A Predictive Capability for Scheduling Furnace Operations in Titanium Production

Abstract

Component manufacturing within the aerospace industry demands that companies perform to the best of their ability, optimising their operations to maximise throughput. Hot rolling of titanium bloom products at TIMET was identified as an area where significant time and energy savings could be made. Bloom refers to a length of titanium with a large square or near-square cross-section. A range of different alloys and dimensions of bloom products are processed by TIMET on their rolling mill. Blooms are initially loaded to one of eight possible furnaces prior to hot rolling. Certain bloom types are simply preheated then finalrolled; whereas others have as many as two reheat operations and three rolling processes, while incorporating a flame-cutting operation to bisect the bloom prior to final reheat.This paper describes the development of a Genetic Algorithm model of this process, built within Microsoft Excel and utilising Palisade RiskOptimizer software [1]. The purpose of the model is to optimise the scheduling of blooms loaded to the furnaces, with the aim to reduce the total processing time for all jobs, termed the makespan. Use of this model will increase machine productivity whilst providing the associated monetary savings to the company's bottom line. This paper outlines the steps taken during construction of the model providing details of the measures taken to incorporate decisions and rules inherent within the manufacturing processes.The composition of the problem is that of a job-shop scheduling configuration with eleven machines (furnace loader, eight furnaces, rolling mill, and flame-cutter). Each furnace can process a maximum of four blooms at a time; therefore the blooms are grouped into job packets of 1, 2, 3 or 4 when considered for the model. The processing times for the tasks form the foundation of the model and were collected from actual bloom processing campaigns on site. These times were obtained for furnace loading, rolling, flamecutting, and for the furnace to reach the target temperature. There are many possible configurations of bloom number, bloom dimensions, furnace number, and target temperatures; therefore process times are stochastic variables that follow a statistical distribution (identified as log normal). This production scheduling model deals with uncertainty by simulating such times and minimising the mean makespan.The model was constructed in a stepwise fashion, ensuring validation after each modification or addition. Initial models scheduled only one bloom type on two furnaces and the rolling mill. Each subsequent model was built cumulatively on previous models leading to a greater number of machines and range of products, up to 11 machines and 4 product types. The current model can find an optimal schedule that satisfied the given constraints after 12,557 simulations.