Transparent Organic Light Emitting Diode (OLED) displays are utilized in various fields, such as digital signage, smart windows, and automotive Head-Up Displays (HUDs), due to their dual characteristics of high transmittance and self-emission. With th...
Transparent Organic Light Emitting Diode (OLED) displays are utilized in various fields, such as digital signage, smart windows, and automotive Head-Up Displays (HUDs), due to their dual characteristics of high transmittance and self-emission. With the increasing outdoor application of these displays, it has become crucial to analyze degradation under ultraviolet (UV) conditions, which affect the films, and to ensure product reliability. However, conventional degradation analysis has primarily relied on linear or exponential models, which fail to sufficiently account for outliers and the non-linear degradation characteristics inherent in actual data.
This study analyzes the impact of outliers in transparent OLED degradation data on predictive performance and proposes a degradation analysis method that combines outlier removal with non-linear regression. Experiments were conducted using a full factorial design involving UV exposure, color, brightness, and UV-blocking films. A total of 4,992 relative luminance data points were obtained from accelerated testing over 2,166 hours under 32 distinct conditions. Model training was restricted to the 0–2,030 h interval. Symbolic Regression (SR), linear, and exponential models were constructed after applying outlier removal exclusively to the training dataset. Furthermore, the study analyzes differences in model performance based on the outlier removal method by comparing RANSAC and Z-score preprocessing techniques.
Regarding outlier removal, the Z-score method failed to eliminate rapidly fluctuating outliers due to its dependence on variance. In contrast, RANSAC effectively removed outliers by identifying data points with large residuals based on the underlying trend. Symbolic Regression generated models expressible as mathematical formulas within a complexity range of 4 to 6. Notably, the model applying RANSAC-SR demonstrated superior predictive performance, exhibiting the lowest prediction error at the final measurement point, which was excluded from the training set.
Analysis of condition-specific characteristics using the derived degradation models revealed that degradation was accelerated under UV irradiation and high-luminance driving conditions. Additionally, it was confirmed that degradation analysis using the RANSAC-SR model is feasible and yields statistically significant results.
This study presents a methodology to enhance degradation prediction accuracy by overcoming the limitations of conventional outlier processing and non-linear modeling through the RANSAC-Symbolic Regression model. This approach reconstructs degradation data into a continuous model capable of time-dependent analysis. The proposed method is expected to be applicable not only to transparent OLED displays but also to the analysis and reliability evaluation of various displays and materials exhibiting degradation characteristics.