi-field coupling test platform. The multi-field coupling test platform applies mechanical and environmental conditions according to the preset test scenario requirements. During the application process, the luminous brightness, resistance value and film adhesion signal of the sample are acquired in real time to obtain the real-time application data sequence and the real-time measured performance degradation time series curve. The degradation confidence prediction module is used to input the real-time operating condition applied data sequence and the measured performance degradation time series curve into the degradation confidence prediction network to obtain the degradation confidence prediction result, wherein the degradation confidence prediction network is trained in a self-supervised manner. The uncertainty quantification module is used to weight the degradation confidence prediction results to obtain a comprehensive confidence score, and bind the comprehensive confidence score with the real-time measured performance degradation time series curve and the real-time operating condition application data sequence, and add it into the operating condition space; The working condition change path generation module is used to extract the working condition application data sequence corresponding to the minimum comprehensive confidence score in the working condition space as the target point when the spatial confidence of the working condition space does not meet the preset requirements, and generate multiple candidate continuous working condition change paths pointing to the target point with the real-time working condition application data sequence as the starting point. The stability assessment module is used to select the path with the highest information gain from multiple candidate continuous operating condition change paths through a path confidence assessment network, and drive the test platform to execute automatically. This process continues until the spatial confidence of the operating condition space meets the preset requirements. Based on multiple measured performance degradation time-series curves stored in the operating condition space, the stability of the flexible optoelectronic display material is assessed, and the stability test assessment results are obtained.

Images