110th Anniversary: On the Departure from Heuristics and Simplified Models toward Globally Optimal Design of Process Equipment

Despite the large advances in computational tools attained by the Process Systems Engineering community, industry for the most part still performs basic design of process equipment using trial-and-verification procedures guided by heuristic rules. In this opinion article, we present a discussion on how to depart from the use of these heuristics-based procedures, most step-by-step, sometimes computer-aided. We believe that there are direction changes, some incipient and some in full development already, toward the use of optimization tools for the task. The academic literature is dominated by mixed-integer nonlinear models, solved using different techniques (mostly stochastic or mathematical programming-based). These procedures have practical limitations that have hindered the migration of practitioners away from current heuristics and simplified model-based tools. We discuss these drawbacks and propose solutions. We first show how the use of commercially available discrete values of the design geometrical variables followed by reformulation can render linear models, solved using mixed-integer linear programming, sometimes integer linear programming. We also show how reformulation or judicious discretization of continuous variables can lead to a large reduction in the number of nonlinearities, such that they can be solved using commercial and noncommercial global solvers with larger efficiency and robustness. Next, we present the use of set trimming to aid global optimization of equipment designs, as means of reducing the search space. Then, we present the technique of smart enumeration, which can be used instead of MINLP models and eventually after set-trimming. We also discuss changes in modeling and propose to move away from simple models, where we argue for abandoning physical properties as well as transport-like coefficients that are based on averages, to use properties and transport coefficients calculated locally inside and along the equipment. Then, we discuss the impact of the above proposals on flow sheet synthesis and retrofit problems. As these advances show higher memory and speed needs, we discuss how parallel computing (using the cloud, small computer clusters, or supercomputers) can address the aforementioned challenges.