Melt Scheduling to Trade Off Material Waste and Shipping Performance

Kedar S. Naphade, S. David Wu, Robert H. Storer, Bhavin J. Doshi
2001 Operations Research  
The ingot formation or "melt" process is the first step in many steel making operations. This process involves melting steel and alloys for desired chemical composition, then pouring it into a variety of ingot molds. Complex technological and resource constraints can make the planning and scheduling of such processes extremely challenging. In this paper, we report our experience in developing solution methods for this "melt scheduling" problem at BethForge, a division of the Bethlehem Steel
more » ... oration, and a manufacturer of custom-made heavy steel forgings. We describe the main issues associated with generic melt scheduling problem as well as constraints that are specific to BethForge. The problem at BethForge is particularly challenging due to the need to keep the ingot at a high temperature before forging, their high product variety, and the need to consider trade-offs between two conflicting objectives. We first formulate the base melt scheduling problem as a mixed integer program. We then decouple the scheduling decisions into two levels and develop a local search algorithm based on Storer and Wu's problem space neighborhood. Our aim is to generate a family of efficient schedules that allow decision makers to balance the tradeoff between two criteria. Computational experiments are performed using data from BethForge. The melt scheduling procedure developed herein has been implemented and installed at BethForge. It made fundamental improvement in their melt scheduling process. Steel ingots form the raw material for a variety of steel products that are processed through forging, heat treatment and other similar steel manufacturing processes. Steel manufacturers either have ingots delivered to them or have ingot formation processes at the front end of the plant. The ingot formation or "melt" process is extremely critical because of its capital and energy intensive nature, and the effect it has on the material flow of the downstream production processes. These factors, coupled with a complex set of resource and technological constraints, make melt scheduling a very challenging problem. When we started our study at BethForge in 1995 melt scheduling was carried out manually by seasoned schedulers who were intimately familiar with the complex engineering rules and constraints involved in the process. Later in 1995, Bethlehem Steel decided to move the melt operation from within the Bethlehem plant to Steelton, PA (some 80 miles away), where ingots for other steel products are also poured. The forging, heat treatment and machining processes remain in the Bethlehem plant. Steel ingots are poured on a weekly basis at Steelton, then transported by rail carts to Bethlehem. This new process creates major challenges to the manual scheduling system as the melt operation is now limited to a specific time slot during the week, strict weight restrictions are imposed by the Steelton melting resources, and frequent rescheduling becomes necessary due to variations in crew configuration, maintenance plan, resource availability, and rail transportation constraints. Our study herein is part of a joint NSF/Private sector research initiative where we seek to restructure and develop more robust and practical solutions for production scheduling in industry. In Naphade et al. (1996) and Doshi et al. (1996) some of the preliminary results in melt scheduling at BethForge were documented. The proposed melt scheduling procedure has been coded, tested and demonstrated to BethForge management. The procedure has later been integrated into their planning and scheduling system. In this paper we report in detail our experience in developing the overall solution methodology for this particular problem and computational results obtained for real instances at BethForge. The paper is organized as follows: Section 1 explains the process of steel ingot formation, Section 2 provides a detailed description on the melt scheduling problem at BethForge, Section 3 present a MIP formulation for the stated melt scheduling problem, and Section 4 describe a two-level heuristic for the solution of the problem. In Section 5 we presents computational experiments using actual order data, and in Section 6 we discuss experiences and insights gained from implementing the scheduling system at BethForge.
doi:10.1287/opre.49.5.629.10611 fatcat:m6okxrs5ova5ri34eat2dh6etu