| Sensory evaluations are often the ultimate measure of quality for many food products but food process controls rely on instrumental measurements. Effective techniques are needed to map desired sensory quality targets into instrumental process set points. In this research, fuzzy set and neural network techniques were used to determine food process set points that would produce products of certain desirable sensory quality. Fuzzy sets were applied to represent the intrinsic fuzzy characteristics of sensory responses without unverifiable assumptions, and neural networks were used to model the relationship between process and sensory variables. Rice cake production was used as a model process. Product sensory attributes were evaluated by a trained panel. Multi-judge responses were formulated as fuzzy membership vectors, which in turn were formed into fuzzy membership matrices of multiple sensory attributes. Multi-input multi-output (MIMO) neural networks were successfully used to determine process control set points for the manipulatable and instrumentally measurable variables. Experimental validation showed that the process set points led to products with desired sensory attributes. |
