Bhosekar, Atharv. Supply chain optimization and modular process design using machine learning-based frameworks. Retrieved from https://doi.org/doi:10.7282/t3-r19s-a356
DescriptionGlobalization and the sudden increase in the exchange of information, trade, and capital all around the world, driven by technological innovation, has given rise to complex global supply chain networks. Since optimally designing such networks can yield significant profits in the long run, supply chain optimization is an area of active interest. The motivation for the problem considered in this work is two-fold. First, in a modern supply chain network, data plays an important role. However, since traditional optimization solvers cannot readily make use of this data, there is a need for frameworks that can utilize data to make optimal decisions in such networks. Second, there is a growing interest in considering multiple levels of decisions while designing the supply chain. However, due to differences in scale, level of details, and computational expense of the resulting integrated model, the problem of integrated decision-making is challenging. This work aims to propose machine learning-based frameworks that address these challenges.
The problem of optimal inventory allocation is first solved in a multienterprise supply chain network where the supply chain model is available in the form of a complex simulation. Further, historical data of the process or data generated from process simulations is used to design the process while simultaneously considering the total cost of the process as well as the flexibility of the design obtained. The framework is applied to modular processes. It is shown that using machine learning-based frameworks, process-level details can be incorporated at the supply chain design stage. This approach allows quantitatively assessing the benefits of modular processes such as design standardization, reduced transportation cost due to decentralized manufacturing, and optimal production facility location. Finally, the study is extended to address the problem of multiperiod supply chain optimization under product demand uncertainty. The results demonstrate the efficacy of the machine learning-based optimization framework proposed in this work and yields a set of solutions that minimize the risk as well as the expected total cost of the supply chain network.