Mixed integer optimization has been used extensively to model greenfield and plant expansion designs across multiple industries, including semiconductors, water utility, forest products, paper/FMCG, chemicals and more. Greenfield plant models may be used to test possible equipment and real-estate configurations for profitability, cash flow, NPV and ROI.
Real-world Example
Many of our customers have had great success using mixed-integer optimization to conduct greenfield evaluations. An example evaluation conducted by one such customer involved analyzing various plywood manufacturers. The main goal was to try different combinations of capital equipment and different configurations of the plant floor with different sizes of inventory capacities in order to determine the best possible mix of equipment and layout, while taking into account:
- Construction cost
- Capital equipment cost
- Real estate cost
- Labor cost
- Material handling cost
- Demand profiles
- And more.
The overarching goal was, of course, to maximize ROI and profitability.
The optimization model demonstrated how subtle changes in inventory surge capacity or variations in a single piece of equipment could change the whole dynamic of the plant’s performance, either adding or subtracting millions from the lifetime value of the assets. A key ingredient to success in this sort of modeling exercise was access to domain experts who helped define the right questions to ask, assist with data and help interpret results.
Mixed Integer Optimization Over Simulation
Simulation methods can also be used for such a modeling effort as designing a new plant. However, by using the simulation approach we miss many of the dynamic reactionary insights gained by the use of optimization. This is because optimization techniques can reroute flows to overcome new bottlenecks that arise in ways that a simulation engine cannot.
Therefore, I would strongly recommend anyone designing a new facility to use a mixed integer optimization model in their design efforts. Mixed integer optimization, by definition, will find the best possible result. Simulation simply finds a result, but not necessarily the best result.
