A generalized data-driven production planning model: algorithmic foundation and simulation-based performance assessment

Production planning models must consider the non-linear relationship between resource workload and cycle times when timing work releases to meet demand. This is particularly crucial in semiconductor manufacturing where cycle times are of the order of ten weeks and subject to fluctuations at high utilization levels. Data-driven (DD) production planning formulations have been proposed to address this issue as an alternative to clearing function (CF)-based formulations. DD models characterize a production system through a set of discrete system states representing the multivariate relationship between work in process (WIP) and throughput of all products under steady-state conditions. The cycle times are usually not negligible in relation to the period length. The WIP-throughput relationship alone does not sufficiently capture the congestion behavior of the system under these conditions. The planning model must also correctly identify the periods in which the throughput will be realized as output. We investigate the temporal relationship of WIP and output under different cycle time regimes, and propose the concept of flow patterns that specify the expected input and output quantities over several periods based on the WIP at a specific point in time, and extend the concept of system states. To accommodate these patterns, we establish a DD planning formulation that generalizes previous DD models and outperforms other DD variants and the Allocated CF (ACF) formulation in computational experiments using a large-sized wafer fab model.