Premium
Investigating Reliability on Fuel Cell Model Identification. Part I: A Design of Experiments Approach
Author(s) -
Tsikonis L.,
Van herle J.,
Favrat D.
Publication year - 2011
Publication title -
fuel cells
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.485
H-Index - 69
eISSN - 1615-6854
pISSN - 1615-6846
DOI - 10.1002/fuce.201100036
Subject(s) - weighting , optimal design , design of experiments , reliability (semiconductor) , sensitivity (control systems) , covariance , mathematical optimization , computer science , identification (biology) , constant (computer programming) , basis (linear algebra) , set (abstract data type) , covariance matrix , solid oxide fuel cell , algorithm , mathematics , statistics , power (physics) , engineering , chemistry , electronic engineering , biology , geometry , quantum mechanics , radiology , programming language , medicine , physics , botany , electrode , anode , machine learning
A model‐based Design of Experiments method was employed for the optimization of measurements on a solid oxide fuel cell (SOFC). Based on a simplified SOFC model, a variation of the D‐optimality was used as optimization criterion for the calculation of optimal experimental designs (determinant of the covariance matrix with weighting factors). Solutions for different numbers of design points were calculated and the behavior of optimization criteria as functions of the number of design points as well as of the number of repetitions of measurements was analyzed. A new type of graph was introduced which depicts the behavior of optimization criteria for constant number of measurements. This approach showed that, for constant numbers of measurements, the precision is higher and therefore the reliability in the cell's model identification is improved when repeated measurements of a small set of optimal design points are effectuated, instead of many different measurements. Finally, a sensitivity analysis was performed showing the influence of the parameter values on the values of the optimization criterion and the optimal measurements. The used methodology and its theoretical conclusions may be used as a basis for development of diagnostics tools or filtering existing data for optimal parameter estimations.