(648c) Multistage Distributionally Robust Mixed-Integer Optimization for Integrated Production and Maintenance Scheduling | AIChE

(648c) Multistage Distributionally Robust Mixed-Integer Optimization for Integrated Production and Maintenance Scheduling


Feng, W. - Presenter, Zhejiang University
Zhang, Q. - Presenter, University of Minnesota
Feng, Y., Zhejiang University
In manufacturing, production tasks are frequently performed to meet product demands, which generally cause the degradation of equipment units (also referred to as assets). As a result, maintenance activities are required to maintain a minimum level of performance and avoid equipment failures, which in turn affect the production schedule, e.g. if unit shutdown is required for maintenance. Obviously, managing production and maintenance activities independently without consideration of their tight relation can be severely suboptimal, which has motivated recent efforts in jointly optimizing production and maintenance. In this work, we extend the integrated production and maintenance scheduling model proposed by Biondi et al. (2017) to consider both offline and online maintenance tasks. In addition, changes in asset performance (here reflected in the production capacity) due to online maintenance and changes in the asset’s health state are also taken into account.

The main novelty of this work lies in the consideration of uncertainty, both in product demand and in the residual useful life indicator, where the latter captures the inherently stochastic degradation process and uncertainties originating from measurement errors and inaccuracies in predictions models. We propose a multistage distributionally robust optimization framework based on a decision rule approach developed in previous work (Feng et al., 2020), which considers both binary and continuous recourse decisions in every stage. We minimize the expected cost under the worst-case distribution within a Wasserstein ambiguity set. A tractable mixed-integer linear programming (MILP) reformulation of the problem is derived, and the effectiveness of the proposed methodology is demonstrated in an extensive numerical case study as well as in a real-world ethylene plant case. The results show that the proposed distributionally robust optimization approach can lead to significantly improved out-of-sample performance while maintaining computational tractability.


Biondi, M., Sand, G., & Harjunkoski, I. (2017). Optimization of multipurpose process plant operations: A multi-time-scale maintenance and production scheduling approach. Computers & Chemical Engineering, 99, 325-339.

Feng, W., Feng, Y., & Zhang, Q. (2020). Multistage robust mixed-integer optimization under endogenous uncertainty. Under review.