This paper presents a systematic methodology to optimize the module instance configuration of an evolving product family (PF). The proposed methodology seeks to maximize the total profit of a PF for a given planning horizon by taking into account the interdependencies of modules at both the product level and the PF level. The module configuration optimization problem can be viewed as a stage-wise sequential decision process. Dynamic programming (DP) is suitable for modeling such problems. The DP-based methodology proposed in this paper breaks up the PF module instance configuration optimization problem into smaller DP optimization problems involving module groups based on the independence assumption of profit change due to the module replacement strategies. The aggregation concepts of independent module groups and module clusters are used to decrease the state space in the DP model. A module group is defined as a group of interacting modules linked by the replacement dependence relationships in a PF. A module cluster is defined as the modules within a module group that are strictly inter-dependent on each other in replacement actions. A DP model is established for each of the module groups to optimize the PF through individually optimizing the module configuration of the individual module groups. In the DP model, the time points with equal intervals during the planning horizon are considered as stages; the decision of module configuration strategies is defined as the control variable; the feasible combinations for the modules within one module group are selected as the states at each stage; and profit change (benchmarked with respect to profit without any module replacements) is treated as the contribution function that needs to be maximized. In the deterministic model, the expected change in profits and expected time of module instance availability are assumed to be deterministic. In the stochastic model, the profit change and the time of module instance availability are treated as uncertain events. The Monte Carlo method is used to simulate the total profit change distributions subjected to the uncertainties of data and module instance availabilities. We use an illustrative PF example to demonstrate how the suggested models can be used to optimize the PF architecture.

This content is only available via PDF.
You do not currently have access to this content.