Light-emitting diode (LED) applications have expanded from display backlighting in computers and smart phones to more demanding applications including automotive headlights and street lightening. With these new applications, LED manufacturers must ensure that their products meet the performance requirements expected by end users, which in many cases require lifetimes of 10 years or more. The qualification tests traditionally conducted to assess such lifetimes are often as long as 6,000 hours, yet even this length of time does not guarantee that the lifetime requirements will be met.

This research aims to reduce the qualification time by employing anomaly detection and prognostic methods utilizing optical, electrical, and thermal parameters of LEDs. The outcome of this research will be an in-situ monitoring approach that enables parameter sensing, data acquisition, and signal processing to identify the potential failure modes such as electrical, thermal, and optical degradation during the qualification test. To detect anomalies, a similarity-based-metric test has been developed to identify anomalies without utilizing historical libraries of healthy and unhealthy data. This similarity-based-metric test extracts features from the spectral power distributions using peak analysis, reduces the dimensionality of the features by using principal component analysis, and partitions the data set of principal components into groups using a KNN-kernel density-based clustering technique. A detection algorithm then evaluates the distances from the centroid of each cluster to each test point and detects anomalies when the distance is greater than the threshold. From this analysis, dominant degradation processes associated with the LED die and phosphors in the LED package can be identified. When implemented, the results of this research will enable a short qualification time.

Prognostics of LEDs are developed with spectral power distribution (SPD) prediction for color failure. SPD is deconvoluted with die SPD and phosphor SPD with asymmetric double sigmoidal functions. Future SPD is predicted by using the particle filter algorithm to estimate the propagating parameters of the asymmetric double sigmoidal functions. Diagnostics is enabled by SPD prediction to indicate die degradation, phosphor degradation, or package degradation based on the nature of degradation shape of SPD. SPDs are converted to light output and 1976 CIE color coordinates using colorimetric conversion with color matching functions. Remaining useful life (RUL) is predicted using 7-step SDCM (standard deviation of color matching) threshold (i.e., 0.007 color distance in the CIE 1676 chromaticity coordinates).

To conduct prognostics utilizing historical libraries of healthy and unhealthy data from other devices, this research employs similarity-based statistical measures for a prognostics-based qualification method using optical, electrical, and thermal covariates as health indices. Prognostics is conducted using the similarity-based statistical measure with relevance vector machine regression to capture degradation trends. Historical training data is used to extract features and define failure thresholds. Based on the relevance vector machine regression results, which construct the background health knowledge from historical training units, the similarity weight is used to measure the similarity between each training unit and test unit under the test. The weighted sum is then used to estimate the remaining useful life of the test unit.