A Bayesian model for the failure rate estimation of photonic components

Estimating the reliability of new, products is an important part of new product development. There are several ways in which this reliability can be estimated, including the use of: field data, empirical reliability prediction models, test data, and reliability physics based models. Field data is usually the most desirable for estimating the reliability of a product since it represents the deployment conditions and therefore represents all applicable stresses to which the item is exposed. There is typically not enough operational field data to estimate the reliability of products in a statistically robust manner, and therefore using field data alone leads to low statistical confidence inferences on the measure of failure rates. This paper presents a practical and structured approach to estimating the reliability of components used in photonic systems. This approach consists of a generic model that allows for the "fusion" of various data types (e.g., tests, models, simulations), and merges available data with a Bayesian approach to form the "best estimate" of component reliability. This paper explains the theoretical robustness of the method, and demonstrates the Bayesian method. The approach is particularly suitable for photonic components given the short development times typical in the telecommunications industry and the resulting relative scarcity of field data.