Designing the design of experiments (DOE) – An investigation on the influence of different factorial designs on the characterization of complex systems

Aleksandar Jankovic,Gaurav Chaudhary,Francesco Goia

doi:10.1016/j.enbuild.2021.111298

Aleksandar Jankovic, Gaurav Chaudhary + Show 1 more

Open Access

https://doi.org/10.1016/j.enbuild.2021.111298

Copy DOI

Abstract

Although a general set of guidelines and procedures for performing the design of experiments (DOE) exists, the literature lacks a recommended course of action for finding and selecting the optimal design of experiments among a large range of possible designs. This research tries to fill this gap by comprehensively testing more than thirty different DOEs through nearly half a million simulated experimental runs. The performance of various DOEs in the characterization of the thermal behaviour of a double skin façade (DSF) is assessed by comparing the outcomes of the different designs and using the full factorial design (FFD) as the ground truth. Besides the finding for the specific case study used in this investigation, this research allowed us to obtain some broad conclusions on the behaviour of different DOEs, which are summarized and translated into recommendations and a general decision tree chart for selecting the suitable DOE(s). The outcomes of this study help researchers and designers to apply DOEs that consider the extent of nonlinearity and interaction of factors in the investigated process in order to select the most successful and the most efficient designs for the specific process characterization.

Highlights

Developed primarily for agricultural purposes by British statistician Sir Ronald Fischer in the 1920s [1], the design of experiment (DOE) as a statistical method has been widely applied in different fields of science and industry, especially to support the design, development, and optimization of products and processes [2]
Results and Discussion, we present the outcome of our investigation of the case study based on the statistical analysis of variance performed on full factorial design (FFD) and its comparison with other DOEs
It was impossible to produce one FFD that encompassed all the ventilation modes since the air buffer (AB) mode does not have the same number of factors

Summary

Introduction

Developed primarily for agricultural purposes by British statistician Sir Ronald Fischer in the 1920s [1], the design of experiment (DOE) as a statistical method has been widely applied in different fields of science and industry, especially to support the design, development, and optimization of products and processes [2]. Well-established, general guidelines and procedures are available to support the implementation of DOE methods [3]. These steps include defining the objectives and response variables, determining factors, levels, experimental design type and experiment execution. The problem statement leads to establishing the objectives based on which the performance indicator (response variable) needs to be defined. As an essential step in the whole process, the factors affecting the performance indicator and how they are discretized, the number of experimental runs, and a suitable array need to be defined in the second stage [8]. The third stage covers the performance of the experiment according to the designed array and collection of data

Methods

Results

Conclusion