Blackbox optimization applied to Design of Experiments

The pharmaceutical industry relies on a process known as batch manufacturing to supply their goods to the public. Each batch takes time to ends, involves many parameters and is very expensive. Quality tests are performed at the end of each batch. The calibration of parameters is a crucial task and Design of Experiment (DoE) is the traditionally approach used. DoE is a predefined sequence of tests on parameters. Each test consists to run a batch with a given set of values of parameters and to measure the performance (quality tests) of that batch. DoE compiles obtained results and gives its best values of parameters. This project aims to explore an adaptive approach of DoE that consists to dynamically exploit the information from previous tests to determine the new set of values of parameters for the next test, in order to converge to a better set of values of parameters than a classical DoE approach, possibly with a less number of tests.

Faculty Supervisor:

Charles Audet

Student:

Partner:

Sanofi

Discipline:

Mathematics

Sector:

Health and Related Sciences & Technology; Manufacturing; Other services (except public administration); Professional, scientific and technical services; Wholesale trade

University:

Polytechnique Montréal

Program:

Accelerate