AI Cupper: A Fuzzy Expert System for Sensorial Evaluation of Coffee Bean Attributes to Derive Quality Scoring

摘要

In the coffee industry, “cupping” is the process of sensorial evaluation of coffee beans, also known as sample evaluation. This process is done for three major reasons: 1) to determine the actual sensory differences between coffee samples; 2) to describe the flavors of the samples; and 3) to determine preference of product. In totality, cupping targets the measurement of the coffee's quality, which is expressed with a final numerical score. When cupping, the expert judge writes down the individual components’ scores and ranks their intensities for reference. Fuzzy logic has been employed for sensory evaluation of chhana podo (a baked dairy product), also for mango pulp and litchi juice. Moreover, a similar work exists only to train the Honduran Coffee Cuppers. This paper introduces a fuzzy expert system, AI Cupper offering an intuitive way for representing the judge's knowledge by linguistically modeling his perception of the coffee attributes through sensorial evaluation. It is capable of training cuppers when evaluating coffees from several countries and even has the capacity to learn as cupping data flows through it. The system was tested and has shown more than 95% of matching results compared with the experts’ results.