(755d) Turnaround Optimization for Continuous Chemical Plants

All chemical plants require periodic turnarounds, or planned shutdowns, for maintenance purposes. When dealing with large industrial complexes and integrated sites, the scheduling of such turnarounds becomes important, as each plant may need raw materials in specific ratios that can be sourced from one or more plants in the site, and products from the plant may be required downstream as raw materials to other plants as well. 

The turnaround time-scale of each turnaround task is on the order of weeks, and intervals between turnarounds are on the order of years. Each specific turnaround can cost millions of dollars to perform, and in addition, incurs lost sales. This makes turnaround scheduling decisions crucial from an economic viewpoint. Thus, the turnaround scheduling task is a long-term plan for managing the operation and preventive maintenance tasks on entire plants that are part of an enterprise-wide or site-wide network. 

The integrated site can be represented as a network with mass balances and ratio constraints. The scheduling of a maintenance task for a plant at a particular time affects the production plans of plants that are upstream as well as those downstream to it. Intermediate storage tanks may be available between plants to hedge against lost sales during shutdowns. Each plant may have to undergo multiple kinds of turnarounds, and each turnaround is required to be performed within a certain time interval from the previous turnaround on a particular plant. In addition, there are demand constraints on each of the final products, production capacity constraints on each of the plants, constraints on the availability of maintenance manpower and the possibility of selling intermediate products to the market. Each turnaround task is associated with manpower and take-down costs. The objective is to maximize long-term profit by managing inventory, intermediate sales, and scheduling decisions, while satisfying the above constraints. 

We formulate a discrete-time mixed-integer programming model to solve this problem. As it is generally not possible to have discretizations as fine as one would like, we validate our solution using a discrete-event simulation model, which we also use to test the robustness of the schedule to certain uncertainties. We present results on a large-scale industrial test case.