The first layer handles capacity allocation. This layer distributes orders over time by due date at casting, levels casting demand loads respecting major capacity constraints, allocates “swing orders” to multiple casters, and generally satisfies global constraints.
The second layer is the heat building model, which combines orders into heats, loads heats into available capacity, and concatenates similar heats, while respecting the next level of capacity constraints.
The third and final layer creates the detailed production sequence at steelmaking, ladle treatment and continuous casting, manages the liquid iron inventory balance, and respects the most detailed constraints.
The optimization layers iterate until the total KPI balance is achieved. The first layer executes, and control is passed to the next level down. If infeasibility is detected at the second or third layer, control is returned to the next level up for resolution. The model cycles up and down between layers until the total KPI balance requested by the planner is achieved.
Thanks to the decomposition of the scheduling problem into layers and the usage of the best fitting combinations of solvers for each layer the computations are executed as efficiently as possible.
Scheduling information system
The new method of scheduling was used as a base for development of a solution for the Czech production company. Also the Czech company specialists and LOGIS project team members defined a set of key requirements for a new solution.
Whether assessing the relative merits of multiple scheduling scenarios, or reacting to shop floor problems that require immediate schedule changes, model execution speed is simply fundamental, and must be accomplished without compromising schedule quality.
No model, however sophisticated, can fully represent the richness and variation of physical reality. Therefore the planner must be able to easily make manual adjustments, with complete visibility to any potential constraint violations.
The quality of any melt shop and caster schedule can be measured along vectors that correspond to the goals for scheduling. The planner must have real-time visibility to the full set of Key Performance Indicators that reflect the value of the schedule, after every model run and after every manual adjustment.
The goals for melt shop and caster scheduling can have different relative importance for different steel plants, and can vary for the same plant as business conditions change. For example, when alloy prices are high relative to steel prices, it’s more important to control over-grading. The planner must be able to adjust the priority of caster scheduling goals, and have the model produce schedules that comply.
The planning environment evolves constantly: new facilities added, new products introduced, changed commercial priorities, etc. The melt shop and caster scheduling model must represent constraints in a way that allows planners to make model changes without long IT wait times. Constraints should be represented in the model as data and the planner should be able to add, change or delete constraints as needed.
To implement these and other requirements LOGIS specialists developed a new solution. During the implementation project the model of melt shop and caster scheduling was set up in the solution. Example of the working space of a planner with Gantt chart of melt shop and caster is provided on the Figure 4.