Electronic part obsolescence forecasting based on time series modeling

The obsolescence of electronic parts can cause a serious problem to sustain systems. In order to properly manage the obsolescence problems, reasonable and efficient obsolescence forecasting is required. In this paper, a time series based obsolescence forecasting method is proposed for electronic parts. The proposed method consists of three phases: a time series alignment of a part family for capturing important patterns, Box-Cox transformation for improving forecasting accuracy, and time series modeling for forecasting life cycle curves. In comparison to an existing method of product life cycle curve forecasting with evolutionary parametric drivers (Sandborn’s method), the proposed method can predict life cycle curves effectively while overcoming the limitations (requiring evolutionary parametric drivers, and assumption of Gaussian distribution) of the Sendborn’s method. A numerical example was provided to demonstrate the three phases of the proposed method. The results showed that the life cycle curves of flash memories could be predicted without the assumptions that the Sandborn’s method requires to generate.