Intelligent techniques for cigarette formula design

This paper proposes a novel intelligent system for improving product formula design with sensory evaluation. The analyses and tests we carried out have shown that the proposed intelligent system is efficient for cigarette quality management, formula maintenance and new product design. Genetic algorithms, neural networks, support vector machines (SVMs) and fuzzy set method have been combined with expert knowledge in this system. The corresponding specialized knowledge can be extracted from trained neural nets or SVMs for mapping from tobacco cigarette chemical properties to sensory-quality indexes, classification of tobaccos, analysis of the correlation between chemical ingredients and sensory-quality indexes, and cigarette formula management and design.