Surrogate-based feasibility analysis for black-box stochastic simulations with heteroscedastic noise

Zilong Wang, Marianthi Ierapetritou

Research output: Contribution to journalArticlepeer-review

4 Scopus citations


Feasibility analysis has been developed to evaluate and quantify the capability that a process can remain feasible under uncertainty of model inputs and parameters. It can be conducted during the design stage when the objective is to get a robust design which can tolerate a certain amount of variations in the process conditions. Also, it can be used after a design is fixed when the objective is to characterize its feasible region. In this work, we have extended the usage of feasibility analysis to the cases in which inherent stochasticity is existent in the model outputs. With a surrogate-based adaptive sampling framework, we have developed and compared three algorithms that are promising to make accurate predictions on the feasible regions with a limited sampling budget. Both the advantages and limitations are discussed based on the results from five benchmark problems. Finally, we apply such methods to a pharmaceutical manufacturing process and demonstrate its potential application in characterizing the design space of the process.

Original languageEnglish (US)
Pages (from-to)957-985
Number of pages29
JournalJournal of Global Optimization
Issue number4
StatePublished - Aug 1 2018

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Management Science and Operations Research
  • Control and Optimization
  • Applied Mathematics


  • Adaptive sampling
  • Feasibility analysis
  • Stochastic Kriging
  • Stochastic simulation
  • Surrogate modeling


Dive into the research topics of 'Surrogate-based feasibility analysis for black-box stochastic simulations with heteroscedastic noise'. Together they form a unique fingerprint.

Cite this