Selen Cremaschi

Adaptive sequential sampling for surrogate model generation with artificial neural networks

Abstract Surrogate models – simple functional approximations of complex models – can facilitate engineering analysis of complicated systems by greatly reducing computational expense. The construction of a surrogate model requires evaluation of the original model to gather the data necessary for building the surrogate. Sequential sampling procedures are proposed for determining and minimizing the required number of samples for efficient global surrogate construction. In this paper, two new adaptive sampling algorithms – one purely adaptive and one combining adaptive and space-filling characteristics – are proposed and compared to a purely space-filling approach. Our analysis suggests a mixed adaptive sampling approach for constructing surrogates for systems where the behavior of the underlying model is unknown. Results of the case study, optimization of carbon dioxide capture process with aqueous amines, revealed that the mixed adaptive sampling algorithm may reduce the required sample size by up to 40% compared to a purely space-filling design.

Process synthesis of biodiesel production plant using artificial neural networks as the surrogate models

Abstract Biodiesel is an attractive biofuel because it can be used directly with the traditional petro diesel engines, either as a substitute or as a blending component. There are several alternatives that can be used to produce biodiesel. In this work, we developed a superstructure optimization model to synthesize the optimum biodiesel production plant, i.e., the one that gives the minimum net present sink. To reduce the computational cost of solving the resulting disjunctive programming, the surrogate models utilizing artificial neural networks (ANNs), have been developed to replace the unit operation, thermodynamics and mixing models. The optimum solution with the alkali-catalyzed reactor (obtained in five CPU seconds) has a total net present sink of about

A perspective on process synthesis: Challenges and prospects

41 million, which differs less than one percent from the result obtained by modeling the solution in a process simulator. However, this level of accuracy required a large amount of data to train the ANNs.

Heuristic solution approaches to the pharmaceutical R&D pipeline management problem

Abstract This paper gives the authors perspective on some of the open questions and opportunities in process synthesis focusing on separation systems as the application. Driven by energy and environmental concerns and challenged by introduction of new raw materials, this author anticipates significant advances in: (1) novel approaches that integrate experimental studies and process synthesis activities, and multi-scale and surrogate models for accurately capturing the behavior of these unconventional mixtures, (2) systematic generation of alternatives for processing these mixtures, and (3) global, robust, and stochastic optimization for identifying the optimum alternative. This paper is an extended version of a conference paper ( Cremaschi, 2014 ) presented at the 8th International Conference on Foundations of Computer-Aided Process Design.

Experimental Study of Low Concentration Sand Transport in Multiphase Air–Water Horizontal Pipelines

Abstract The paper presents two heuristic approaches, a shrinking horizon multiple two-stage stochastic programming (MTSSP) decomposition algorithm and a knapsack decomposition algorithm (KDA), for solving multistage stochastic programmes (MSSPs) with endogenous uncertainty, specifically focusing on pharmaceutical research and development (R&D) pipeline management problem. The MTSSP decomposition algorithm decomposes the problem into a series of two-stage stochastic programmes, which are solved as resources become available. The KDA decomposes the MSSP into a series of knapsack problems, which are created and solved at key decision points on a rolling horizon fashion. Based on the results of the six case studies, both the MTSSP decomposition algorithm and the KDA generate implementable solutions that are within three percent of the rigorous MSSP solution obtained by CPLEX 12.51. Both methods showed several orders of magnitude decrease in the CPU times compared to ones that were required to solve the rigorous MSSP.

Variants to a knapsack decomposition heuristic for solving R&D pipeline management problems

The ultimate goal of this work is to determine the minimum flow rates necessary for effective transport of sand in a pipeline carrying multiphase flow. In order to achieve this goal, an experimental study is performed in a horizontal pipeline using water and air as carrier fluids. In this study, successful transport of sand is defined as the minimum flow rates of water and air at which all sand grains continue to move along in the pipe. The obtained data cover a wide range of liquid and gas flow rates including stratified and intermittent flow regimes. The effect of physical parameters such as sand size, sand shape, and sand concentration is experimentally investigated in 0.05 and 0.1 m internal diameter pipes. The comparisons of the obtained data with previous studies show good agreement. It is concluded that the minimum flow rates required to continuously move the sand increases with increasing sand size in the range examined and particle shape does not significantly affect sand transport. Additionally, the data show the minimum required flow rates increase by increasing sand concentration for the low concentrations considered, and this effect should be taken into account in the modeling of multiphase sand transport.

A Quick Knapsack Heuristic Solution for Pharmaceutical R&D Pipeline Management Problems

Abstract The knapsack decomposition algorithm (KDA) (Christian and Cremaschi, 2015) decomposes the R&D pipeline management problem into a series of knapsack problems, which are solved along the planning horizon. It yields tight feasible solutions, and improves the solution times by several orders of magnitude for large instances. This paper investigates the impact of problem parameters and size, and KDA decision rules on KDA solution quality and time. The decision rules are (1) timing of new knapsack problem generations, and (2) formulation of the resource constraints in knapsack problems. The results revealed that the KDA decision trees were insensitive to problem parameters, and the KDA solution times grew super-linearly with linear increases in the length of the planning horizon and the number of products. The results suggest that the KDA where knapsack problems are generated after each realization with the original resource constraint yields the most accurate solutions in the quickest time.

A new representation for modeling biomass to commodity chemicals development for chemical process industry

Abstract The paper presents a novel knapsack decomposition algorithm for solving multistage stochastic programs (MSSPs) with endogenous uncertainty, specifically focusing on pharmaceutical R&D pipeline management problem. The algorithm decomposes the MSSP into a series of knapsack problems, which are created and solved at key decision points on a rolling horizon fashion. Based on the results of the five case studies, the algorithm generates implementable solutions with several orders of magnitude decrease in the CPU times required to solve the rigorous MSSP.

Efficient Surrogate Model Development: Optimum Model Form Based on Input Function Characteristics

Abstract A new representation combining network and stage-gate frameworks to study the impact of capital and research & development (R&D) decisions on the evolution of biomass to commodity chemicals system is presented. The network representation is used to universally express the interconnections between the processing technologies and to track the material flow among them. The stage-gate representation is used to express the discrete nature of technology maturity levels. The corresponding mixed-integer nonlinear program is developed and solved for a case study. In this case study, ethylene and propylene can be produced from naphtha and/or biomass. The results of the sensitivity analysis reveal that the raw material costs are the dominant factors that dictate the optimum investment decisions and production plan for this system.

Estimation of percentiles using the Kriging method for uncertainty propagation

Abstract Surrogate models statistically relate input data to output data, which are collected by running the complicated system simulation. This study focuses on the identification of the proper surrogate-model form using computational experiments. Eight different surrogate-modeling approaches are evaluated using thirty-five challenge functions with various shapes and numbers of inputs. Data sets for training the surrogate models are generated using Latin Hypercube, Sobol, and Halton sampling methods. In general, ANN, ALAMO and ELM provided the lowest root mean square error and maximum absolute errors for all functions tested.