In this paper, a systematic framework for synthesis and design of processing networks under uncertainty is presented. Through the framework, an enterprise-wide optimization problem is formulated and solved under uncertain conditions, to identify the network (composed of raw materials, process technologies and product portfolio) which is feasible and have optimal performances over the entire uncertainty domain. Through the integration of different methods, tools, algorithms and databases, the framework guides the user in dealing with the mathematical complexity of the problems, allowing efficient formulation and solution of large and complex enterprise-wide optimization problem. Tools for the analysis of the uncertainty, of its consequences on the decision-making process and for the identification of strategies to mitigate its impact on network performances are integrated in the framework. A decomposition-based approach is employed to deal with the added complexity of the optimization under uncertainty. A network benchmarking problem is proposed as a benchmark for further development of methods, tools and solution approaches. To highlight the features of the framework, a large industrial case study dealing with soybean processing is formulated and solved.
The process industry sector is characterized by large capital investments, which are necessary for construction of production sites and facilities. The erection and commissioning of large production sites implies the use of massive amounts of economical, environmental and societal resources. The accuracy of the decision-making and of the design process is therefore of crucial importance, both for the enterprise which is committed to the investment, and for the human society in which the enterprise operates. Several tools have been developed and adopted in order to guide, support and facilitate the decision making process in capital investment projects, such as process management and project portfolio management (Project Management Institute, 2008).
Recent developments in Process Systems Engineering (PSE) have been focusing on formulating and solving processing network problems under the framework of enterprise-wide optimization (Grossmann, 2005). In this approach, the decision-making problem is cast in the form of superstructure optimization, which is formulated and solved as a Mixed Integer Linear or Mixed Integer Non Linear Programming (MIP or MINLP), making use of the integer and binary variables to represent discrete and binary choices.
The main strength of the enterprise-wide optimization approach is in its ability to provide comprehensive and transparent inputs to the decision makers, through a systematic and quantitative analysis. On the other hand, it poses several challenges, due to the size and complexity of the mathematical problem to formulate and solve, as well as to the amount of data which are required (Varma, Reklaitis, Blau, & Pekny, 2007). Often, the nature of the problem requires the formulation of large scale non-linear and non-convex problems (Karuppiah & Grossmann, 2006) whose solution to global optimality is still an open problem. Finally, the inclusion of data uncertainty in the decision-making problem causes a significant increase in problem size and complexity (Dua and Pistikopoulos, 1998, Karuppiah and Grossmann, 2008 and Paules and Floudas, 1992; Sahinidis, 2004). Because of this complexity, formulation and solution of real industrial problems require considerable time and resources investment, as well as deep knowledge of optimization theory and algorithms.
For these reasons, we believe in the importance of developing of a systematic framework for enterprise-wide optimization particularly to motivate and facilitate its use in practice. The integration of state-of-art methods, tools and solution strategies, in a framework for enterprise-wide optimization has in fact the potential of increasing the productivity of the workflow needed to formulate and solve this class of problems; and thereby to enable the use of this powerful tool in industry and public sector, supporting transparent and efficient decision-making process.
In line with these considerations, in this manuscript we propose a systematic framework for synthesis and design of processing networks under uncertainty. The framework is based on the integrated business and engineering framework developed earlier (Quaglia, Sarup, Sin, & Gani, 2012a), which is extended to include decision-making under uncertainty.
The structure of the manuscript is as follows. In Section 1 the framework is described, by highlighting the mathematical formulation of the problem and the integration among the different methods and tools. In the Section 2 a Benchmark Network Problem (BNP) is proposed, and its formulation and solution according to the proposed framework is discussed. In the Section 3, the capability of the framework to deal with the size and complexity of an industrial problem is demonstrated, by formulating and solving a large scale case-study which is about synthesis of soybean processing network under uncertainty. Finally, conclusions and future works are presented in the last section.
A systematic framework for synthesis and design of processing networks has been proposed. By integrating methods, tools and solution strategies in a software infrastructure, the framework facilitates and optimizes the workflow required for formulation, solution and analysis of enterprise-wide optimization problems. Instead of focusing on the direct solution of the problem of decision-making under uncertainty, the framework guides the user through some preliminary steps in which the problem, the uncertainty and its consequences on the decision-making process are analyzed. The result is a comprehensive package, which contains, alongside the optimal network, a large amount of other information that constitute a valuable input for business, engineering and process management decisions. Moreover, the framework integrates an incremental solution strategy, which allows the solution of complex problems.
A network benchmark problem (NBP) has been proposed and solved to demonstrate the framework capability. It is the intention of the authors that the NBP should be treated as a benchmark problem for the development of integrated business and engineering decision-making methods and tools.
Finally, a large scale industrial problem has been studied, demonstrating the capability of the framework to manage the complexity of a real problem, subject to market, technical and raw material quality uncertainty. The solutions obtained through the application of the framework showed outstanding performance in mitigating the consequences of the uncertainty.
On the base of the presented results, our future works will focus on integrating sensitivity and ranging analysis methods and tools in our framework. Moreover, the framework will be employed for the formulation of case studies, selected from different industry segments (e.g. water and wastewater networks, production networks, biorefinery, etc.) in order to test and demonstrate its flexibility.
