Optimization of manufacturing supply chains incorporating dynamic lead time and lot size modeling

Modern manufacturing and distribution practices have recently undergone significant restructuring from “make and sell” to “sense and respond”, with time-based competition having a significant impact on production system design and operation. Recent state-of-the-art Lean Manufacturing and Advanced Planning & Scheduling practices have certainly contributed to productivity growth. However, they remain primarily deterministic and thus unable to take advantage of further improvement promised by available probabilistic information.; Our proposed research overcomes state of the art limitations of ( i) fixed lead time (LT) assumptions, and (ii) the intractable nature of combinatorial (Mixed Integer Linear Programming) formulations of production systems characterized by lot size (set up) requirements. We develop methodology that exploits the natural time-scale-based decomposition of decisions with different scope and functionality. Our methodology builds on tractable decentralized cell specific sub problems associated with a short time scale and a fluid based master-problem associated with a longer time scale, to optimize, or continuously improve plant wide production planning. We employ an iterative interaction of a single production planning master problem in the long-time-scale Planning Coordination Layer (PCL), with multiple decentralized cell sub problems in the short-time-scale Performance Evaluation Layer (PEL). The master problem employs a Linear Programming optimization algorithm, while sub problems employ a multi-class Open Queuing Network (OQN) model. The OQN model is a fluid queuing network approximation used to model Dynamic Lead Time (DLT) delays reflecting lot size and uncertainties in service and set up times. Performance evaluation and sensitivity analysis is performed at the sub problem level over all of the short-time-scale units within each long-time-scale unit of the master problem, rendering them dynamic during the master problem's horizon. This enables dynamic piece-wise-linear approximation of non-linear queuing and set up delay response surfaces as a function of the production and set up schedule determined by the master problem.; Computational results show the effectiveness of dynamic LT modeling for optimal production and setup scheduling. Comparison with industrial practice and state-of-the-art approaches demonstrates the ability of our approach to improve LT performance in excess of 20%. We attribute the efficiency improvements to the value of the dynamic LT information employed. Finally, computational complexity appears to increase only linearly with problem size supporting the practicality of our approach.