Assessments of experimental designs in response surface modelling process: Estimating ventilation rate in naturally ventilated livestock buildings

• Five experimental design methods are compared in this study for the performance in RSM modelling.
• The FDS tool is evaluated in this study.
• The effect of terms reduction and response transformation on the model accuracy are discussed.

Precise modelling the ventilation rate through a naturally ventilated livestock building can benefit the control of indoor climate and reduction of ammonia emission. In terms of agricultural dairy buildings, the modelling of ventilation rates may involve in several variables, including the opening sizes at side walls and the outdoor wind conditions. A statistical modelling process requires knowing how the experiment is designed and what modelling technique is followed. In this paper, several different methods for design of experiment (DOE) such as central composite rotation design (CCRD), optimal design (OPD), Box–Behnken design (BBD) and space filling design (SFD) were compared for their accuracies of the acquired models and numbers of experimental runs. Response surface methodology (RSM) was applied and discussed for modelling the ventilation rate in relation to those variables. Results demonstrated the BBD had the best performance in the model development. The fraction of design space (FDS) tool was also evaluated for its ability in comparing different DOE methods and results showed that this tool performed inadequately in comparing between traditional DOE methods such as CCRD, BBD and FFD and modern DOE methods, such as OPD and SFD.