Simulation and Statistical Methods in Proactive and Strategic Obsolescence Management

Abstract

In fields with sustainment-dominated systems, where the sustainment costs are larger than the original system costs, managing the life-cycle costs is a non-trivial task. These fields include aviation, power plants, medical, military, telecommunications, and civilian safety. In recent years, many of these fields have begun to rely on commercial off-the-shelf (COTS) parts whose lifetimes are driven by market forces outside of the control of the system managers, serving to exacerbate obsolescence issues. While the procurement lifetimes of commercial electronics typically span 12-18 months, many of these fields have systems in excess of 30 years. Efficient management of supply chains and obsolescence issues is crucial to keep key systems operational. This dissertation provides proactive and strategic diminishing manufacturing sources and material shortages (DMSMS) (or obsolescence) management frameworks to forecast availability risk and identify optimal technology refresh strategies. First, a proactive technique evaluates reducing the bias associated with maximum likelihood estimation (MLE) for part-level forecasting. Second, a model is presented to measure the availability risk across a system of parts, taking into account DMSMS issue mitigation. Third, optimal technology refresh strategies are explored, focusing on the trade-off between lifetime buy and technology refreshes mitigation decisions for a system. The first contribution of this research identifies the amount of oversampling required to reduce the bias in Weibull parameter estimates for generalized Type I censoring. A common method to estimate Weibull parameters with censored data is to use MLE, but bias in parameter estimates tends to increase as the percent of censored observations increases and/or the sample size decreases. The bias can affect the accuracy of forecasting or simulating procurement lifetimes in a DMSMS context. This dissertation combines previous work to reduce the bias in the Weibull shape parameter and oversampling, with a technique akin to synthetic minority oversampling technique (SMOTE), to reduce the bias in the scale parameter. Using the Kullback-Liebler divergence, the amounts of oversampling that results in a decrease in the estimated distribution's deviation from the true distribution are identified according to the sample size and the percent of censoring for a data set. A case study is presented using microelectronic parts to highlight application of the proposed methodology. The second contribution of this research quantifies the risk associated with parts as they move through the procurement life-cycle curve. Often a single part becoming obsolete, or being non-procurable, does not lead to a serious DMSMS issue as these are quickly mitigated with like-part substitutions. In addition, having more than one manufacturer for a part is a recommended strategy to minimize the risk associated with DMSMS issues. A discrete event simulation (DES) model is created using a finite-source capacitated queuing model and two risk metrics are developed. The DES results are verified with analytic results. Additionally, model extensions are presented to improve the realism of the model. The third contribution of this research presents a strategic DMSMS management study which compares the trade-off between lifetime buys (LTBs) and technology refreshes over a system's lifetime. This research provides a way to quantify and compare proactive and reactive DMSMS management strategies. Two models are compared: the first uses a ranking and selection (R\&S) method while the second uses a rolling horizon (RH) framework. Optimal strategies from both models are compared with sensitivity to the relative costs between LTBs and technology refreshes. The solutions provide DMSMS managers a set of (near-)optimal scheduling strategies at minimal cost.