In-situ Monitoring and Anomaly Detection for LED Packages Using a Mahalanobis Distance Approach

Owing to the long lifetime and high reliability of light emitting diode (LED) packages, few or any failures should occur during a short-term or accelerated life test. Therefore, a time and cost-effective qualification test for accurately predicting the long-term lifetime of an LED package is a critical economic and business requirement for adoption of new LEDs. Previous research usually applied offline photometric measurements to collect the direct performance degradation data of LEDs (e.g. luminous flux and chromaticity coordinates). However, these methods incurred measurement errors and significant testing costs. In this paper, an in-situ monitoring method with sensing the indirect performance data (e.g. lead temperatures, input driven current, and forward voltage) is proposed to detect the health of LEDs. In this proposed method, a data-driven method using a Mahalanobis distance (MD) approach is employed to detect early anomalies of LEDs before failures happen and transformed MD values are defined as a real-time health indicator to reflect the LED's degradation. The experimental results show that the proposed MD-based anomaly detection approach can provide an early anomaly warning at around 45% of lifetime before actual failure happens for all test LEDs evaluated.