Modular Design Optimization using Machine Learning-based Flexibility Analysis. (June 2020)

• Record Type: Journal Article
• Authors: Bhosekar, Atharv; Ierapetritou, Marianthi
• Is Part Of: Journal of process control. Volume 90(2020)
• Page Start: 18
• Page End: 34
• Publication Date: 2020-06
• Subjects: Modular design -- Design optimization -- Feasibility analysis -- Flexibility analysis -- Design under uncertainty -- Machine learning -- Operability; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2020.03.014
• Languages: English
• ISSNs: 0959-1524

Abstract

Highlights: This work proposes a novel approach for flexibility analysis problem with the help of machine learning-based classifier models. A multiobjective design optimization framework is proposed for maximizing flexibility and minimizing the cost. The proposed design optimization framework takes advantage of the modular design of the process by analyzing the feasibility of each module separately. Computational studies conducted using a case study of an air separation plant, demonstrate the efficacy of the proposed framework. Abstract: Recent studies on modular and distributed manufacturing have introduced a new angle to the traditional economies of scale that claim that large plants exhibit better efficiencies and lower costs. A modular design has several advantages, including higher flexibility of decisions, lower investment costs, shorter time-to-market, and adaptability to market conditions. While design flexibility is a widely studied concept in the process design, modular design provides an interesting new opportunity to the design optimization problem under demand variability. In this work, a framework for modular design under demand variability is proposed. The framework consists of two steps. First, the feasible region for each module is represented analytically with the help of the historical data or the data from a simulation using a classification technique. In the second step, the optimal design choice is obtained by integrating the classifier models built in the first step as constraints in the design optimization problem. The design optimization problem is first solved considering a single objective, i.e., minimizing the total cost or maximizing the flexibility. These two objectives are then addressed simultaneously using a multiobjective optimization framework that considers the tradeoff between maximizing the flexibility of design and minimizing the cost. Computational studies conducted using a case study of an air separation plant, demonstrate the efficacy of the proposed framework. Several advantages of using a modular design, as well as data-driven methods in the decision-making process in the design step, are discussed.