Study On Advanced Computing Techniques Based Consumer's Behaviour Characterizing For Sensory Evaluation Of Industrial Products

In many industrial companies, the activities are developed around consumer'spreference. In this situation, sensory evaluation can be applied as a powerful tool forexploiting market and quantifying human perception. Currently, sensory evaluation iswidely applied in enterprise in order to maintain and strengthen links betweenconsumers and products. To solve the problems and the difficulties spring up in sensoryevaluation domain, the present work, carried out in the framework of sensory dataanalysis, aims to characterize the behaviour of sensory panel, to determine relationshipsamong different panels and to forecast consumer's behaviour and preference accordingto sensory criteria of design experts.First, we proposed 2-level structure of sensory evaluation. We abstracted fouressential elements in the context of sensory evaluation and presented the subdivision ofeach element by using the format of set expression. Based on the four elements, wedefined Design-Oriented Sensory Evaluation (DOSE) and Market-Oriented S(?)nsoryEvaluation (MOSE). We analyzed the inner structure of two types of evaluations andthe difference between each other. The 2-level structure has been applied into theintegrated textile/clothing supply chain in order to analyze the requirement ofenterprises.Then, at the level of expert evaluation (DOSE), we proposed an effectivenormalisation model of sensory data. Using this model, sensory data from differentindividuals expressed with different formats can be normalized on a common optimalscale in order to guarantee information quality and uniformity.Next, the analysis methods and models of DOSE data were explained in details.According to the degree of consistency of data variation, we defined calculating methodof distance and sensitivity inter and intra-panels not only at the level of individual butalso the terms used. The obtained results of distance and sensitivity were then transformed into fuzzy numbers for physical interpretation. We also proposed alinguistic dictionary among evaluation terms and a procedure for improving the qualityof sensory evaluation. We compared our primary methods in this part with the classicalmethod (i.e. PCA) which is widely used in the domain of sensory analysis. Thecomparison result validated the superiority of our methods.Also, according to the design principles of consumer questionnaire and taking thecharacteristics of MOSE in mind, we designed a questionnaire for fabric handevaluation. For satisfying the practical requirement from enterprises, we built a datafiltration process to pretreat with MOSE data collected. This filtration processguarantees the quality of basic data used by the analysis step.Moreover, we presented the analysis method of MOSE data. According to thecharacteristics of the structure of MOSE data, we introduced the concept of equilibriumconsumption for estimating consumer's data variation between two different samples.This method can effectively solve the several local maximums problem existing inconsumer's data. According to the degree of consistency of data variation, we definedthe distances or dissimilarities between two different evaluation terms used byconsumers. Based on these dissimilarities, we developped a model for interpretingconsumer's terms and consumer's preference using normalized expert's evaluationterms. Finally, combining the methods and models presented above, we proposed aprocess for forecasting the consumer preference of new products.The research work of this thesis was carried through in the context of thedevelopment of product economy in European developed countries. We had effectivecommunion with some large French enterprises and conducted a large number of datacollection in the different countries (China and France) during the research proceeding.These provided a solid base for our research work and assured the practical value ofresearch harvest of this thesis in the integrated context of international cooperation. Atpresent, some parts of the research harvest have been applied in the analysis projects ofsensory data for some large companies in Europe.