Background Food sensory science and flavour analysis are key processes in new product development, and is essential in understanding consumers by bridging the gap between product characteristics and consumer perception and acceptance. Scope and approach This article provides a critical review of computer-based approaches to flavour and sensory analysis, including optimal design approaches to sensory experimental designs, and incorporation of nonlinear modelling methods such as artificial neural network into the analysis of results. The advantages and disadvantages of these methods, as well as their statistical background will be discussed. The incorporation of these statistical and mathematical methods into existing analytical processes is briefly covered, along with an overview of available computer software packages. Key findings and conclusions Food flavour and sensory analysis is an information gathering process, and can be divided into two main stages: (1) the design of the experiment; (2) analyses and interpretation of results. The choice of an analytical procedure in sensory and flavour science is crucial in obtaining information correlating food products and consumers. Traditionally, sensory analysis is based on classical experimental designs and linear multivariate analysis techniques. Computer algorithm-based methods such as optimal designs in the design of experiments, and artificial neural network as a non-linear regression method may be used in conjunction with current methods, or adopted to overcome potential shortfalls of existing methods.

Design of experiments and regression modelling in food flavour and sensory analysis: A review

http://www3.stat.sinica.edu.tw/statistica/oldpdf/A18n425.pdf

On general minimum lower order confounding criterion for s-level regular designs

An optimal design should minimize the confounding among factor effects, especially the lower-order effects, such as main effects and two-factor interaction effects. Based on the aliased component-number pattern, general minimum lower-order confounding (GMC) criterion can provide the confounding information among factors of designs in a more elaborate and explicit manner. In this paper, we extend GMC theory to s-level regular designs, where s is a prime or prime power. For an $s^{n-m}$ design D with $N=s^{n-m}$ runs, the confounding of design D is given by complementary set. Further, according to the factor number n, we discuss two cases: (i) $N/s<n\le (N-1)/(s-1)$, and (ii) $(N/s+1)/2<n\le N/s$. We not only provide the lower-order confounding information among component effects of D, but also obtain three necessary conditions for design D to have GMC.

Analysis on $$s^{n-m}$$ s n - m designs with general minimum lower-order confounding

Construction of some s-level regular designs with general minimum lower-order confounding

Based on the effect hierarchy principle, a good design should minimize the confounding among the lower-order effects. Thus, it is important to obtain the confounding information of effects of a des...

Lower-order confounding information of inverse Yates-order two-level designs

In this paper, we extend the general minimum lower-order confounding (GMC) criterion to the case of three-level designs. First, we review the relationship between GMC and other criteria. Then we introduce an aliased component-number pattern (ACNP) and a three-level GMC criterion via the consideration of component effects, and obtain some results on the new criterion. All the 27-run GMC designs, 81-run GMC designs with factor numbers and 243-run GMC designs with resolution or higher are tabulated. The Canadian Journal of Statistics 41: 192–210; 2013 © 2012 Statistical Society of Canada



Dans cet article, nous generalisons le critere de l'amalgame general minimal d'ordre inferieur (GMC) aux plans d'experience a trois niveaux. Dans un premier temps, nous passons en revue les relations entre le GMC et les autres criteres. Par la suite, nous presentons une representation numerique du nombre de composantes de repliement (ACNP) et un critere GMC a trois niveaux en considerant l'effet des composantes. Nous presentons aussi quelques resultats sur ce nouveau critere. Nous avons catalogue tous les devis GMC a 27 iterations et a 81 iterations ayant un nombre de facteurs egaux a et les devis GMC a 243 iterations ayant une resolution ou superieure. La revue canadienne de statistique 41: 192–210; 2013 © 2012 Societe statistique du Canada

Zhiming Li

Papers

Three-level regular designs with general minimum lower-order confounding